Carries the four fixes that landed after the v1.34.1 tag plus the two found while
reviewing them.
The workout collapse that shipped in that batch only worked on the workout list,
because it keyed a session on a field three of the five callers never fetched.
Those three kept averaging one session into a person's baseline more than once.
The service-worker fallback version had drifted a release behind with nothing
checking it; both halves now have a gate.
The service worker derives its cache names from the version injected by
`sw-version.js`, and falls back to a literal compiled into `sw.js` when that
import fails. The prebuild step rewrites both from `package.json`, so both were
correct as long as a release rebuilt. One did not, and the literal trailed the
release by a version while every test stayed green.
Nothing checked it. The generated file is compared against `package.json`, and
the behavioural suite deliberately reads the literal out of the source so its
cache-name assertions cannot drift from the literal they exercise. Reasonable
apart, and together they left the literal itself unverified.
The effect is confined to the degraded path and is still worth closing: a client
that cannot load `sw-version.js` computes the previous release's cache names, so
it keeps serving the old shell and its cleanup treats the current caches as the
stale ones.
Re-anchors the literal and adds the equality as a gate, since it is a question
with an exact answer. A version bump committed without a rebuild now fails
locally instead of shipping.
The canonical picker keyed a session on sport, start, end and duration. Only the
workout list selected the end timestamp; the Coach narration, the per-sport
baseline and the deep-link cluster passed rows without it, so those rows carried
no session identity at all and the duplicate survived. One record counted once in
the list and two or three times everywhere else, including in the averages a
person is compared against.
Drop the end timestamp from the identity. On a workout with pauses it measures
wall-clock time while the duration counts only moving time, so two re-sends of
one session can differ in that field alone and the stricter key rejected a
genuine match. It also bought nothing in exchange: nobody starts a second
workout of the same sport in the same instant with the same duration.
The deep-link cluster now selects the duration it needs. A structural guard
freezes the call-site set and records which file owns each caller's select, so a
new surface cannot join the set without fetching the identity, and the identity
cannot grow a field only some callers fetch.
The badminton mapping added the enum member in the Zod registry and left the
generated spec behind, so the drift gate failed. One line, `- badminton`.
Third failure this branch revealed in sequence: the formatter hid a red test,
the red test hid this. Each gate only runs once the one before it passes, so a
single unformatted line kept two real problems invisible.
The spacing pass on this branch tightened `insight-status-card` to `gap-1.5`
and left `mood-insights-sections` at `gap-2`, so two sibling cards that are
meant to read as one family stopped matching. `assessment-spine-order.test.ts`
exists to hold exactly that contract and went red as soon as the formatter
stopped masking it: `format:check` failed first in the job, so the unit tests
never ran and the divergence shipped unnoticed onto the branch.
Bring the mood card to the same value rather than widen the other one back.
The direction of travel is the tighter rhythm, and both allowed steps are on
the sanctioned spacing scale.
The test's expected string moves with the contract, not around it: it still
asserts that both cards carry the SAME value, which is the whole point of it.
The formatter joins one JSX line that was left wrapped. No behaviour change.
It is the only thing standing between this branch and a green gate: typecheck,
tests, e2e, integration and both image builds already pass, and `format:check`
fails on this file alone.
- preserve prior rolling identity independently of local-day flooring
- scope exact resolution by schedule and era while advancing resolved grids
- include open rolling slots for display without widening compliance bands
- select active canonical recurrence slots instead of rebuilding from tick day
- scope intake suppression and dedup metadata to the occurrence local date
- lock and rotate refresh families inside one database transaction
- revoke CAS losers and insert successors before commit
- sign refresh artifacts only after committed state
- Normalize IPv6 literals before transition-prefix inspection
- Reject mapped, compatible, 6to4, and NAT64 private embeddings
- Share the verdict with pinned public and operator dispatchers
- document public-only install and data-free offline behavior
- record scoped current and legacy cache eviction
- capture green final-boundary and session-hook verification
- delete current and legacy HealthLog static page and data caches
- preserve unrelated CacheStorage namespaces during activation and logout
- keep IndexedDB query-cache removal in the session-end boundary
- remove the dashboard root from immutable precaching
- evict same-version root entries left by older workers
- retain public logos and favicon for offline support
- document passkey enrollment and reset recovery RED boundaries
- record focused expected failures and green parity evidence
- preserve production, schema and scanner-owned files untouched
- document expected cache-isolation and transition-form failures
- record preserved public and redirect behavior
- capture atomic task commits and verification evidence
- execute the operator CLI against real PostgreSQL credential tables
- require complete target revocation and unrelated-admin isolation
- force a mid-transaction database failure to prove rollback
- require password, passkey, TOTP and WebAuthn proof choices
- order fresh-factor completion before challenge creation
- keep successful enrollment in one cookie-bound ceremony
- refuse standard embedded-private IPv6 resolver answers pre-connect
- preserve public IPv6 to IPv4 fallback connectivity
- prove followed redirects repeat the pinned DNS verdict
- Remove the user-facing private-host bypass contract
- Require redacted operator approval guidance for legacy failures
- Preserve usable public and approved-private connected states
- require fresh existing-factor proof on both cookie-only legs
- bind registration challenges to user, purpose, session, expiry and one use
- cover bearer, ambient-cookie, stale, foreign and replay attacks
- cover mapped compatible 6to4 and NAT64 private embeddings
- exercise both literal URL and raw DNS address verdicts
- preserve public IPv4 native IPv6 and transition forms
- exercise signed-in service-worker install at CacheStorage boundary
- require data-free offline navigation and immutable public assets
- require current and legacy session cache cleanup
The 15-minute reminder tick computed windowDuration ONCE per schedule
as windowEnd - windowStart, then applied that SAME span to every
timeOfDay slot in the schedule. For a single-dose schedule that span
genuinely is the one slot's window, but for a multi-dose schedule
(e.g. windowStart=08:00/windowEnd=18:00 with timesOfDay=[08:00,18:00])
it silently gave the 08:00 slot a 600-minute grace before even
entering ORANGE phase — ten hours nobody configured — while an
identically-intended single-dose schedule with a narrow legacy span
(08:00-09:00) correctly escalated on schedule but then had only a
30-minute radius for a logged-but-late dose to suppress its reminder.
Same misapplied field, opposite-looking symptoms: one medication
silently never reminded, the other kept escalating despite being
logged as taken.
resolveSlotWindowDurationMinutes resolves each slot's duration
independently, checked in priority order: an explicit per-dose
doseWindows entry for that exact timeOfDay (the actual per-dose rule
set in the dose-window editor) > the schedule's reminderGraceMinutes
(documented for exactly this purpose but never actually read by the
reminder tick until now — the write path already reads it via
graceToleranceMs) > the same default an unconfigured "point" dose
gets everywhere else in the app, bounded by the gap to the nearest
sibling slot so two doses can never share a window. A single-dose
schedule with neither override keeps the exact legacy windowEnd -
windowStart behaviour, unchanged, for backward compatibility with
schedules that predate the per-dose doseWindows system.
The core property this restores: a dose's resolved window never
depends on how many sibling doses share its schedule. Two
identically-configured doses now resolve identically regardless of
whether one of them happens to have a second dose elsewhere in the
day.
Full suite (18k+ tests) passes with no regressions; 13 new tests
cover both real-world cases that surfaced this (a single 08:00-09:00
schedule and a twice-daily 08:00/18:00 one) plus the sibling-
independence property directly.
AGENTS.md duplicated the project description, version, stack table and a tree
of the source layout, then went untouched for about thirty releases. It
announced v1.4.0 as current and called the UI Dracula-themed, both false for a
long time. An instruction file that is confidently wrong is worse than none.
Replace the copies with a pointer to CLAUDE.md and keep only what is genuinely
specific to running as an agent here: the generated-client read ban, the gate
including TZ=UTC, reading output rather than exit codes, where working notes
live, and how the worktrees are used.
v1.33.1 claimed custom cycle symptoms were carried in backups. They were not,
and the same release turned the old silent drop into a hard failure, so the
accounts the fix was written for went from losing one symptom link to being
unable to restore at all.
Both ends of this defect had passing tests for the whole time it was broken.
The section builder returned the symptoms, the wire schema declared them, the
restore read them, and each of those three was covered on its own. Nothing
covered the object literal in between, which copied section keys out one at a
time and never listed `customSymptoms`.
So the test spans the seam and asserts on rows. It populates an account with a
symptom of its own on a day-log, builds the file through the real
`buildFullBackupPayload`, serialises it, parses it back through the real
`parseBackupPayload`, and drives the real admin restore. Asserting that the
payload carries the field would only re-test the writer that was already green.
Deleting the symptom row before the restore is the case the field exists for,
not a contrivance: `restoreCycleData` clears day-logs, cycles, and the profile
but never the symptom catalogue, so a restore onto the same instance still
finds the row sitting there. That is why the older round-trip test stayed green
through the whole thing.
Mutation check: hand-pick the cycle section's keys again and the restore
answers 500 on `Unknown cycle symptom keys: custom:sacroiliac-ache`, which is
the production failure verbatim.
The assembly picked `manifest` out of the records section by hand, one line
under the comment explaining that both sections are spread precisely so nobody
has to pick their keys. It resolved to the same value the spread had already
placed, at the same position, so the emitted file is byte-identical either way.
Left in, it is the pattern that lost the custom cycle symptoms sitting next to
the note about how that happened.
The sync function was widened to answer with what it wrote rather than a bare
count. This mock still resolved to a number, so the handler destructured
`imported` off it, got undefined, and reported `measurements_imported: NaN`.
Nothing in the type system could see it: `vi.fn().mockResolvedValue(0)` carries
no relation to the real signature, so a widened contract drifts past every mock
in the tree unnoticed. This was the only sibling; the class is worth a sweep of
its own.
`IntradayCumulativeProfile` shipped classified as backed up and carried by
nothing. The drain folds the per-sample rows into twenty-four running totals
and deletes the originals in the same transaction, so past the grace window
the table is the only copy of a day's curve: a restore rebuilt the account
without it, reported success, and left the same-time baseline reading "still
learning" for a fortnight.
Builder and restore sit together in one file, the way the profile and cycle
pairs do, because the restore route delegates and a grep of the route alone
lies. The section spreads into the payload rather than being copied key by
key. Two curves for the same metric and day, or a curve that is not a whole
day, throw naming the day instead of writing a row every reader drops.
The pair is proven end to end rather than at each end: the round-trip suite
builds a real payload from a populated account, serialises it, parses it back
through the real parser, drives the real restore, and asserts on the rows
written. Remove the read and it goes red.
Two parallel lines of work reached the same conclusion independently: a
baseline band over a quantity that cannot go negative must not report a lower
edge below zero. One of them answered it locally with a `Math.max(0, …)` at the
call site; the other gave every metric's numeric domain a single home and made
`buildBaselineBand` take the metric so the floor comes from the declaration.
Keep the second and drop the first. The required parameter is what surfaced
this at all: the new call site could not compile without naming its metric,
which is the whole reason it was made required rather than defaulted.
The section builder returned them, the wire schema declared them, and the
restore has read them since v1.33.1 — including the throw that replaced the
silent drop. The payload assembly copied its keys out one by one and was never
extended, so every backup written carried an empty list. An account with a
custom symptom did not get the silent drop that release removed; it got a
restore that failed outright on a key nothing had written back.
Spread the cycle and records sections instead of hand-picking their keys, so a
field added to either section interface reaches the wire by construction. Declare
labelEncrypted on the cycle section, which the builder already emitted.
The drain's existing transaction mock had no profile delegate, so the fold
could have thrown inside the per-day error boundary and every assertion would
still have passed. Cover the merge directly: the upsert lands before the
deleteMany, the stored curve matches where the samples actually fell, and a
serialised tick that finds the rows already drained writes nothing.
The rail gains a same_time_baseline kind and a server builder, which is the
growth path that type documents for itself. It speaks only when today is
actually outside the typical band: a card that reports doing exactly what you
usually do teaches people to stop reading the rail. Dismissing it is keyed by
day and metric, not by the hour, so it stays dismissed instead of coming back
every sixty minutes.
The Insights grid gains a section of four tiles, one per cumulative metric,
each honestly quiet until the history accrues.
The two-trajectory chart it needs did not exist, so DeltaSparkline gains a
second mode rather than a second chart family: the comparison curve rides the
same container, the same fixed height and the same shared y-domain as a dashed
muted line behind the live one, which is the only way the two can be read
against each other.
Six locales. The wipe and backup plans learn about the new table: the profile
cannot be rebuilt once the samples are gone, so it is carried, not derived.
The total shown after a backup upload skipped the nutrient day totals, so a file
that carried and restored them reported fewer records than it held. Same defect
as the payload gaps this branch closes, on the reporting side. Counts the
nutrient days, the health profile and the custom metrics with their readings.
The targets reference panel graded weight against the WHO BMI band even
for an account that had entered its own target, so the panel under the
weight chart contradicted the chart above it. An override now wins and
the card's source reads Custom, the same posture the glucose builder has
always taken with a user-pinned band. The BMI classification stays
WHO-derived: naming which category a reading falls in is a reference
fact, not a goal.
Thirty-three suites read a registrar as source text and matched
`boss.work` next to a queue constant. They now match `createAndWork`,
and the seven schedule tuples that grew a `cronIsTheRetry` third element
assert that element rather than accepting any arity — a guard that
accepts more than it did before is a guard that notices less.
Two negative assertions were tightened while they were open: the
"has no cron" checks on the arrival and workout-insight queues ended at
a literal `]`, so a cron added WITH send options would have slipped past.
The behavioural repairs carry the real change. Handlers that resolved to
undefined now resolve to their outcome, so the assertions name the counts
instead of asserting nothing. The reconcile tick's test said it swallows
a failure instead of throwing; it reports a failed outcome now, and the
test says that, with the comment pointing at the retryLimit that carries
the old anti-retry-loop concern.
The classification guard proved every model has an opinion and would have
passed unchanged while a model classified as carried was read into the payload
and never written back — which is what happened to the nutrient day totals, the
custom mood tags and the custom cycle symptoms, three releases running.
Split BACKED_UP into the models with both ends and the models without, each of
the latter carrying what an account loses meanwhile, so the debt is a register
rather than a silence. The check reads declared writer and restore files rather
than grepping for a route, because the route delegates: a review that grepped
only the route concluded the cycle data was never restored and had to be
retracted. Declaring the file list caught its own first omission — measurements
are read through the pager, not directly.
Adds a round trip that builds a real payload, parses it through the wire schema
and drives the real restore, asserting on the rows written rather than on the
payload. Dropping the cycle section spread makes it fail the way production
did: the restore throws on a symptom key nothing wrote back.
Covers what the required parameters now make impossible to forget: the
scorer takes the band whole and never widens it, a missing target scores
the bare trend, the resolved yardstick rides the weight component, the
shared builder refuses the reference band that sits under an override,
the snapshot gates its mood pillar on the module map, an explicit target
reaches both the score and the dashboard bands, and weight/BMI record
direction follows the user's own goal.
F1 adds the golden the score has never had: the rollup read branch and
the live read branch scored against one series encoded both ways. On the
canonical one-weigh-in-a-day cadence they agree exactly. On an uneven
cadence they part company, and the test pins that the whole difference is
the per-day mean collapse - the rollup branch equals the live branch fed
those same daily means - with a companion assertion proving the fixture
is not a no-op.
runJob's claim is behavioural, not typed: an ok:false outcome has to
reach pg-boss as a failed job, reported once, with the handler's own
facts and the thrown error's stack rather than runJob's frame. That is
what the retention cleanups now rely on, so it is pinned.
The rule's tests carry its blind spots in the valid list, where a reader
will trip over them: an aliased receiver is not detected, and neither is
a handler that returns ok:true after swallowing.
One guard per defect, each verified by breaking it on purpose first.
A non-negative metric's baseline `low` never goes below zero — swept over
every type in the catalogue with a series whose spread exceeds its
median, plus the reported step shape rendering an edge of exactly 0. The
axis assertion sits in the same file, because the point of the fix is
that the two read one fact.
No insights page passes an English word as a unit — the page sources are
scanned for `unit=` / `yAxisUnit=` literals and held against the set of
language-neutral symbols, and every `unitKey` is resolved in all six
locales. That second half matters: a key passed as a prop is invisible to
the `t()` call-site coverage sweep, so nothing else would notice a
missing translation.
An integer-valued metric never renders a fractional part — the step page
is rendered in German with no precision prop at all and must produce no
comma-decimal, on the Min/Max card and on the band sentence below it.
Two types were missing from the plausibility map, so the first guard also
asserts the map covers the enum.
All 104 bindings across the five registrars now go through createAndWork,
whose handler type is (jobs) => Promise<JobOutcome>. The forty-two inline
handlers return what they did. A handler that falls off the end no longer
compiles, and the healthlog/job-handler-outcome rule refuses a bare
boss.work( anywhere under src/lib/jobs except the one module that owns
the wrapper. The rule's header states its own limit: it catches the
omission, not a handler that claims ok:true after a swallow.
Retry policy, reviewed queue by queue because the rethrow changes what
runs, not only what is visible:
- Twenty-one cron passes carry cronIsTheRetry (retryLimit 0). They are
the retention DELETEs, the OAuth-state prunes, the hourly maintenance
sweeps and the nightly per-sample drain — every one of them a
statement that will fail identically a second later. pg-boss defaults
to two retries with no delay, so converting them without this would
have run the same doomed statement three times a night. The next tick
is the retry and it arrives with the whole gap as backoff.
- The policy rides on the schedule, not on createQueue: create_queue()
ends in ON CONFLICT DO NOTHING, so a queue-level default would be
inert on every already-running instance.
- Everything else keeps its retries deliberately. A TLS probe, an
off-host upload, a restore drill and a provider poll all fail
transiently, and the reminder ticks only fail at job level when the
pass itself could not run.
Six integration cards and the lab-OCR dialog keyed their tone off `res.ok`.
A sync that imported nothing because every reading was refused rendered a
green tick and a count, and a re-scan in which every confirmed row turned out
to be a duplicate raised a green toast saying zero readings were saved.
Each surface now renders the outcome the route resolved: a tick only when
everything landed, a warning when some of it did not, an error when none of
it did, and a neutral note when there was nothing new. The lab-OCR toast
picks its variant the same way.
Replaces the directory-scoped import guard with one that covers the whole
component tree and pins per-file occurrence counts, so a listed file cannot
quietly add a second tick. Its docstring states what it cannot catch.
Both were classified as backed up when the backup plan was written and carried
by nothing. The custom metric is live data loss: a series the account defined
itself is the one kind no integration can re-sync, so a restore rebuilt the
account without the metric and every one of its readings, and reported success.
Builder and restore live in one file beside each other, the way the cycle pair
does — a grep of the restore route alone does not find either, because the route
delegates to both.
Readings nest inside the metric that owns them rather than referencing it by
name, so the restore has no parent to look up and no lookup to miss. A duplicate
metric name and an undecodable ciphertext field both throw and name the value.
Intake events whose medication did not restore now throw with the unresolved
reference instead of being filtered away into a dose history that comes back
short while the restore reports success.
Compare today's running total at the last completed local hour against this
person's own typical total at that same hour, over a trailing four weeks of
stored intraday profiles. Same estimator as the vitals baseline — median for
the centre, median plus or minus k times MAD for the band — so a reader who
understands one understands the other.
It refuses to speak below fourteen usable days and says so through the
existing insufficient arm rather than a band built on three days. It ships no
projected day total: extrapolating a partial day is a guess, and this exists
to replace guesses.
Steps first; active energy, distance and flights ride the same engine.
Three defects on one card of the step sub-page, all of them a surface
answering a question that belongs to the metric.
"Your usual range is -5,595.6-13,387.6 steps", under a chart whose own
y-axis refuses to draw a negative step. Same app, same question, two
answers: the axis clamped and the sentence beneath it did not. Both now
read the same fact. `VALUE_RANGES` already declares every metric's
plausibility floor for ingest; `value-domain.ts` reads it and answers
"can this metric be negative" for the padded y-domain and for the median
+/- k.MAD band alike. The band's lower edge floors at zero for a
non-negative metric; `spread` stays the unclamped dispersion so a
deviation-in-sigma calculation is unaffected. `buildBaselineBand` now
takes the metric it is describing as a required argument, so a new caller
cannot quietly reintroduce a negative edge. Two types carried no
plausibility range at all, which let any number through ingest and read
as "unknown" on the negativity question: walking steadiness gets the same
0-100 percent band as its two gait siblings, the loud-listening event a
floor of 1 with a generous ceiling.
"13,387.6 steps" also said steps to a reader of German. Five sub-page
units are English words, not symbols - steps, flights, falls, years, and
a "count" that was a poor label in any language. They now travel as i18n
keys through a new `unitKey` on the metric-page scaffold and resolve in
all six locales. The thirteen real symbols (bpm, mmHg, %, m/s, kg/m2,
...) stay literals; translating a symbol would be the opposite mistake.
And the Min read "104.0 steps". Precision was a per-page prop defaulting
to one decimal, so every page had to remember that its metric is a count
- and the step page did not. It follows the metric now: a discrete metric
(tallies, category events, closed integer instrument scores, ordinal
levels) resolves zero decimals wherever it is rendered, and a page only
passes the prop to override that.
The weight pillar has been graded against something for two years and
never said what. The score card's weight row now carries one line: the
user's own target band when they set one, or a plain statement that the
row scored the trend with no target at all. The band rides the analytics
DTO in canonical kg and the page formats it into the reader's display
unit, so an imperial account reads pounds.
The weight sub-page shades the same yardstick: its chart bands follow an
explicit target when one exists and fall back to the height-derived band
only when it does not.
A cumulative metric has no meaningful latest reading — a partial day's step
count is a fragment, not a value — so the only honest comparison is against
the same person's own typical total at the same local hour. That needs an
hourly substrate and there was none: the nightly cumulative drain collapses
per-sample Apple Health rows into one daily total after the grace window and
hard-deletes the originals in the same transaction. Today's curve was
derivable; every older day's was already gone.
Fold the rows into twenty-four running totals and write them down first, in
the same transaction and before the delete. The merge is additive so a day
drained twice keeps both halves, and the rows are re-read under the advisory
lock the drain already takes so an overlapping tick folds nothing.
One table, one nightly range delete to bound it.
Sixty-two handler functions across twenty-four modules now resolve to a
JobOutcome instead of undefined. The counts they already wrote to their
wide event ride out in `did`, so a run that deleted nothing and a run
that could not run are no longer the same shape.
Three rules decided each conversion:
- a catch that swallowed becomes jobFailed, keeping its warning line;
- a catch that already rethrew keeps its throw, because the throw
already fails the job and rewrapping it buries the stack;
- a fan-out pass is judged on the pass. One account's revoked grant
belongs on that account's integration ledger, not in a queue-wide
failure that retries the whole cohort, so per-leg failures ride out
as a users_failed count — visible for the first time.
The retention cleanups are the behaviour change worth naming: every one
of them used to warn and return, so a purge that had not run for weeks
looked like a purge with nothing to do. They fail now. Their registrar
sets retryLimit accordingly in the commit that converts the bindings.
JobFailure carries the thrown error's stack; without it every one of the
101 queues would report the same runJob frame.
The section builder returned them, the wire schema declared them, and the
restore has read them since v1.33.1 — including the throw that replaced the
silent drop. The payload assembly copied its keys out one by one and was never
extended, so every backup written carried an empty list. An account with a
custom symptom did not get the silent drop that release removed; it got a
restore that failed outright on a key nothing had written back.
Spread the cycle and records sections instead of hand-picking their keys, so a
field added to either section interface reaches the wire by construction. Declare
labelEncrypted on the cycle section, which the builder already emitted.
The Health Score graded the weight pillar against 22 x height squared,
expanded by 2 kg either side: a target nobody set, nothing disclosed,
and the only producer of the pillar's band. For anyone with a height on
file it pre-empted both the honest trend-only path the module header
promised and the null-redistribution behind it, while the target the
user had actually entered sat unread in User.thresholdsJson.
computeHealthScore now takes the band whole plus the yardstick it came
from. Both callers resolve it through one shared builder next to
buildHealthScoreBpInputs, which hands over a band only when
getEffectiveRange reports a real user override; every other account
scores the bare trend. defaultWeightTargetFromHeight is gone.
Same pass, same defect class elsewhere: the dashboard bands, the
weight-drift verdict rung and the weight/BMI record direction all
ignored an explicit target too, and the snapshot never resolved the
mood module, so a mood-off account got a mood pillar in the hero ring
and none in the Insights card. moodEnabled and the weight yardstick are
required parameters now, so a caller that forgets one fails to compile
instead of quietly scoring against a default.
Six provider syncs returned a bare count, which narrowed their failures away:
a Nightscout window in which every SGV row was refused by a constraint
returned the same 0 as a quiet night, and an Oura or Polar run whose
collections all failed returned 0 too. The routes then answered 200 with a
count and the settings card drew a green tick over it.
Every sync now reports what it imported AND whether any part of it did not
land, and the route resolves the outcome server-side so the web card and the
native client cannot disagree. Google Health keeps its 502 for a run that
failed and wrote nothing, but a run that failed after writing some resources
is now a partial rather than a discarded result. The labs OCR commit carries
the same resolved outcome.
Behavioural cases first: the Google mapper keeps the raw `exerciseType`
when the sport falls back to "other", reports a missing one as null
rather than an empty string, and the ingest writes it under
`metadata.googleExerciseType` on both the insert and the update arm — the
update is what a full re-sync goes through, so a row written by an older
build repairs in place. The key is omitted rather than nulled when a
session carries no type, so a re-sync cannot wipe what an earlier one
recorded.
Then the structural guard, because the behavioural ones only cover the
provider that was reported. It freezes the set of sport-label mappers and
the set of files creating Workout rows, and asserts every mapping ingest
carries its raw label into metadata. Fitbit has the same gap and is
excluded on purpose — it ships on its own, small and isolated, per the
standing rule for that provider. The exception is an assertion, not an
omission: when Fitbit gains the write, the guard fails and forces the
entry to move. An exception nobody has to retire is how this gap survived
in the first place.
The rule that a write which wrote nothing must not read as a success was
scoped to the import panel. Provider syncs need the same rule, so the
classifier moves to src/lib/outcome/written-outcome.ts and the tick plus the
success colour move to one presentation component the whole tree renders
through. Adds the sync shape: a run reports what it imported and whether any
part of it failed, and the outcome is resolved from both.
A pg-boss handler signalled success by falling off the end of a function.
Every binding in this tree resolved to undefined and nothing read the
value, so a handler that caught an error, logged a warning and returned
was indistinguishable from one that finished its work: pg-boss marked
both completed and the failed-job readers saw nothing.
JobOutcome makes the completion a value the handler has to construct.
runJob turns ok:false into a reported, rethrown failure so the queue's
retry policy applies. createAndWork is the one binding path; its handler
type is (jobs) => Promise<JobOutcome>, so a handler that returns nothing
no longer compiles.
The type cannot make a handler honest — jobDone() after a swallow still
compiles. It converts omission into commission, which is the smaller and
far more reviewable class.
An unmapped Google `exerciseType` resolves to the literal "other", and
until now that was the end of the label: nothing else carried it, so a
workout Google typed as a sport the map does not know became
indistinguishable from a genuine `OTHER_WORKOUT`. Carry the trimmed
upstream value through `GoogleHealthMappedWorkout.sportTypeRaw` and write
it to `Workout.metadata.googleExerciseType`, the shape Polar, Strava and
WHOOP already use. Insert and update share one payload, so a full sync
repairs rows written by an older build.
The comment above the type map claimed an unmapped type "still persists"
because the column is free-text. It does not, and it did not when the
comment was written. Prose that asserts what the code does not do is the
same defect as the missing write, so correct it here rather than later.
The v1.33.1 credit block named a reporter for the dashboard layout while
the body of the notes described nothing of the kind. The fix was in the
release; only its description was missing, which leaves a credit pointing
at nothing and a reader unable to tell what changed on their own instance.
Eight workstreams, two migrations, and a pass by an independent reader that
found four things this release was claiming about itself that were not true.
Two of the fixes change what people receive. Medication reminders went out
every fifteen minutes because the dedup key could not tell one dose slot from
another, and "handled by the app" was silencing every channel rather than the
one it names. Both are called out in the notes, because a user who set that
flag as a general mute will notice.
Two came from reports: an OpenAI-compatible gateway that the provider list
had no room for, after an earlier answer from us said otherwise, and four
places where a timestamp followed the browser instead of the profile.
The backup half is the one worth reading. Three record types were carried and
could not be restored, or could not be carried at all, and each failed in a
different direction: one dropped silently, one stopped the restore outright,
one was discarded while the restore reported success. What a backup holds is
now declared in one file with a written reason for everything left out, and
the roughly thirty types that are still carried by neither end are named there
instead of waiting to be discovered by someone who needed them.
The Del chip added beside the document card measured 4.06:1 against the card's
background, under the 4.5:1 floor, and axe failed the vault on it. So the
change that made the destructive key visible introduced a barrier of its own.
It uses the full foreground token now. The design rule says muted for meta
text, and a key hint is meta — but it is the one kind of meta that has to be
readable at a glance, so here the contrast rule wins and the reason sits next
to the class list rather than in a commit nobody will read again.
Found by running the e2e suite locally before pushing rather than after.
The guard tested whether the argument list contained the substring "timeZone".
`timeZoneName: "short"` contains it. That option appends "GMT+1" to the output
and leaves the zone as the browser's, so a formatter written that way sailed
through while being exactly as wrong as a bare one — and looking more careful
than a bare one, which is worse.
Tightening it to a colon-terminated property then reported nine correct call
sites, because the shorthand `{ timeZone }` is what most of this tree writes.
The honest response to that would have been to allowlist all nine, which is
how a guard stops guarding. It accepts both forms now, and a test pins both
directions rather than leaving the regex to be trusted.
Two exclusion reasons in the backup plan aged inside their own release and are
corrected: push attempts now also hold the reminder dedup anchors this release
introduced, and the thumbnail's "rebuilds from the stored bytes" only holds on
the disaster-recovery path, since a portable export carries no bytes to rebuild
from.
Four things an independent read found, three of them things this release said
about itself that were not true.
A custom mood tag travelled without its label, its icon, its sort order and
its SCALE. The label is the smaller loss: the person who named a factor
"Migräne" would have got a key back. The scale is the real one — the inverse
flag marks a factor where a high value is bad, and dropping it does not lose a
label, it silently reverses the meaning of every rating already recorded
against that tag. Custom cycle symptoms lost their label the same way. Both
now carry the ciphertext verbatim, like every other encrypted column in a
disaster-recovery payload.
The plan file claimed, in the present tense, that a guard requires both a
payload reader and a restore branch for every carried model. No such guard
exists. The commit that introduced the file said so honestly; the file itself
said the opposite, and roughly 31 models have neither end. BACKED_UP means
"must be", not "is", and both the plan and its test now say that plainly. The
wrong sentence is recorded rather than quietly swapped, because it is the same
defect the file was written to prevent.
The export route still told operators that labs, allergies, illness episodes
and workouts could be exported but never restored. The restore route
re-creates all of them, and since this release nutrient day totals too. The
scope now points at the plan instead of repeating it beside the code, which is
how that sentence came to be wrong.
And the export card promised "every record you own". It does not, so it no
longer says so — six locales.
The resolver is online-first, so mounting the databases fills the ASN
and carrier but does not by itself keep login IPs on the host. Point at
the IP_GEO_LOOKUP_DISABLED section that does.
The hint said switching a widget off here turns it off in the mobile
app. This page writes the flag; whether the client applies it on its
next read is the client's business, and there is an open report that it
does not. Say where the setting lives instead of promising a result
this surface does not control. Six locales.
BMI renders on the native dashboard and was configurable from nowhere.
Two individually-correct decisions produced it: `bmi` is a catalogue id
with no web render path, and the web layout editor filters those out
because a toggle over something this page cannot draw would be a silent
no-op. The widgets PUT accepts and persists them all the same, and the
native client writes them, so an account ended up holding rows it never
set and could not change from any surface.
The sortable list stays exactly as it was. Below it, the ids the web
does not draw but the layout carries get their own labelled group with
the same two flags and no reordering — placement belongs to the client
that draws them. An account with no native client has nothing there and
sees nothing. The module gate applies as it does above, fail-open.
Refs #581
A focused card deleted on Backspace as well as Delete. The timeline's
roving tabindex gives a card focus the moment someone tabs into the
grid, Backspace is the reflex "go back" key, and a card carries no
delete control at all — so the reflex press removed a medical document
with nothing but a transient undo toast behind it.
Delete alone now deletes. The key resolution moves into a pure helper
in vault-utils so the contract is testable without a render harness,
and the card stops keeping that contract in a code comment: the overlay
button carries aria-keyshortcuts, and a kbd chip appears while the card
holds focus. The label is per-locale (Entf / Suppr / Supr / Canc).
The switch was described three times as applying to what you upload
next. A genuine off-to-on flip also enqueues a pass over the documents
already stored without a summary, capped at 200 per flip — scanned
medical documents going to a third-party provider, from copy that reads
as forward-looking only.
Drop "newly uploaded" from the sub-label and give the disclosure the
sentence it was missing, including the cap. Six locales. The behaviour
is right and stays as it is; only the sentence changes.
A guard pins the number in the copy to MAX_ENQUEUES_PER_RUN, so moving
the cap without touching the copy fails.
The runtime has always read GEOLITE2_DIR, but the variable was absent
from the compose `environment:` whitelist, from both env examples, and
from the docs. A variable that is not on that whitelist never reaches
the container even when the operator's `.env` sets it, so the override
existed in code and was unreachable from the documented path.
Put it on the whitelist, document it in both examples and the manifest,
and add `docs/self-hosting/geolite2.md`: fetch the databases with a free
MaxMind account, mount the directory read-only, point the variable at
it. The compose file carries the bind-mount line commented out beside
the app service.
The admin notification told the reader to set MAXMIND_LICENSE_KEY in
this repository's Actions secrets — something nobody running the
published image can do, about a fallback that sends their sign-in
addresses to a third party. It now names the mount, names the online
provider it is actually querying, and links the runbook. Six locales.
It also no longer claims the carrier column stays empty; the online
tier has filled the carrier from the ISP field since v1.25.8.
A blank GEOLITE2_DIR now resolves to the default instead of the process
working directory: compose substitutes an unset variable to the empty
string, which is not nullish.
Refs #659
The Bestand tab was 1,139 lines of one client file holding two concerns that
only meet at the row shape: the list of containers a person holds, and the
three dialogs that change it. Registration, per-item correction, packaging
economics, the summary and the row rendering all shared a scope, so reading
any one of them meant scrolling past the rest.
Move-only. The dialogs go to `inventory-dialogs.tsx` with the day-input
conversions their fields produce; the row shape and the cache-invalidation
they both need go to `inventory-shared.ts`; the section keeps the summary and
the list at 458 lines. Every export the tab offered is still offered here, so
no caller changes. No copy changes, no new keys.
Two suites pin these contracts by reading the file's own source — the register
dialog's unit conversion and the correction dialog's sparse patch. Their
`readFileSync` now names the file the code moved to. The assertions are
unchanged, byte for byte; only the path they read is.
The retention-purge comments still cite migration 0277 for indexes that landed
in 0276 — the renumbering moved the directories and left the prose behind. The
schema is trustworthy because its comments are, so a wrong number is not
cosmetic.
The import worker's `already_recorded` reason claimed the skip covers a dose
recorded on any other intake surface. It does not: `source` is part of the
live-row unique, so a hand-logged dose and an imported one at the same slot are
different keys and both are written. Say what the mechanism actually does, and
say plainly that cross-surface dedup is not there.
The per-medication import and the account-wide dose-history import queued
their work through ninety near-identical lines each: row lock, stale-job
sweep, active-job refusal, job row, pg-boss send, and the fallback that fails
the row when the queue is not there. Five steps written twice drift, and the
drift shows up as an import that hangs on one surface and not the other. The
status projection moved out of these routes for the same reason; this is the
write half.
Scope is the only real difference, and it is one argument. What the two
surfaces legitimately say differently — the 409 wording, the envelope, the
kickoff audit line — stays at the call site.
The replay key had the same shape of problem in smaller form: the export
reader owned a function for it and the hand-listed body inlined the same
template. Both now call one function, with a test that pins them equal for
the same medication and instant.
The supply tab carried its own copy of the two container kinds whose first
opening starts a degradation clock, with a comment saying it mirrored the
state machine's list. Nothing enforced the mirror — the state machine types
itself against the generated Prisma client, which the browser bundle cannot
carry, so a copy was the only way to say it.
Move the list into a client-safe module with no imports and let both sides
read it. The state machine re-exports it under the Prisma enum, so a renamed
or dropped container kind is now a compile error rather than a member that
quietly stops matching.
The resolver reads the profile's sex for the body-fat and pulse bands, and
nothing asserted it. A resolver that stopped reading the field, or read it and
defaulted to one side, passed the whole suite. Pin both recorded sides and the
neutral band that OTHER and an unanswered profile share.
`binaryReferenceSex` was already the declared owner of the shrink from the
three stored values to the two a sex-split reference table has rows for, but
eight sites still narrowed inline — including a byte-for-byte private copy in
the effective-range resolver and a ternary 37 lines above a `toProfileSex`
call in the same file. Each of those is a place the fallback can drift.
Delete the copy, replace the ternaries, and name the pair once as
`BinaryReferenceSex` so a signature that needs it stops re-listing the two
literals. No behaviour changes: every site already resolved OTHER and "no
answer" to the same neutral or absent value, and the existing suites pass
untouched.
A migration container starts empty, so applying 0281 there proves the
statements parse and nothing else — the backfill's WHERE matches no rows and a
broken expression passes green. Every self-hoster runs it against rows that do
carry a leg, and those are the only rows it exists for.
Seed the pre-migration shape, run the migration's own UPDATE read out of the
committed file, and assert an attributed row comes out still knowing which leg
broke, an unattributed one is left alone, and a second run does not nest the
array.
The batch route has accepted a `syncTrigger` since v1.32.8 and put it on the
wide event, where no human reads it. Half of what was promised got built: the
value is written and appears in three places in the codebase, none of which is
a screen anyone looks at.
Persist it on the account, alongside the instant of the most recent batch that
arrived without the app being opened, and publish both from
`GET /api/integrations/healthkit`. The card now names how the last sync arrived
and when data last came in on its own. `healthKitLastSyncedAt` alone reads
identically whether the phone delivers by itself or only while someone has the
app open, which is why "is background delivery working" could not be settled
from either side.
A batch that declares no trigger writes null rather than leaving the previous
one standing — the field describes this sync, and an older client's silence is
not evidence about it. A foreground batch leaves the background timestamp
alone; overwriting it would erase the only answer the card has.
Also here: WHOOP, Fitbit and Google Health join the workout-freshness table.
All three write workouts, and without an entry their cards could report every
measurement type fresh while the workout leg had been silent for weeks. And the
status pill drops a generic `aria-label` that replaced its own accessible name,
so a screen reader announced the same words for a healthy connection as for a
broken one.
`failing_leg` was one slot, so two legs of the same provider failing in
overlapping windows overwrote each other. The row remembered whichever failed
last, and that leg recovering then cleared an error the other one was still
causing — the ledger forgot a live failure and painted the card green over it.
The slot becomes a set. A failure adds its leg, a success removes its own, and
the error state, the per-kind buckets and the streak anchor clear only once the
set is empty. The array is ordered oldest failure first, so the last element is
the leg that wrote the message `last_error` holds; when that leg recovers while
others are still failing, the message is cleared rather than left describing a
leg that is now healthy.
An unattributed failure adds nothing and removes nothing. A single-leg provider
therefore behaves exactly as before, and a multi-leg provider no longer loses
what its other legs already reported.
The column never reached a client — it is absent from the status serialiser and
from the OpenAPI contract — so it is replaced rather than shimmed.
The module-gate branch cited the client-managed skip as its precedent for
advancing past a cycle. That skip no longer exists in this file, and the
reason the gate advances is its own: a disabled module has nothing left
to deliver.
`apns-collapse-id` is a replace directive: a push carrying an id
supersedes the earlier undelivered push with the same id. Every
medication reminder carried the literal string MEDICATION_REMINDER, so
three medications due at 08:00 arrived on the phone as one notification
naming one of them. The user takes that one and misses the others with
nothing on screen to suggest anything was dropped.
Key the collapse on the per-slot tag the Web Push sender already
coalesces on, falling back to the event type when a payload carries none.
Grouping stays on `threadId`, which is what grouping is for. Discreet
mode keeps the generic id — no record-identifying string on the lock
screen — and a tag over Apple's 64-byte ceiling falls back rather than
costing the send.
The per-phase "already sent today" guard read the Telegram message
ledger, and that table is written only when a Telegram send succeeds. A
user on email, APNs or ntfy never had a row, so the 15-minute tick
re-derived the same overdue slot and sent it again on every pass — around
fifty mails for a single missed dose.
Anchor the guard in `push_attempts` instead, the ledger the daily
briefing and the illness escalation already use. Two properties are
deliberate: the anchor is stamped for every dispatched slot regardless of
what any channel returned, because keying it on a delivered push would
re-open the flood the moment a channel starts failing; and it is stamped
before the dispatch, so a crash costs one phase's notice rather than a
repeat every quarter hour.
The key carries the schedule id and the slot's real HH:mm alongside the
phase and the local date. Both are load-bearing: two legacy
single-window schedules on one medication share the empty first-class
time-of-day, so a coarser key would collapse the evening dose into the
morning one. The tests assert the other direction too — the next phase
still escalates, and the next local day starts clean.
The Telegram ledger keeps its message-tracking duties; it is no longer
the dedup key. Anchor rows carry a sentinel channel and stay out of the
operator's notification health and diagnostic reads, which count
deliveries, not decisions.
`medication.clientManaged` and `measurementReminder.clientManaged` mean
the iOS client schedules the reminder locally, so the server should not
push it. That is what the OpenAPI description, the settings copy and the
iOS contract all say. What the crons did was skip the dispatch entirely,
so a user who flipped the switch also lost Telegram, ntfy, webhook,
email and Web Push — channels they may well read on a desktop while the
phone holding the local schedule sits in a drawer.
Move the gate into the dispatcher's channel loop, where it applies to
APNs and nothing else. The skip is not counted as an attempt, so
`dispatched` keeps meaning "something reached the user", and it stays
countable: the same wide-event keys the crons emitted are emitted here,
plus a `skipped` row in the push-attempt ledger.
The Vorsorge path is the sharper half. It suppressed every channel and
then advanced `nextDueAt`, which dropped the reminder out of the
preventive-care card and out of the daily digest: a reminder that never
reached anyone was recorded as handled. The advance now happens only
where it always belonged, after a dispatch that a channel actually
delivered. When nothing delivers, the reminder stays overdue and visible.
The restore resolves a mood entry's rated factors by key and asks only for the
seeded catalogue. A tag the user created is not in it, so the lookup came up
short and the whole restore threw 'Unknown mood factor keys'. The file was
complete; it simply could not be read back.
That is the same cause as the cycle symptoms two commits ago, with the
opposite ending. There, an unresolvable key was filtered out and the loss was
silent. Here it stopped everything. One shape of missing definition, two
answers, neither of them the right one.
The account's own tag definitions travel now, get written before the factors
resolve, and the lookup accepts the seeded catalogue or the owner's own. The
seeded rows stay out of the file for the same reason as before: a restore must
not rewrite another instance's reference data.
Four more test mocks needed the new delegate, which is again the compiler and
the suite naming the call sites rather than me hunting for them.
The export has written nutrient day totals since v1.29. The canonical schema
never declared the field, so `.passthrough()` carried it through parsing and
no reader ever looked at it: every water and vitamin total in the file was
discarded on restore, and the restore reported success.
Declared in the contract, cleared and re-created beside its siblings, and
counted in both places that count — what the file holds and what the restore
replaced. A count that stops at the domains someone remembered is how this
stayed quiet.
restoreCycleData was loud about one missing reference and silent about
another, three lines apart. An unknown cycle reference threw; an unknown
symptom key was filtered out of the list, so the day-log came back with one of
its symptoms quietly gone and the restore reported success.
The cause was upstream: the seeded catalogue is reference data every instance
has, but a symptom the user created lives only in their account and was never
carried. The key had nothing to resolve against, and the filter turned that
into silence.
Both ends now. The account's own definitions travel with the backup, the
seeded rows deliberately do not — shipping those would let one instance's
restore rewrite another's reference data. And the resolve step throws on an
unresolvable key rather than filtering it, so the same function answers the
same question the same way twice.
Widening the builder's client type made four export tests fail on a missing
delegate, which is the sweep working: those are the call sites, named by the
compiler instead of found by reading.
The mirror of the wipe plan, for the same reason and against a worse failure:
a list that was right when written and stops being right every time a model is
added. On the delete side that leaves data behind; here it means data does not
come back, and nothing says so until someone is restoring a backup because
something already went wrong.
The list is longer than the wipe list by sixteen models, and that gap is the
point. A wipe may lean on the database — deleting a medication cascades to its
schedules, dose changes and inventory events — but a restore has no cascade,
so every child has to be named here.
Every exclusion carries its reason in prose, and the guard enforces a length
floor on it. That floor is not pedantry: the first draft of this file said
'Same as WithingsConnection' four times, which is a pointer rather than a
reason, and a pointer silently inherits whatever the thing it points at
becomes. Writing the decision where it is made is the only way it stays
reviewable.
The guard found two things in its own author's work on the first run: a model
with no verdict at all, and a reason too thin to be one.
The restore side of the guard — that every carried model also has a restore
branch — lands with the payload work it will police.
A formatter with no timeZone follows the browser, which is a different
question from the one this app answers. Four surfaces had drifted apart that
way before anyone noticed, and one of them printed a shared report a day off
from the page it came from.
The admin host chart went with them rather than onto the allowlist. Server
samples are real instants, and an operator reading their own instance should
not find a spike at 14:20 here and 15:20 two screens away.
Two deliberate limits, both written into the test. It does not match
toLocaleString, which in this tree is overwhelmingly a number formatter, and
burying four real hits under every thousands separator is how a guard gets
silenced. And it cannot tell a Date built from calendar arithmetic — a month
label, a weekday header, no instant behind it — from one parsed out of stored
data, so those four sites carry a written reason instead. The value is the
ratchet: a new bare formatter has to be argued for before it ships.
The bucketer hard-coded Europe/Berlin for day, week and month boundaries, so
a reading taken on the last evening of a month landed in the wrong month's
bar for everyone who does not live there. Two smaller siblings did the same
thing: the Today chip asked the browser for a time instead of the profile,
and the score card's tooltip formatter carried a locale but no timezone.
The boundary timezone is a required option rather than a defaulted one. A
default would have left any caller that forgets to pass one still wrong and
still silent; required makes the compiler name them, which is how the three
call sites and the eight test sites in this diff were found rather than
guessed at. The tests were always asserting Berlin's calendar without saying
so, and now they say so.
Two fixture mistakes worth recording, both caught by running the tests rather
than trusting them: summer time is already in force in Berlin on 31 March, so
the first instant chosen was April on both sides; and 31 March 2025 falls in
the same ISO week as 1 April, so the week case needed its own instant where
one side is still Sunday.
The page at c/[token] resolved the owner's timezone; the PDF download beside
it resolved nothing and fell through to the display default. Same link, same
reading, two dates — and the one a practice files is the one that was wrong.
Resolved in the shared download gate rather than in the route, so the PDF and
the FHIR bundle cannot drift apart the way the page and the PDF did. The
owner's clock preference travels with it for the same reason.
The fixture is deliberately west of UTC: Berlin is the fallback, so a Berlin
fixture cannot tell a resolved timezone from a forgotten one.
The rotation-coverage guard greps the script for a quoted column name.
Listing the new column in the header comment satisfied that grep without
the script rotating anything, which made the guard pass on a script that
would have left the column behind. The comment now names it unquoted and
says why.
Names LiteLLM, OpenRouter and vLLM, states what it does not share with
the OpenAI provider and why, and keeps the Local provider's own section
intact — it still speaks the same wire and remains the right pick for a
model on your own machine.
The properties this provider has to keep: it resolves against the base
URL that was saved for it, the pinned OpenAI arm cannot see that URL and
still posts to api.openai.com, and the OpenAI key never travels to the
gateway. Each is asserted at the wire, not on internals.
The SSRF floor is checked through the same alt-notation classes the
Local provider is checked against, and the pinned tags are shown to keep
requiring a public host even when the operator opened every private one.
The AI settings connect-src is pinned to what it already allowed: the
gateway is called server-side, so no CSP change is needed, and a future
change that needs one will turn this red.
Its own form with its own three fields, its own entry in the provider
dropdown and in the fallback-chain editor, and copy in six locales that
says plainly where the OpenAI key does not go.
The chain editor offers the entry but the default chain does not carry
it: it resolves only against a base URL somebody saved.
The read endpoint reports the gateway's base URL and model, and the
bearer as presence only. OpenAPI regenerated.
The chain gains a seventh entry. It builds the OpenAI client against the
base URL the user saved, with the bearer they saved for it and the model
they named, and it reads none of the OpenAI columns — the pinned
api.openai.com arm stays pinned.
Host policy is the Local provider's, unchanged: a public host always, a
private one only where the operator allowlisted it, which is what a LAN
LiteLLM or vLLM needs. Egress keeps riding safeFetch.
The JSON-mode dialect the Local client learns per endpoint moves into a
shared module and now serves both clients, so a gateway that rejects
response_format is detected once and retried without it.
The entry stays out of the default chain: it can only resolve against a
base URL somebody configured.
Three dedicated User columns — base URL, optional bearer, model — for a
provider whose endpoint the user chooses. They are separate from the
OpenAI columns on purpose: the OpenAI arm pins api.openai.com, so it must
share no column with a user-supplied host, and a gateway must never
receive the key issued for OpenAI.
The bearer joins the encrypted-column registry and the rotation script;
the three columns join the account-wipe reset set.
Four specs navigated and then read the DOM once, with nothing in between
that waits for anything but the network. An idle network says nothing about
whether React has rendered, so the read can sample a tree that is still
mounting and the assertion then blames the application for a race inside the
test. Two such reads went red on consecutive CI runs, on different specs.
The settings over-scroll guard also slept a fixed 500 ms, which is the wrong
shape twice over: too long when the page is ready, too short on a loaded
runner, and it never says what it is waiting for.
The structural guard runs in the unit suite rather than in Playwright,
because e2e runs at PR-CI only and flaky counts as failed there — one
unguarded read otherwise costs a whole pipeline before anyone sees it. It
refuses to count a wait whose rejection is thrown away, since that reads
like a guarantee and is not one, and would leave the check defeatable by
appending a catch. The one deliberate swallow is on the allowlist with its
reason: a genuine regression there still fails at the width assertion below
it, which reports the measured number instead of a timeout message.
A release asset is served with content-disposition attachment and nosniff, so
an image embedded from that URL renders as a broken image in the notes. The
raw path serves it as an image, which is the only thing that actually works.
A chart shows the shape and hides the numbers. An appointment needs the
numbers, and so does anyone moving readings into a spreadsheet.
The table reads the same expression the chart hands Recharts, after the range
slice, the week/month bucketing and the display scale, and it formats through
the chart's own value and date formatters. Nothing is re-derived, so a cell and
the tooltip cannot print two numbers for one point.
Collapsed by default, with the point count in the trigger so the size reads
while it is closed; the region scrolls in its own box rather than growing the
page, and the table primitive already scrolls a wide row set sideways. The
caption names the grain, because a weekly mean labelled as a daily reading
would be a lie, and the grain now has one source shared with the bucketing.
Derived overlays get no column: a moving average is drawn from the points, not
measured. An empty window renders no table at all, since the chart's own empty
state already says there is nothing there.
Opt-in per mount. The metric sub-pages take it; the dashboard tiles do not,
because a table under every tile buries the dashboard it summarises.
Refs #636
The tag crosstab already ran a with-versus-without Welch test per structured
tag, with a D to D+1 pairing mode whose stated example was an evening tag
against next-day recovery. It only ever had one channel at that lag, and a
recovery score is a wearable composite plenty of accounts never record. The
rated-factor board next to it has tested next-day resting heart rate and HRV
since v1.14.0, but a rated factor is a continuous score, so a binary tag never
reached either vital.
Add both to the tag map. Same engine, same per-side day floors, same
Benjamini-Hochberg family, same card. The fetch filter derives from the map, so
the channels come along without a second query.
The delta read-out goes neutral to match the rated-factor board: with resting
heart rate on the board, a positive delta is worse on one row and better on the
next, and a green plus sign would claim a verdict the association cannot carry.
Refs #616
The preamble already says once what the two states mean and that neither is
absence; repeating it per row would have doubled a manifest that rides every
turn.
The unit suite mocks Prisma, so it proves the discriminant and nothing about the
query underneath it — a wrong column would have left every test green and every
account answering no_data_unconfirmed in production. Seeds the reported case and
the four other shapes: an empty record, another account's rows, tombstoned rows,
and a history that mixes units.
Rounds the out-of-window mean to one decimal, like every other mean the snapshot
surfaces, and annotates a failed manifest probe rather than letting it degrade in
silence.
The mutation check found both holes: an addendum test that only looked for the
new reason codes passed with the old declare-absence-and-move-on sentence still
in place, and a block-failure check whose window was wide enough to see the
neighbouring block's annotation.
The fixture is the report: a fortnight of glucose readings from April 2024 on an
account that displays them everywhere else. The clock is pinned so it cannot age
out of the window it was written to sit outside of.
Covers the four states an empty window read can resolve to, that a never-recorded
metric and a historical one differ in the SAME run, that a mean is withheld
across mixed units rather than fabricated, that the manifest advertises the range
instead of "absent", that both instructions branch on the reason code, and that a
miss carrying figures reaches the prose verifier's authoritative set. Adds the
medication-compliance page to the failed-read sweep and pins the narration
blocks' failure annotation.
A Coach retrieval tool that read the window and found nothing returned
`reason: "no_data"` — the same code for two situations that have nothing in
common: the record holds nothing for that metric, and the record holds a
substantial history for it, all of it older than the window that was searched.
Nothing downstream could tell them apart, so the Coach said the honest-sounding
sentence for the wrong case, and then, having been told there was nothing there,
changed the subject. Someone with a year of imported glucose readings was told
the readings did not exist, on the same account whose glucose page was showing
them.
The window cap stays. It exists because a multi-year history does not fit a
prompt budget, and widening it would only move the same failure. What was
missing is the ability to say a different true sentence.
- `tools/availability.ts` — one grouped aggregate per miss, no rows
materialised, resolving four states: `no_data` (probe confirmed empty),
`outside_window` (rows exist, older than the searched window; carries the
count, the date range in the user's own timezone, and a bounded per-series
mean/min/max), `unavailable_in_scope` (rows exist inside the window and the
domain still produced none — a module is off), and `no_data_unconfirmed` (the
probe itself failed; absence was never established, so it is not asserted).
- Every empty read in the executor routes through it. A bare `no_data` can no
longer be returned without the record having been checked.
- The DATA INVENTORY carries the same distinction, so a domain the window
cannot reach is advertised with what it holds instead of as absent — the model
no longer skips the fetch, and a vague follow-up keeps the subject.
- Both instructions that told the model to report absence and move on now
branch on the reason code, and forbid answering about one metric by reporting
another.
- The out-of-window aggregate joins the authoritative set for the prose
verifier, so the sentence the fix exists to make sayable is not elided.
- A cross-cutting narration block that throws is now counted rather than
vanishing into a `null` that looks like honest absence.
- The medication-compliance page tells a failed read apart from an account with
no medications, instead of inviting someone to re-enter what it already holds.
Refs #648
`Partial<typeof item>` inferred the literal types off the base fixture, so an
override could not swap the container kind or null a date. The fixture takes
an explicit row shape instead.
Hardcoding the container kind in the batch-import consumption path or in the
refund path left the whole suite green — neither had a scenario that read a
clock-less container. Both now do: a back-dated import chunk against a tablet
pack stays IN_USE with no deadline, and a refund that brings units back to a
long-opened pack reads IN_USE rather than expired. The pen counterpart of the
import case is pinned alongside.
The fake client's `updateMany` learns the `id: { in: [...] }` form the batch
claim uses; it only understood a single id before.
The no-runway arm of the low-stock evaluation now has its written-off count
asserted too.
The comparison now includes `updated_at`, so a second application that
rewrites an already-correct row is visible even when the value it writes is
identical. Without it the assertion passed whether or not the guard that
skips untouched rows was there.
The first pass of migration 0276 restored the rows the in-use clock had
already flipped, but every clock-less container still IN_USE carried the same
clock-derived `expires_at`, most of them with the deadline still ahead. The
daily expire scan would have worked through them over the following weeks and
reproduced the whole defect one row at a time.
A clock-less container's `expires_at` is now set from its printed expiry —
usually nothing, which is how it stays out of that scan. Pens and ampoules
keep theirs; their clock is real.
Submitting the correction dialog is out of reach for the SSR component
tests, so the two behaviours that only show up on submit are pinned against
the module source: a date rides along only when it changed, and a cleared
field sends null rather than an empty string. The route side of the same
contract is covered by a real request test.
The row printed the opening date and the printed expiry with nothing between
them. It carries the same middle dot the unit figures use.
The docblock now says out loud what the enum cannot distinguish: AMPOULE
covers both a punctured multi-dose vial, where the post-opening window is
real, and a carton of single-use glass ampoules, where it is not. The kind is
never auto-selected, so picking it is a statement about an opened container,
and the opening date is editable when it turns out not to be.
The supply tab rendered a container's kind, its unit count and a state badge,
and nothing else. A row could read "Expired" with nothing on screen to say
which date caused it, and the correction dialog only ever offered the
remaining-unit count — so the opening date, which the intake consumption hook
writes without the user ever typing it, could not be taken back.
Each row now carries the two dates it has, and the correction dialog edits
both alongside the unit count. Each field rides along only when it actually
changed, so saving a stock correction never rewrites a date left untouched,
and clearing the opening date sends null, which the route reads as "this was
never opened" rather than "leave it alone". Beside the opening date the
helper text says whether the kind carries a 30-day window at all.
The headline stops contradicting itself. "0 of 0 doses remaining" next to
"+ 54 units expired" read as "nothing was ever recorded" beside "stock was
written off"; when the available pool is empty and only expired stock is
left, one line says so instead. The low-stock push follows: no usable supply
with written-off units on hand is a different errand from an empty pack, and
`expiredUnits` now reaches the payload, the notification metadata and the
wide event.
Migration 0276 reverses the rows that reached EXPIRED only through the
post-opening clock on a container that never had one. It moves a row back to
IN_USE when an opening date is recorded and ACTIVE when none is, and sets
`expires_at` from the printed expiry — which for these rows is usually
nothing at all, because they recorded none.
The predicate is exact rather than approximate: nothing but the printed
expiry and the in-use clock can write EXPIRED, so a clock-less container
whose printed expiry is absent or still ahead can only have got there through
the clock. Pens and ampoules keep their EXPIRED, so does anything whose
printed date genuinely lapsed, and so does a drained container. No date is
invented and no printed expiry is fabricated.
The integration test seeds the before-state, applies the migration SQL from
disk, and asserts both what moves and what stays — the testcontainer starts
empty, so the migration running at boot proves nothing on its own.
A supply container went EXPIRED 30 days after its first use no matter what
kind of container it was. That window exists because puncturing an injection
pen exposes its contents and the product information caps how long the rest
stays usable; a blister of tablets has no such limit, every tablet sits in
its own sealed cavity, and pressing the first one out says nothing about the
rest. The state machine never read `containerType`, so a tablet pack opened
five weeks ago had its remaining stock written off.
The clock now keys off the container kind. PEN and AMPOULE carry it: both are
a single sealed reservoir that the first dose breaches. BLISTER, INHALER,
BOTTLE and OTHER do not: their units stay individually sealed, or the
arrangement is simply unknown, and a fixed 30 days would be a number nobody
entered. The printed expiry still applies to every kind, so an unknown
container is not an unbounded one.
`computeExpiresAt` takes the kind too, so a clock-less container with no
printed date now persists no `expiresAt` at all and the daily expire scan
never sees it.
The per-item PATCH gains a way back out: `markAsFirstUseAt` accepts null and
clears the opening date, which is the correction a container needs after the
intake hook auto-opened it from a back-dated dose.
Also drops `daysRemainingInUse`. It had no caller anywhere in the tree and
its docblock described a chip that does not exist.
The wrap assertion gated on the section heading, but the signals card paints a
skeleton heading first and swaps it for the resolved one — so the element could
detach between the visibility check and the measurement, and the skeleton
heading carries no action slot to measure in the first place. Gate on the
caption itself and read both rects in a single evaluate.
The Insights overview panned sideways at every phone width. SectionHeading
pinned its trailing action slot with shrink-0, which sizes a flex item to its
max-content width. That slot was designed for a compact control, but every
caller passes prose today — the provenance method caption, the device-score
subtitle, the sleep-quality subtitle. A sentence at max-content measures
1229 px in English and 1464 px in German, and since the row sits inside the
AuthShell <main> — whose overflow-y:auto makes overflow-x compute to auto —
that width turned the page into a horizontal scroller. min-w-0 lifts the flex
min-width:auto floor so the caption wraps to the row instead; the parent's
flex-wrap still moves the slot to its own line before anything is squeezed.
The existing width guards could not see this. Both measured
documentElement.scrollWidth, which never budges when <main> absorbs the
overflow into its own scroll area, and the insights case ran against an empty
mocked payload that fits any viewport by construction.
- Add e2e/utils/horizontal-overflow.ts: asserts the document AND the <main>
scroll container, and names the widest offending element on failure.
- Add e2e/mobile-horizontal-overflow.spec.ts: sweeps /insights, /, /measurements
and /medications at 360/390/393/402/430 px against populated fixtures, plus a
canary proving the measurement can see through <main>, and a check that the
insights fixture really populates the page.
- Rewrite the cycle spec's empty payload as a populated one and sweep every
phone width; its /insights case moves to the new spec.
- Point mobile-density.spec.ts at the shared assertion — it carried the same
document-only blind spot.
Three legs, none of which stands alone.
`destructive-control-guard.test.ts` is the structural one. A registry
names every destructive control, what it takes, and how recoverable that
is; three detectors then make FORGETTING to register the failure mode.
The obvious guard — every trash icon must sit near a dialog — fails both
ways and fails quietly: it misses the control that destroyed the most
here, a PUT carrying two empty strings with no trash icon involved, and
it cries wolf on the trash icon that drops a row from an unsaved draft.
Both classes are live in this tree, and a guard suppressed on half its
hits teaches people to suppress it. So the judgement lives in the
registry, where it is written down and reviewable, and the mechanism only
guarantees that a new control cannot arrive unnoticed and a removed one
cannot leave a lie behind. What it cannot do is on the record in the
docblock: it cannot prove the registry complete, and it cannot prove an
entry true. Skipping a confirmation is possible, but it costs a written
argument, and the length floor is enforced.
`confirm-button.test.tsx` proves the dismissal contract once, where it
belongs: the trigger carries no click handler, the caller's callback
reaches exactly one node, the dialog action, and cancel never sees it.
An earlier draft also asserted that a closed dialog renders no confirm
control; that passed with `open` hard-coded true, because Radix portals
do not render server-side at all. It was measuring the renderer, so it
is gone rather than left green.
`about-me-clear-scope.test.tsx` covers the finding itself: the request
carries no allergies key at all, the clearing call reaches the control
only as `onConfirm`, and the allergy field stays editable so the targeted
way to clear it survives.
The existing settings suite grows to name every newly guarded control by
slot, plus the threshold override switch, whose off position performed
the same irreversible delete as the Reset button beside it.
Every assertion was mutation-checked.
Settings → AI "About me" had one Clear control wired to
`save.mutate({ aboutMe: "", allergies: "" })`. One tap, no dialog, and
the allergy line went with the note — while the toast it raised said
"About-me note cleared", which is what the control looks like it does.
The allergy line is the first thing a clinician reads and the column is
overwritten in place, so there is nothing to restore afterwards.
Scope the control rather than only gating it: it now writes `aboutMe`
alone and omits `allergies`, which the PUT leaves untouched by contract,
so the payload cannot carry the allergy line away even if the dialog is
ever dropped. Clearing allergies stays possible in the allergies field
itself, which is a deliberate act on the thing being emptied. The
confirmation on top guards the note.
The same one-tap shape ran through thirteen other controls that
overwrite a column or drop a credential with nothing to restore from.
They route through the shared confirmation now: both Codex disconnects
(personal and the instance-wide shared account), the AI server key, the
profile photo, the threshold reset — all, per metric, and the override
switch, whose off position performed the same irreversible delete from a
control that reads as a display toggle — the source-priority order, the
dashboard layout, the Insights overview, a single target range, a
remembered Coach fact, a past Coach plan, a logged cycle day and a custom
cycle symptom.
`ConfirmButton` moves to `components/ui`, since it is no longer a
settings-local helper, and gains `ConfirmDialog`: the same dialog without
a trigger, for destructive controls that are not buttons. A switch has to
stay on when the dialog its off position raised is cancelled, and a
control that could not reach for the primitive reached for nothing —
which is a fair part of why those were the ones that shipped unguarded.
Copy lands in all six locales.
The description promised that null or an empty string clears any field.
An empty string clears the free-text ones and now gender, but email is
validated as an address and 422s on it, and the display preferences are
not clearable at all. Name them rather than let a client find out.
Adds the cases that were missing everywhere except the FHIR conformance
suite: an empty string clears without taking its siblings down, OTHER
round-trips through the profile and onboarding writes, an unsupported
value is still refused and says which field it is about, the Strain
model scores an unanswered account on neither side, the dashboard bands
give OTHER the neutral fallback, the snapshot payload carries the value
instead of flattening it, and the doctor-report cover omits the row
rather than claiming a value the account never chose.
Refs #645
The profile validator has accepted MALE, FEMALE and OTHER for a long
time, but the published contract documented two, and a client written
against that contract had no way to know the third existed. The profile
PATCH is one write, so a rejected gender took every sibling field with
it - a time-format change lost to a field the person never touched.
Gender now clears on an empty string, the same contract every other
profile field documents, and an unsupported value is refused with a
message that names the field and its accepted values instead of the
validator's own prose. Onboarding validated against a narrower enum
than the profile schema and 422'd the whole submission; its form could
not record the third value at all.
The rest is the same defect in other clothes. The Strain model scored
an unanswered or OTHER account on the male Banister coefficients; it
now uses the midpoint of both published models, matching every other
sex-split formula here. The dashboard snapshot emitted null for an
account that answered OTHER, which is indistinguishable from no answer.
Three casts asserted a two-value type over a three-value column. All of
them now carry the stored value and narrow once, at the boundary of a
reference table that genuinely has no third row, through one named
helper in src/lib/profile/sex.ts.
Refs #645
The envelope was assigned after the loop, so the loop's IN_BED skip had
no observable effect - removing it changed nothing, which makes it a
dead branch rather than a guarantee. Assign the envelope first and let
the skip protect it.
A device can write the same period twice: an original envelope and a
corrected one, same start, later end, distinct external ids. Both rows
belong to the same writer, so the writer collapse keeps both, and the
per-stage totals added them. One night served a 10 h 13 min in-bed
headline next to a 20 h 17 min in-bed legend on the same screen, and
the hypnogram drew the night twice over.
Every per-stage total is now the measure of the union of that stage's
spans rather than the sum of the rows, and the same merged spans feed
the hypnogram segments and the time-asleep headline. Stages merge
independently: two stages describing one minute is a contest, and the
writer collapse is this file's mechanism for that - arbitrating again
would collapse the summary-shaped writers, which stamp every stage
total on the sleep end instant and nest by design.
stages.IN_BED is now the in-bed envelope itself, the same union
inBedMinutes reports, so the legend cannot contradict the headline and
a night whose envelope only a losing writer recorded still reports it.
The seven documented statuses are four kinds of thing, so they get four
rulings instead of one.
Taken and Skipped state a decision about a dose and each maps onto something
already recorded here — a row with a take time, and a row marked skipped. That
second mapping was checked rather than assumed: skipped is documented as a
deliberate skip and held apart from autoMissed, which is the marker for a dose
nobody ever acted on, and the compliance re-fold counts a skipped slot in its
own column and never as taken. So a deliberate skip arrives as one.
Not Interacted, Not Logged and Unspecified are absences rather than
statements. Written as taken they would invent a dose; written as skipped they
would invent a decision, because not answering a notification is not choosing
not to take something; written as pending they would put outstanding doses
from years ago in front of someone today. Snoozed describes what happened to a
reminder. Notification Not Sent is about the software. None of the five
becomes a row.
All five are refused, but each under its own name, because "the file says
nothing about this dose" and "this row is about a reminder" are different
things and a person acts on them differently. Three reasons where there was
one, each with its own sentence in six locales.
An inactive medication's history imports and the medication stays inactive.
Preserving what happened and putting a medication back on the active list are
two acts, and only the first belongs to an import.
Three things the tests said were covered and were not.
A column flipped from honoured to ignored moved both sides of its own
assertion, because the check read the verdict off the table it was checking.
The verdicts are now written out as literals, so changing one is a decision
somebody states twice.
The preview block was unreachable from an SSR-only test, so nothing would
have noticed it rendering through the outcome view — which would classify
what a preview *would* write as a success and put a tick on a run that never
ran. It moves into its own component that takes the verdict as a prop, the way
the CSV result view already does, and the preview and the archived-row report
are asserted directly.
The account-wide import narrows its in-progress check to account-wide jobs, so
an import running on one medication's page does not refuse it. Nothing pinned
that; now something does.
Also correct two comments about the earlier replay-key fault. The repeated
value in the reported file was not a dose quantity — it was copied from the
example this dialog itself displays. The field never carried identity, and the
submission was never at fault.
The dose-history import gets its own card in Settings, because the file
covers a whole regimen and the per-medication dialog on a medication's page
covers one medication.
Preview before writing. The dry run parses the file, matches every row
against the record, and answers with the whole verdict: rows read, rows that
would land, rows refused grouped by reason, and the medication names that
match nothing — verbatim, so they can be added and the file re-run. Nothing
is written until the second button.
The column rulings sit on the card before a file is chosen, one line each,
naming whether the column is used, reported, or not read and why. A column
read and dropped in silence is indistinguishable from a column honoured, and
that is how somebody comes to believe an importer looked at something it
never did.
Outcomes render through the same primitive the other three import surfaces
use, so a run that wrote nothing cannot carry a success tick. The preview
borrows the reason lines and deliberately not the outcome sentence: a preview
has written nothing, and classifying it by the counts it would produce would
put a tick on a run that never ran.
The fixture is the export a self-hoster attached, de-identified as it
arrived: 3,395 rows, sixteen medications, three and a half years, two UTC
offsets, three statuses, two units, blank cells in four columns. A
hand-built fixture would have carried one status and one offset and proved
nothing about the file that prompted the work.
The integration pass runs it through the real derivation into a real
Postgres, because the claim is about rows: 3,387 land, 8 are refused because
the file records no decision about them, and imported plus refused equals
every row read. 2,799 of the doses were taken after they were due, which is
the fact that survives only because the scheduled slot and the take are two
columns.
Structural cover so the verdicts cannot rot: every column of both shapes has
exactly one ruling, and every parse-time refusal is spelled the same as a job
skip reason — a reason that exists in only one vocabulary renders as "reason
not recorded", which is a refusal nobody can act on.
The intake importer took a hand-written list of dates and times for one
medication. A self-hoster with a ten-column export covering sixteen
medications and years of history was left translating it by hand, which is
where the scheduled time and the time a dose was actually taken get
flattened into one — the distinction the record wants most.
The export keeps them as two columns, so read the file's own shape.
`MedicationIntakeEvent` already models that pair as `scheduledFor` +
`takenAt`; an imported dose now populates both instead of anchoring itself
on the take. `Skipped` becomes a deliberate skip with no take time, which is
how every other skip path writes it. A dose whose actual amount differs from
the amount the schedule expected carries the difference as a per-dose
override; an unremarkable amount stays absent, which is what absent means
there.
Timestamps carry their own UTC offset (`+1030` and `+0930` on adjacent rows
of one file) and it is read literally: the epoch is computed from the stated
offset rather than handed to the platform's lenient date parser, and a value
without an offset is refused instead of assumed to be the server's zone.
An export covers a whole regimen, so the job is account-wide and each
payload entry names its own medication — migration 0276 widens
`medication_intake_import_jobs.medication_id` to nullable for it. The
per-medication route is unchanged and keeps naming its medication on the job
row. A medication is never created from an import file; unmatched names come
back verbatim so they can be added properly and the file re-run.
Every column gets a verdict, and `?dryRun=1` returns the whole one before
anything is written.
The intake import built its replay key from the row's optional `zaehler`
field. That reads as a per-row counter — the documented example showed one —
but nothing required it to be unique and nothing checked. A reported export
carried the dose quantity there instead: 1 tablet, on all 28 rows. Every entry
queued under the same key, one dose reached the ledger, and the other 27 came
back reported as duplicates of a dose they had nothing to do with.
The key is the instant now, and `zaehler` is out of the contract rather than
accepted-and-ignored. Nothing is lost by dropping it: the live-row unique on
`(user_id, medication_id, scheduled_for, source)` already makes an imported
dose unique per instant, so a re-import of a file first imported under the old
key still dedups.
The telling is the other half. Two different facts were counted as one number:
entries of a submission that collide with each other, and doses the ledger
already holds. They are separate reasons now, grouped with a count, and the
dialog renders them through the same classifier and the same outcome line as
the CSV and JSON import cards — so an import that wrote nothing can no longer
answer in success tone. That rule shipped for the settings import panel and
this surface was not part of it; the guard now lists every import surface by
path, so a fourth one has to be added deliberately. `skippedInvalid` is gone:
it was hardcoded to zero on every path and reported a capability the importer
never had.
Alongside it, the other half of the environment report — a scheduled job that
fails every run has to say so. `recordError()` increments one unnamed counter,
so the admin status page could show a number and not which queue produced it,
and the module's own screen showed a zero day count with nothing to
distinguish "nothing stored yet" from "every run failed". pg-boss has held the
answer the whole time: a job that exhausts its retries keeps its queue name,
its error and its timestamp. The admin status now names the failing queues
with counts and the last error, the environment overview reports its own
account's failed runs, and neither needed a change at any of the thirty
`recordError()` call sites — a visibility fix that has to be remembered per
call site is the same failure one level up.
Refs #650, #646
Migration 0217 created the type as `environment_location_source`; the
datamodel declared `EnvironmentLocationSource` with no `@@map`, so the client
compiled every write against `public."EnvironmentLocationSource"`. Postgres
answered 42704, the upsert rolled the transaction back, and the module sat at
"0 days recorded" from the day it shipped — one enum out of forty-nine, and the
only one without a map.
No migration follows. `prisma migrate diff` against a database built from the
committed migrations reports the enum rename and the `environment_contexts.source`
column rebuild before the map and nothing after it: the database was never the
side that drifted, and renaming a live type to match a datamodel typo would be
the wrong direction.
Two guards, because the two questions are different. The unit guard replays
CREATE TYPE / ALTER TYPE RENAME / DROP TYPE across the migration set and holds
every enum to a type that survives it. The integration guard compares all three
name classes Prisma emits — enum type, table, column — against a real migrated
Postgres, which is the only place a name mismatch is observable at all: it
survives generation, typecheck, and every mocked test untouched.
Refs #646
Mutation-checked against the production paths they cover: the escalation
verdict, the leg guard in `recordSyncSuccess`, the streak anchor the verdict
dates itself by, the status envelope threading it, and the pill's relative
timestamp.
Also give the four legacy per-provider status routes the verdict the
consolidated envelope already carries. They answered `connected: true` plus a
`state` string and nothing else, so a client reading one of them saw a live
connection over a pipe that had delivered nothing for weeks.
A sync failure the classifier bucketed as `transient` had no exit. The park
test read `kind === "persistent"` and the persistent streak anchor was only
ever stamped by a persistent failure, so a streak could run for a thousand
attempts with the persistent counter at zero, the state frozen at
`error_transient`, and no ladder it could climb. `transient` is a hypothesis
about retryability, and the classifier defaults every timeout, 429, 5xx and
unrecognised response to it precisely because the wire cannot tell a reboot
from a permanent break. That default is right for one attempt and wrong for a
thousand: an unbroken streak retried hourly for a full day has refuted it by
experiment. Record it as `persistent` from there, which feeds the park ladder
that already exists — 48 h of unbroken failure end to end before an
integration stops attempting.
The counters could not accumulate anyway. `IntegrationStatus` is keyed
(user, integration) while Withings runs four legs on four crons and WHOOP four
resources on four more, and a success cleared the error, every bucket, the
streak anchor and the state whatever leg had recorded them. The ECG tick at
:41 wiped what the sleep tick recorded at :15 — including a reauth park — so
the strike ladder was reset several times an hour and the card painted a green
"connected · 12 minutes ago" over a data class that had stopped arriving. A
success now clears only what it is entitled to clear.
Two columns carry it: `failing_since_at` dates the streak, `failing_leg` names
its owner. The migration backfills the anchor for rows already in an error
state so a connection broken for weeks escalates on its next attempt rather
than restarting its clock.
The defect is about rows that survive, so a mocked client proves nothing. The
new integration test seeds a row into every table in the schema, driven by
information_schema and pg_constraint rather than a fixture, runs the wipe and
asks the database what is left. A table added later is seeded and asserted
without anyone remembering to.
It holds all four halves: nothing in the wipe list survives, everything in the
exemption list does, everything reached only by a database cascade is gone,
and the account row comes back reset but able to sign in. A second file pins
the allergy and checkup-reminder deletes actually removing their row.
The reproductive-health note gains a Deletion section. It taught people to
reason carefully about a host being imaged and never said what pressing
delete does; now it says which of the three deletions removes the row and
which only hides it.
The nightly tombstone purge asked for `deleted_at IS NOT NULL AND deleted_at
< cutoff` and the audit-log purge asked for `created_at < cutoff`. Neither
predicate had an index behind it: every deleted_at index in the schema is
partial on IS NULL, the live-rows predicate, and every audit_logs index leads
with user_id or action. Both purges planned a sequential scan and then paid
index maintenance per removed row, in one unbounded statement, on a
connection carrying statement_timeout=60000. Past a certain backlog the
statement aborts, the transaction rolls back, nothing is purged, and the next
night runs the identical statement against the same rows. The mechanism that
gives deletion a horizon could not close on the instances that needed it.
Migration 0277 adds the indexes the predicates ask for; the three tombstone
ones are partial on IS NOT NULL, so on a healthy instance they cover only the
tombstones and shrink as the purge drains. The purges now walk the backlog in
batches of 5 000 by primary key, capped at 200 000 rows a run.
Both handlers used to swallow every error into a wide-event warning and
return, so pg-boss recorded a successful job either way: a purge failing
every night looked exactly like a purge with nothing to do. They now rethrow,
and a run that stops at the cap records that a backlog remains.
The wipe deleted thirteen tables against a schema that carries more than
eighty user-scoped ones. Mental-health assessments, lab results, the
menstrual record, allergies, family history, the document vault, every AI
conversation and live clinician share links all survived a confirmation
that promised to permanently delete all health data. So did
MeasurementRollup, the per-day reconstruction of the measurements the same
transaction had just removed.
The list now lives in src/lib/data-wipe/wipe-plan.ts next to its own
exemption list, and a structural guard enumerates schema.prisma and fails
when a model belongs to neither. Adding a model can no longer widen the set
of things that survive a wipe by default. The same guard covers the User
row's own columns, which had drifted the same way.
The boundary is explicit: everything the account owns goes, except what is
needed to sign in again. The confirmation and the toast say so before the
person types the confirmation rather than after.
Checkup reminders and allergies promised permanence over a soft delete with
no restore path; both now hard-delete, so the dialog is true. The lab
biomarker dialog was checked and is already accurate — it maps to the
hard-deleting /api/biomarkers/[id] route, not to the soft-deleting lab
result route.
Every entry in the display-transform registry declares the decimals its
unit is read at. That number was carried to the render layer and read by
exactly one screen out of all the ones that show a converted value, so
everywhere else the raw product of a kilogram and 2.20462262185 reached
the surface. Pounds came out with a dozen digits it never needed.
Move the rounding into `applyDisplayTransform` and
`applyDisplayTransformDelta`, so a surface gets a correctly-scaled number
whether or not it remembers to ask. Patching the weight cell would have
left the same trap behind inches, Fahrenheit, miles per hour and miles.
Two boundaries stay exempt and are documented as such. A transform that
does not rescale is the exact identity and is never rounded, so a stored
78.45 kg keeps rendering as 78.45 on the metric branch and every
untransformed metric is byte-identical. `invertDisplayTransform` — the
entry boundary — keeps full precision, because a value rounded on the way
out and inverted on the way back would drift the stored canonical number
a little further on every save; the entry surfaces quantise to their own
two-decimal canonical dialect instead.
`applyDisplayTransformUnrounded` is the one documented escape hatch, for
arithmetic whose result is not rendered. The target adapter needs it: its
guardrails round inward, and rounding the 30 kg floor to the nearest
tenth of a pound first gives an edge that inverts to 29.98 kg and the
server rejects it.
The weight page and the HealthKit scaffold were folding the factor into
their stat strips by hand. Both now convert through the helper, so the
raw multiplier is only ever handed to the chart, which owns its own
series and formatting.
Formatting is unchanged. Precision is fixed at the conversion and the
surfaces keep the formatter they already used, so switching preference
changes the unit and nothing else about how the number looks.
Refs #627
The scope picker arrived with the standard doctor's report already ticked,
so someone who opened the panel and pressed Generate sent material they
never chose. Nothing is selected now until a person selects it.
The template keeps its definition and loses its automatic application: it
sits on a button at the top of the picker, named, on both surfaces that
mount the picker, and it stays there so the standard set can be restored
after experimenting. One click still gets a complete doctor's report.
Generate and the share-link create refuse an empty scope, with the scope
line beside them saying what to do rather than reporting a failure after
the press; the line also describes the control, so the disabled state is
not silent for a screen reader.
The root layout shipped the active locale's catalog as `<script defer>` and
counted on it executing ahead of Next's hydration scripts. It does not: Next
emits its own chunks as `async` tags AHEAD of anything the layout renders into
`<head>`, and an `async` script runs the moment its fetch settles. So the
370 KB catalog was racing the app bundle, and on a cold load it lost about a
third of the time — `load-locale` initialised with an empty cache, every `t()`
in the hydration render returned its raw key against the server's real text,
and React tore the tree down with a text hydration mismatch. The user saw a
frame of `nav.skipToContent` before the provider's backfill landed.
Register the catalog through `next/script` at `beforeInteractive` instead.
Next's own bootstrap awaits those scripts before requiring a single app module,
and pairs the registration with a `<link rel=preload>` so the download still
starts with the document rather than after it. The CSP nonce rides along, so
the posture is unchanged.
Fold the boot global into the cache lazily rather than snapshotting it at
module evaluation. The snapshot made correctness depend on an ordering the
module cannot see; reading on demand reduces the invariant to something local
and testable — whatever the boot script assigns is visible to the first `t()`
that asks, whenever it lands.
Measured on the dashboard render spec, 30 repeats per project against a
production build: 11 of 30 failed on desktop before, 0 of 60 after.
Everything on that page is server-rendered, so a row is visible long before
the client boundary that owns it is live. A chevron click landed on inert
markup about one run in four, and a `boundingBox()` occasionally came back
null because the node had just been replaced.
The page is one client boundary carrying the export panel and the sharing
card, so the sharing card's own query is a sound signal that the surface is
wired. One helper opens the page and waits for it; every case on that route
goes through it, including the disclosure case that was already doing this
by hand.
Three of the picker assertions resolved to two elements and failed strict
mode; they now name the export mount, and a new case pins that each mount
carries exactly one of every id the picker renders.
The included-data disclosure case asserted a collapsed first render. The
panel opens on a first run on purpose — that is the only place the standard
template is applied, and a template behind a closed disclosure is opt-out
wearing a hat — so the case now asserts the contract that replaced it: open
on first render, and still a disclosure, closing and reopening under the
same control.
Settings → Gesundheitsakte mounts the one scope picker twice: the export
panel at the top, the share-link create form under `#sharing`. Sharing the
component is deliberate, but it also meant every test id inside it resolved
to two elements on that page, so nothing in the picker could be addressed
without guessing which one you got.
The picker now takes a required `surface` and every id under it ends with
that surface — `report-group-row-vitals-export` and
`report-group-row-vitals-share` are two controls and read as two controls.
The summary line beside each picker follows the same rule. A third mount
picks a value and stays addressable by construction.
The capability descriptor named `fhir:read` as the scope a narrow token needs
for the REST face. No code path mints a token carrying it — every mint grants a
wildcard, the medication ingest scope, or an MCP scope — so the face is
reachable by the owner's cookie session or a wildcard device token and by
nothing else. Advertising an unobtainable credential invites a client to go
looking for one. The constant stays where it belongs, as the fail-closed check
each route makes; the descriptor now reports the fact instead.
End-to-end coverage for the scope picker: the first run opens it expanded with
the fenced tier rendered and no group control over it, a group row clears the
44 px floor at 390 px, and a leaf grid opens from its chevron. All on stable
test ids rather than text a breakpoint can hide.
Migration 0273 revokes every live record share link minted before the selection
model, because their frozen scope was "everything except mood and cycle" — a
scope nobody chose. The owner has to see why the link stopped and be able to
replace it without retyping what they already decided, or a deliberate closure
reads as a fault. The list now names that state, says why in one sentence, and
offers a replacement that carries the label, the window and the lifetime
across, leaving exactly the decision that link never carried.
Writing the test for it found a defect in the picker. The group count and the
group checkbox read the catalogue's group rather than the leaves the surface
actually renders, so on the share-link form — where the insurance leaf is
hidden because the server refuses it — the identity row read "1/2" and drew
itself half-selected over a leaf the owner could neither see nor reach, and its
checkbox would have tried to add it. The three helpers take the rendered leaves
now, so a control cannot describe more than it offers.
Two downloads under the token — the FHIR document Bundle and the clinical PDF —
behind the same unlock cookie the document route enforces, serving exactly the
link's frozen selection. That is what the removed FHIR toggle promised, in the
form a practice actually uses: open the link, unlock it, save the file. No new
credential type, no new scope, no mint path that does not exist.
The gate is the document route's, step for step: the same live check, the same
token-scoped cookie, the same blunt 404 for every miss class so a probe cannot
tell "no such link" from "revoked" from "documents-only". It differs in one
number, because generating a bundle is far more expensive than serving a stored
blob and must not share a budget with it. Each download writes its own audit
row. The scope cannot exceed the page's: both resolve the same frozen leaf list
through the same resolver, and the insurance leaf is refused at create, so the
insurance number cannot reach this path at all.
The MCP doctor-visit summary stops claiming presence it cannot support. An
account whose owner never saved a report selection resolves to the empty one on
that surface, and an empty scope produces an empty payload — which now reads as
{ present: false, reason }, the same sentinel the other reads use, instead of a
record with every section quietly missing.
Mutation-checked: removing the unlock check, the rate limit, the documents-only
guard or the audit row reddens its own case; pointing the bucket at the document
route's key reddens the bucket assertion; making presence unconditional reddens
the absence case.
The saved profile's shape moves out of the module that reads the row. The panel
is a client component and parses the blob from the /me payload; importing the
reader dragged the database client into the browser bundle, which typecheck and
the unit suite both pass and only `next build` catches.
Five layers, because the defect had five places to hide.
The catalogue guard checks what the compiler cannot state: every leaf resolves
a label a human can read, every group resolves one, the fenced tier stays last,
and the shipped template intersects the fence at nothing.
The selection guard pins the three minting functions against each other. A
stored blob this build cannot read resolves to the empty selection, never to a
default set — that is what makes the revoked legacy links revocable rather than
silently re-scoped. A request naming an id the server does not know is refused
by name rather than dropped, because dropping it would narrow a scope the
caller did express.
The egress guard freezes the set of aggregator callers with the source of each
one's selection written down. The required brand-typed parameter already stops
a new path compiling without an answer; this stops one being added without a
human writing the answer down, which is the moment a reviewer asks the
question.
The behavioural sweep aggregates every leaf in and out against a fixture that
carries a row for each, and asserts the payload differs. It is parametrised
over the catalogue, so a leaf added later is covered the day it is added. Its
positive control asserts the all-on payload is genuinely populated first, so a
"no difference" result means the gate is inert and not that the account is
empty. Beside it: an unmapped measurement type is excluded rather than passed
through, no table behind an unchosen leaf is queried at all, and an empty
selection produces an empty record.
The artefact layer runs the real route and reads what comes out — the FHIR
bundle, the PDF text and the zip's contents — with a positive control beside
every exclusion.
Mutation-checked: making the gate admit everything reddens the sweep for every
leaf; passing an unmapped type through reddens the fail-closed case; dropping
the mood query's condition reddens the zero-read case; returning the BMI
unconditionally reddens the sweep naming BODY_MASS_INDEX; adding a fenced leaf
to the template reddens the purity guard; removing the panel's seeding block
reddens the first-run and repeat-run cases; dropping the insurance refusal
reddens the share-link case; adding a caller outside the frozen list reddens
the egress guard naming the file.
The panel suite grew past the practice name it was written for and is renamed
for what it now covers: the three states, the fenced tier having no group
control, and the scope line naming its sensitive inclusions.
Integration coverage moves to the leaf model, including the case this release
exists for: an account with data in every domain and a caller that named
nothing gets an empty record.
The OpenAPI contract catches up with the routes — the required selection, the
refused keys, the share summary's new fields, the moved `Patient/$everything`,
the Coverage search and the five typed reads.
The export panel rendered twenty flat switches behind a collapsed disclosure,
and the share-link form rendered none at all — it posted whatever the create
schema defaulted to. Both now mount the same picker: twelve named groups, a
count on each, the leaf grid one chevron away, and a fenced tier below a
separator.
The fence is mechanical, not a label. The sensitive tier has no group
checkbox, so no single control anywhere in the picker can turn on more than
one of mood, cycle, family history, PHQ-9, GAD-7, WHO-5 or SCI at a time.
There is no select-all. The shipped template contains none of them. And the
scope line under the button restates every sensitive inclusion by name rather
than folding it into a count, because a number cannot be checked at a glance
and "incl. Mood, PHQ-9" can.
First run opens expanded with the named standard template applied and the
fenced tier visibly empty, so the user presses Generate having seen twelve
named groups. Every later run opens collapsed carrying their own last
selection, which makes the waiting-room path two interactions.
Only leaves are ever stored. A group's checkbox is derived from its leaves, so
the two cannot disagree, and switching a group off and back on inside one
session restores the pattern instead of flooding it.
The cycle export card is deleted. It posted empty groups under a docblock
claiming every other section was off, and empty groups resolved to the
defaults — so a "cycle only" download carried weight, blood pressure, pulse,
oxygen saturation, body fat, body composition, the medication list, adherence,
labs, allergies and family history. Cycle is a fenced leaf in the one picker
now, reachable for the PDF as well as for FHIR, which is more than the card
offered.
The panel gains a 30-day preset and a custom window, and its error line says
what went wrong instead of interpolating a status code.
The Checkbox primitive learns the indeterminate state it already had the Radix
support for. A partly-selected group drawn as unchecked would say "none of
this is included", which on this surface is a false statement about what
leaves the instance.
The report resolved an absent key to "included". A caller that sent no
sections object, or an empty group inside one, received the full record at
server-chosen defaults: roughly fifty metrics plus the medication list, labs,
allergies and family history. The web panel sent every key explicitly so it
looked fine from the browser, but the share link, the FHIR REST face, the MCP
doctor-visit resource and the cycle-only export card all went through paths
where absence meant yes. No arrangement of booleans fixes a default that means
yes, so the shape goes.
A selection is now an inclusion list of leaf ids. A leaf is the smallest thing
that can be released or withheld: every measurement type, and thirteen
structured sections — identity, insurer, the glucose panel, labs, the drug
list, the administration ledger, adherence, the GLP-1 block, allergies,
illness episodes, family history, mood, cycle. Membership is inclusion,
absence is exclusion, and there is no defaults constant anywhere to drift.
Three locks keep a metric added later off a report chosen earlier. A stored
selection is a list of what was chosen, so a leaf that did not exist then is
not in it. The catalogue is keyed over the measurement enum, so adding a
member without deciding its group fails typecheck in the same commit. And the
aggregator's filter inverts: a type the catalogue does not know is excluded and
annotated, where it used to be passed through.
The aggregator's third parameter is now required and brand-typed, which turned
the three surfaces that had no scope at all into compile errors. Each one had
to name a source, and each now replays the owner's saved selection — the same
object reviewed in Settings. An account that never saved one gets nothing
there rather than a guess.
Identity and insurance become selectable, and a share link refuses the
insurance leaf outright. The share view never decrypted the number; that is
structural now instead of incidental. The KVNR is not even read unless the
leaf was chosen.
Two surfaces gain a record they always implied. A share link serves its frozen
selection as a FHIR bundle and as the PDF, behind the same unlock cookie the
document route uses, with its own rate bucket, its own audit row and the same
blunt 404 for every miss. That is what the removed FHIR toggle promised, in
the form a practice actually uses.
The AI summary leaves the report. Its text came from the cached briefing with
no relation to the selection, so its scope could not be proven, and prose whose
scope cannot be proven has no place on a document whose whole point is a proven
scope.
The PDF table stops being a nine-type whitelist. It prints every selected leaf
that has data, sub-headed by group, in the same order the panel shows — so
selecting resting heart rate no longer means seeing it in the bundle and not in
the PDF. The clinician view is grouped the same way and reads real labels
instead of "Phq9 score".
Migration 0273 adds the saved-selection column, drops the orphaned prefs column
and the two retired share-link scope columns, and revokes every live record
share link minted before this model. Their frozen scope was "everything except
mood and cycle", which is a scope nobody chose; narrowing it would still be a
scope nobody chose. Documents-only shares are untouched. The sharing list shows
each closed link in a pick-again state with its label, window and expiry kept.
The retired allowFhirApi and resourceTypes stay on the create and list
responses as the constant false and empty list. The request refuses them. They
come off the response once the native client confirms it does not decode them.
Six files are split at seams this change already touched: the aggregator into
its gate, its dense-bucket query, its series, its glucose panel, its medication
ledger and its clinical records; the export route's three emitters into their
own module; the clinician view into its sections, its document list and its
downloads.
Two layers, because the defect had two halves and each needs its own kind of
proof.
The structural guard walks the panel's control shape, folds every control
through the wire schema and the resolver the route uses, and fails when
flipping one changes nothing in the resolved selection. It also checks the
panel renders a row for every control it declares, and that every resolved
flag is read downstream, either by gating measurement types or by name in the
aggregator. That is the link nobody had written: the dead switches were dead
for several releases and no test connected "the panel renders this" to "the
export honours this".
The behavioural half runs the real export route and reads what comes out. The
previously-dead measurement switches are asserted against the FHIR bundle and
the PDF table, with a positive control beside each exclusion so nothing passes
on an empty fixture. The screening totals are asserted absent for a caller
that names other sections and for a caller that sends no sections at all, then
present when asked for by name. Allergies and family history are asserted
absent from the bundle and never read from the database, and the package case
opens the zip and checks the bundle inside it agrees with the PDF beside it.
Mutation-checked: reverting the allergy gate, the medication-list gate, the
screener entries or the measurement entries turns the matching cases red and
leaves the rest green; removing a control's fold entry or its aggregator entry
turns the guard red with the control named.
The remaining changes update fixtures the new shapes touch: prefs objects now
resolve the added flags from the defaults, and the share-link fixtures drop
the retired FHIR scope, with a case pinning that the create route now refuses
those keys instead of ignoring them.
The export panel rendered twenty data switches and eight of them changed
nothing. The grouped selection folded down to twelve flat flags and dropped
oxygen saturation, body fat, body composition, resting heart rate, HRV, VO2
max, steps and distance on the way, so a user who switched any of them off
still exported the readings in the PDF, in the FHIR bundle, in the package,
through a clinician share link and over MCP. Each one now resolves to a flag
the aggregator reads, and the rows are excluded at the query rather than
fetched and dropped.
Three more controls were untruthful in their own way:
- The medications switch documented itself as excluding all medication data
and excluded the adherence figures only. The drug list and the per-dose
administration ledger rode along regardless. Both follow the switch now;
the rows are still read, because compliance and the GLP-1 block are computed
from them, and they no longer leave the aggregator.
- Allergies and family history were gated for the PDF and fetched
unconditionally for the FHIR bundle, so a package could carry a compliant
PDF and a violating bundle in the same zip. The reads are gated, and a
deselected section is not read at all.
- The PHQ-9, GAD-7, WHO-5 and SCI screening totals had no per-report control.
They rode the mental-health module alone, which has been default-on since
v1.29.1, so depression and anxiety screeners reached FHIR bundles, share
links and the MCP doctor-visit resource on accounts that never chose to send
them. They get their own section key, defaulting off, with the module gate
kept as the second lock, and the stale privacy-by-default comment that
argued otherwise is corrected.
The share link's FHIR promise is removed rather than implemented. The create
form offered a toggle and a resource-type picker, the values were persisted
and badged on the finished link, and the API description called the toggle a
REST reachability switch, while no route has ever answered a share token with
FHIR or any other machine-readable format. Serving one is a feature with real
security surface and does not belong in a correctness fix. The toggle, the
picker, the badge and both claims are gone; the create schema rejects the two
keys instead of accepting and ignoring them; the columns stay, unread, so no
data is destroyed to undo a promise.
Two files are split along seams this change already touched. The
measurement-type gating maps and the filter that reads them leave the
aggregator for their own module, so the query gate and the output gate cannot
disagree; the export panel's control shape and the selection it builds leave
the component, so the selection contract can be extended without the
rendering. Three accepted-and-discarded wire keys (activity energy, GLP-1,
side effects) are dropped: nothing rendered them and nothing read them.
A spec that arms `page.route(...)` in a `beforeEach` and never drops the
handler races its own teardown. The context closes, an async handler is
still awaiting, its next call rejects with "Test ended.", and Playwright
attributes that to the test that had just passed. With `retries: 2` plus
`failOnFlakyTests`, which spec bites is a matter of timing.
`e2e/setup/test.ts` re-exports Playwright's `test` extended with an auto
fixture that unroutes every page and the context after each test, and
clears the page event listeners a spec subscribes to; `expect` rides
along so a spec needs one import. Auto means a spec author cannot forget
it and does not have to know it exists. Every spec now takes `test` from
there, and `e2e-route-cleanup-guard.test.ts` keeps the import boundary
closed with an allowlist that carries its reason.
The per-file `afterEach` on the cycle spec goes away — leaving both would
read as not trusting the fixture.
Not covered: a page inside a context a spec builds itself. Only
`share-documents.spec.ts` does that, and it owns those contexts.
A HealthKit blood-glucose sample carries no context either — the mapping has
no field for one — so the batch ingest was building the same row the old
CHECK refused. Fold that source into the case that pins the contextless
shape.
Refs #640
The example CSV is the file people copy. It showed a glucose row with a
FASTING context and nothing else, which is the shape a meter produces, not
the shape a continuous sensor produces. It now carries both.
Refs #640
Same shape as the CSV and JSON cards: reaching the done state rendered the
success tick whether or not a single row was upserted. The archive's count
picks the tone now, and an archive that wrote nothing says so.
The clinical-record count stays out of that decision. Skipping
electrocardiograms and lab documents is a documented, deliberate exclusion,
not a refused row, so it must not read as a warning.
Refs #640
A CSV import of a continuous-sensor export wrote nothing and rendered a
green success state. Three faults compounded, and the reporter took all
three apart before filing.
The parser refused every BLOOD_GLUCOSE row with a blank glucoseContext.
Behind it, the CHECK constraint from migration 0021 read both ways at
once: a glucose row MUST carry a context, every other type MUST NOT. The
second half is right and stays. The first was written when the only
glucose writer was the entry form, where a person picks fasting or
post-meal at the moment of the reading. Every machine writer that
arrived since has no such answer per sample, so the Nightscout pull, the
MCP log tool, the Telegram capture and both bulk importers were all
building rows the database then refused.
A missing context now stores as NULL, never as a fifth enum member.
Every per-context reader selects by equality against the four members,
so the reading stays out of the doctor report's fasting rows, the
matching FHIR observations and the per-context targets, while the chart,
the all-time summary and the clinical panel keep it in full. Hand entry
still asks for a context, because at that moment you know it.
The import cards now read the counts rather than the status code.
Nothing written is a failure, a mixed file is a warning, and the refused
rows are grouped by reason with a count instead of fifty identical
lines, with the per-line detail one disclosure away. Both cards render
through one outcome table, so a tick cannot reach a result that wrote
nothing.
Refs #640
`medication_intake_events.idempotency_key` carried a GLOBAL unique from
0001_init. The value is client-supplied, so two people running the same
client can mint the same key — and on a global unique the second one is
simply refused by the constraint. Their dose is not written and nothing
tells them so. The bulk drain made it worse: its replay probe read the
key with no user scope, resolved the first user's row, and reported the
second user's entry as a duplicate carrying a stranger's row id.
Migration 0273 replaces the global unique with `(user_id,
idempotency_key)`, the grain every other client-supplied identity on the
table already uses. Both unscoped reads in the bulk drain now key on the
composite; the other four reads were already narrowed by user.
On existing data this is a relaxation, so nothing needs reconciling: the
old constraint is strictly stronger than the new one, and no pair of rows
can violate an index the previous one already forbade. The migration
still carries a reconciliation step for a database whose global index was
dropped by hand at some point. It never deletes or rewrites a dose — it
clears the dedup HINT on the later row of a colliding pair and keeps
every intake event, timestamp, attribution and inventory stamp intact.
The new index is created before the old one is dropped, so a failure
aborts with the old protection still in place.
A structural guard freezes the read side: no `medicationIntakeEvent`
query may filter on the key without narrowing by `userId`.
The dedup key on the medication intake surfaces is not only `externalId`.
`MedicationIntakeEvent.idempotency_key` is a `@unique` column persisted
with the dose row forever, with no TTL and no sweep, matched by plain
equality on the per-medication POST, the bulk drain and the external
ingest bridge. That makes it an identity, not a replay token, and it
carries the same failure mode: a key that rotates per client launch never
matches its own earlier post, so every retry logs the dose again. The
blank case is sharper still — the column is unique table-wide, so an
empty key resolves to whichever unrelated row stored it first. Refuse
both shapes at the schema on the single-entry paths and per entry on the
bulk drain.
The MCP write path hashes a host-supplied key into the `externalId` half
of its dedup key, so the shape is laundered before anything downstream
could see it and the check has to run on the input. That closes a live
one: every tool handler falls back to `""` when the host omits the
argument, and `""` digests to a constant, so each keyless write after the
first reported `already_logged` and was silently dropped.
The HTTP `Idempotency-Key` header stays out, and now says why in the
guard: `IdempotencyKey.expiresAt` bounds it and the cleanup job sweeps
it, so it is a windowed replay token and rotation is its design.
Extend the structural guard with a provenance rule for identities that
never pass a Zod request schema — a spreadsheet column, an MCP tool
argument. Inside the import and MCP trees, a module handling an identity
must apply the floor itself or import a module that does. That is a real
chain rather than a list of known offenders: a new importer that mints an
`externalId` from file input and imports nothing checked fails it, with
no allowlist to escape into.
Harden the structural guard so it reads the file body with its imports
stripped: grepping the raw source cannot tell a file that APPLIES the
check from one that merely imports it, so deleting the call while
leaving the import slipped straight through.
Add the partial-success case on a second batch envelope — the mood bulk
ingest echoes the offending entry back with its index, status and
reason while the good rows still upsert.
An external id is half of every ingest dedup key — (userId, source,
externalId), (userId, externalSource, externalId), (userId, type,
source, externalId). Re-posting the same triple is idempotent and
returns the existing row, which holds only while the id is stable
across client launches. A client that serialised an opaque platform
object with its default object description sent a per-allocation
memory address instead, so no sweep ever matched its own earlier row
and each one minted a fresh record: a live instance grew 23 phantom
medications in a day, none of which could collect a dose, because the
dose side resolved through the same rotating string.
Add one shared classifier for identifiers that provably cannot be an
identity — an angle-bracketed object description carrying a hex
pointer, a bare 0x address, and a blank value — and wire it at every
surface that accepts a client-supplied externalId.
Single-entry creates refuse through the schema, so the client gets the
standard 422 multi-issue envelope with a message naming the shape it
sent and what to send instead. Batch and bulk surfaces classify per
entry and report the offending row as skipped with reason
unstable_external_id, leaving the rest of the payload to land: one bad
row must not stop a sync. Every refusal carries pinned wide-event meta
(count, shape, surface) and never the id itself.
A structural guard freezes both routes into the floor: a request-side
externalId declaration either applies the field check or carries a
marker naming the file that checks it per entry, and the guard verifies
that file really does. The backup-restore payload is the one exemption,
with its reason recorded — it replays ids the server itself wrote.
Whether a column has a reader anywhere in the tree is a whole-program property,
and grep is the only cheap approximation. Grep is unsound both ways here:
generic names match unrelated identifiers and read as consumed, while
consumption through a select-less read plus whole-row serialization reads as
unconsumed. A gate built on that would drift towards green and stop meaning
anything, which is the failure this whole line of work exists to prevent. So it
stays a review tool with a human reading the output, run from the deep-audit
step and never in CI. The script says so itself and prints the false-positive
sources under every run.
It also carries the convention that makes the tool useful: a change that adds a
column, payload field, or preference contains its consumer, or the schema
comment carries `@internal` for something deliberately never surfaced or
`@pending-consumer(#issue)` for one still coming. The second marker is
re-listed on every run, so a follow-up that never happened surfaces at the next
release audit instead of never. Recorded in the contributor notes alongside the
two questions that are exact enough to be gates.
Every finding behind this work is a two-ended change whose ends shipped in
different releases with nothing holding them together, and every gate the repo
runs proves one end: typecheck proves the export exists, knip proves it is
imported, the OpenAPI check proves the schema matches. None of them notice when
the second end never arrives.
Two questions here can be answered exactly, so they become tests.
The account endpoint rides every app boot and carries the user's preferences.
Its field names are read out of the response literal and each one must be read
by a client module — property access or destructuring, comments stripped,
because a doc comment claiming a consumer is what the remembered practice name
had for its whole dead life. That field would have failed this on the day it
shipped. One field is allowlisted today with its reason written out: the
insurer institution number has no web surface at all, which the entry keeps
visible rather than papering over.
The measurement backup select is now a named constant, compared field by field
against the model in the schema. Both sides are literals, so the comparison is
exact — no name heuristics, nothing that can drift towards accepting. A column
added to the model and forgotten by the backup fails at the commit that adds
it, which is how the aggregation provenance went missing. Scoped to the one
model; extend it as each further model becomes backup-relevant.
The reverse-coverage guard asserts that every key in the bundle is reachable
from a call site. It had stopped doing that for most of the bundle. A template
whose only static anchor was a namespace root registered that root as a subtree
base, so a single per-provider OAuth key marked every key under `settings.*`
reachable, and the same shape did it for `measurementReminders.*`. A cache-key
string that happens to start with a namespace word had quietly done it for
`insights.*` too. The guard stayed green throughout, which is the part worth
noticing: a pair guard degrades into a one-end guard without any signal.
Extract the matcher into a helper with its own unit tests, and change what it
accepts. A template carrying both a literal prefix and a literal suffix now
records the pair and requires both ends to match, so the OAuth template covers
the provider keys and nothing else. A template with no suffix whose anchor is a
bare namespace no longer whitelists anything — it fails the guard by name, with
the two honest remedies in the message. The same door is used by the template,
assignment and concatenation arms, so none of them can reopen it. Template
bodies are read by scanning forward from a known opening backtick rather than
by pairing backticks across the tree, which was drifting and dropping whole
files out of the scan.
Three call sites build their keys whole instead of interpolating onto a
namespace root: the two measurement-reminder due helpers, the mental-health
option labels, and the OAuth error copy, which now names its per-provider base.
Two audit-log action names that merely look like keys carry a written
exemption; the API error catalog is a runtime lookup and is allowlisted with
its reason.
With the matcher honest, 149 keys had no call site left. They belong to an
explainer popover, a hypnogram, a passkey table and an insights hero that were
all rebuilt or removed, and they are deleted from all six locales. The
explainer guard stops pinning its title leaves, which no surface has rendered
since the popover went.
The disaster-recovery backup selected every Measurement scalar except the
aggregation fields, so a restore brought every `stats:` aggregate back with no
provenance and the Apple Health authority ladder came back flat: an export.xml
source-day estimate could then overwrite a native HealthKit statistic.
`aggregationProvenance` now rides the select, the disaster-recovery
serialization, the payload schema and the restore mapping, following
`externalSourceVersion` exactly — including staying out of the reduced
portable shape, which carries the human-readable subset only. The schema field
is optional, so every backup written before this parses unchanged.
Restoring one of those older backups no longer recreates the ambiguity
migration 0263 repaired: an Apple Health row under a `stats:` identity with no
provenance comes back as LEGACY_UNKNOWN, the sentinel the cumulative drain
already writes in that situation, so a native sync or archive reprocess can
repair it. The gate is scoped as 0263 scoped its backfill — other providers
mint `stats:` identities for their cumulative dailies and have never carried
provenance, and ordinary point rows keep NULL.
Migration 0263 added three columns beside the Apple Health `stats:*`
authority ladder. Only `aggregation_provenance` is ever read back: the
external-measurement reconciler selects it and refuses a lower-authority write
against a higher-authority canonical row, and the cumulative drain reads it
too. That column stays.
`aggregation_contributor_count` and `aggregation_selected_source_hash` are
written by three call sites and read by none. No select, no raw SQL, no export,
no backup, no API contract, no repair path. The import-time selection keeps its
deterministic tiebreak, which compares an in-memory map key computed per flush
rather than a value read back from the table.
Both are plain nullable scalars with no index, unique, check or foreign key, so
the drop is a catalog-only operation that touches no measured value.
The doctor-report contract has always described the practice / clinic name as
a stored preference, and `/api/auth/me` has returned
`User.lastReportPracticeName` since it shipped, but nothing ever wrote the
column. Every report asked for the clinic name again.
The export route now persists the sanitised name after the requested artefact
has actually been produced, so a generation that throws cannot overwrite the
remembered value with a half-typed one. Clearing the field persists null, which
keeps the cover line cleared next time. The export panel seeds its input from
the auth payload during render, gated per user id, so a late resolve never
clobbers a field the user is already typing into, and a finished export
invalidates the cached payload so a remount reads the new value.
Interface housekeeping. The overview sections drop their hand-rolled card
frames for a shared shell with one inner rhythm, failed reads on dense metric
pages get a named row primitive that puts the message in the content tier, and
the last off-scale spacing and text sizes go back onto the scale.
The tag chip and the sub-twelve-pixel text step are primitives now, so neither
can be hand-copied at a slightly different value again.
The stage-composition chart card was the last insights tile with a
hand-rolled title row. It now uses the canonical header primitive; the
subtitle and the window toggles keep the shape they had, since the toggles
have to drop to their own row on a phone.
Also sweeps the last off-scale spacing: the two dense tiles inside a
recommendation move onto the sanctioned inner-tile step, and the not-found
button onto the scale. The banned twenty-pixel step is now absent from the
tree. The chart body is untouched.
Seven overview sections had hand-rolled the same card literal, and the copies
had already drifted on the one thing the literal did not pin: the space
between blocks. Three used 12 px, two 8 px, one 16 px, one under the other in
the same column.
`InsightSectionCard` composes `Card`, so the frame comes from the card
contract, and fixes the inner rhythm at 12 px. The daily briefing moves onto
it as well, and the lab-changes list keeps its edge-to-edge rows through the
documented flush variant.
Visible change on the overview column: two bodies open by 4 px, one tightens
by 4 px, and all of them pick up the shadow their card neighbours already
had. Chart bodies are untouched.
The HealthKit metric sub-pages and the vitals dashboard had each hand-rolled
the same failed-read row, byte for byte. The tall centred error card is the
right shape for a list surface but dominates a dense metric strip, so the row
stays — as a named primitive rather than a third copy.
`QueryErrorRow` carries the shipped shape with one correction: the failure
message renders in the foreground tier. An alert is the content of a failed
surface, not meta about it, and both copies had it muted. The retry glyph
matches its tall sibling. Both converted surfaces pass their existing slot
names through, so their hooks are unchanged.
Three integration cards carried the same hand-copied chip literal in the
card-header accessory slot, each free to drift on the next touch; the size
itself was an arbitrary value repeated at three slightly different spellings
across the tree.
- Mint `--text-2xs` (0.6875rem with a matched line-height) as the one
sanctioned step below `text-xs`.
- Mint `TagChip` and adopt it on the Fitbit, Google Health, and Nightscout
cards. It stays distinct from `Badge`, which carries status meaning.
- Sweep the arbitrary micro sizes: the stepper label and the two dose-ledger
outline badges move onto the token; the muted target caption and the muted
ledger meta row move up to the `text-xs` muted floor.
No visible change beyond the two muted rows gaining a fraction of a pixel.
Arm the response waiter before navigating, so a response that lands during
the initial load is still observed instead of timing out on an event that
already happened, and poll for the returning refetch rather than sleeping a
fixed span that a loaded machine can outrun.
Mood has been a first-class module here for a long time, with entries,
structured tags, rated factors and notes. The bridge to the external moodLog
service had nothing left to add, and its Settings card was withdrawn in
a56e6fe61 on the promise of a dedicated removal. This delivers it.
Removed: the hourly `moodlog-sync` queue, cron and handler binding; the
inbound webhook and its shared-secret handling; the reverse push and its four
call sites; the pull-sync module; the per-user credential routes and the
instance-wide toggle; the entry on the `/api/integrations/status` envelope,
the `IntegrationKey` union, the cadence map and the display-name table; the
boot-time secret rotation pass; 22 message keys per locale.
Migration 0271 drops the five `users` credential columns and
`app_settings.mood_log_global`, clears the ledger row for the provider, and
deletes the retired pg-boss schedule so no cron keeps enqueuing work no
handler drains.
Mood data is untouched. `mood_entries` is not named anywhere in the migration
and cannot be reached from it: the dropped columns carry no index, constraint
or foreign key, and `mood_entries.user_id` references `users(id)`, not any of
them. Rows the bridge imported keep `source = 'MOODLOG'`, and every read path
- list filter, source badge, exports, backups - still resolves that value.
`src/lib/validations/moodlog.ts` held the native mood-entry schemas alongside
the bridge's own; it is now `validations/mood.ts` and carries only the former.
The load-bearing one is the panel-coverage test: render the panel with an
envelope carrying every integration key and require each configured one to
surface as a card or as a fallback row. Drop the fallback row and it goes red,
which is the whole point — the previous shape of this bug was a provider on
the wire that no screen rendered, and no card-shaped test could have caught it.
Alongside it: the ladder table with the month-silent fixture that resolves to
stalled, the 24-hour and 7-day boundaries asserted against literals rather than
the constants they check, the per-metric stale flag firing only under a healthy
provider, and a structural guard that the verdict-to-pill mapping is defined in
exactly one module.
The card suites migrate with the semantics: a transient failure now expects the
warning pill everywhere, and the Apple Health suite trades its bespoke badge
states for pill assertions. The e2e mocks gain the verdict shape, the
nightscout entry, and an assertion that the card-less row is on the page.
Three `pillStateFor` implementations disagreed with each other, so the same
ledger state painted red on one card and amber on the next, on one page. All
three are gone; the client projects the server verdict once.
Red now means "your action fixes this" and nothing else. A transient upstream
failure carries the warning tone: telling a user to reconnect against a 503
sends them clicking at something they cannot repair, and the detail belongs in
the inline error line anyway. Three new pill states carry the rest — no recent
data, sync stopped, waiting for first data.
Per-metric freshness has been computed on every status call since v1.27.30 and
rendered nowhere. Every connected card now opens a disclosure listing each data
type the connection has delivered and how long ago, with a quiet one flagged.
Collapsed by default so a 25-type card keeps its shape, but the collapsed
summary already names the count, or the signal hides behind a click. A type a
provider never delivered simply is not listed.
The panel iterates the envelope and renders a plain status row for every
configured entry no card claims. That is the structural half: a connection can
no longer be running, failing and absent from the page at once.
Apple Health drops its bespoke badge for the shared pill, gains the divider and
body inset every sibling carries, and says something when a week has passed
with nothing arriving. Nightscout reads the envelope instead of its own status
round-trip. The OAuth card uses the shared error primitive, all nine cards
carry the divider testid, and the four original connectors gain the "view your
data" link the newer ones already had.
The ledger records what the last attempt did. It has no way to say what is
no longer being attempted: a row frozen at `error_transient` for a month
reads exactly like one retrying every hour, because nothing anywhere read
the age of `lastAttemptAt`.
`sync-verdict.ts` derives the missing reading as a pure read-side
projection: eight verdicts down one ladder, no writer, no cron, nothing that
can desynchronise from the row it reads. `stalled` is the new one, and it
deliberately outranks `failing` — an integration nobody is calling is not
failing, it is not running.
Every entry on `/api/integrations/status` carries the verdict, computed
inline from data the route already holds. That placement is the point: a
card-level verdict misses a card-less provider by construction, which is how
a dead sync stayed invisible. Nightscout joins the envelope so its card can
stop fetching its own status.
The freshness module gains the workout leg (Strava writes no measurements at
all) and a source-parameterised core so the Apple Health route can reuse it.
Its per-metric `stale` flag fires only while the provider itself reads
healthy: that combination is the dead-pipe signature, and when the whole
connection has stopped every metric ages together anyway.
MCP reports the verdict beside `connected`, which only ever meant "not
explicitly disconnected".
Sweep every feet + inches pair the fields can offer and every whole-inch
height across the canonical range, asserting the value comes back
unchanged and lands inside the server's window. Assert the metric branch
with toBe so a re-derivation of a stored height cannot slip through.
Render both entry surfaces under each preference: the metric case keeps
the single centimetres box with the canonical guardrails, the imperial
case shows the feet + inches pair seeded from the stored value. Extend
the structural guard so a surface that drops the adapter, hardcodes a
centimetre guardrail, or hardcodes a unit label trips a tripwire.
Height is a profile column, not a MeasurementType, so the display
transform that carries the metric/imperial preference across every
reading surface never reached it. Add the adapter that closes the gap.
With imperial selected both entry surfaces render a feet + inches pair
instead of a centimetres box; with metric selected the field is what it
was. The adapter converts to canonical centimetres at the entry
boundary, so storage, the wire, the Zod bound, and every consumer of
the value stay unchanged.
Imperial entry quantises to whole inches, which map to centimetres
without a remainder, so save-then-reopen returns the typed pair. The
offered feet window rounds inward against the canonical 50-300 window
so a height the fields allow is never refused on save.
The unit select shares the Settings card with the height field, so a
flip re-expresses the in-flight draft through the new branch rather
than stranding an empty field.
Stable per-segment sleep keys plus the replace-by-window sweep, a one-shot
repair over the past year, Retry-After handling, and the unchanged-value write
skip. Migration 0270 records which connections the repair has been through.
Mirrors the Google sleep suites name for name: the mapper keeps its externalIds
across a re-score that re-classifies and renumbers segments, the sweep stays
bounded to the night's window and never touches the fresh keep-set, a null
window issues no delete at all, and a cleanup failure never fails the sync.
The end-to-end fixture runs one night stored under the old keys through
`syncUserSleep` and pins the ordering the heal depends on — sweep first, then
the rescue re-keying the tombstoned rows in place, no parallel insert — plus a
second pass that writes nothing at all.
Also covers the repair's verdict gates (bounded 365-day window, marker stamped
only on a clean run, watermark untouched, dead token and reauth park returning
without a stamp), the unchanged-value skip including the re-classified segment
it must not swallow, and both Retry-After paths.
Fitbit re-scores a night after the fact, and the per-segment externalId carried
a positional index plus the stage label. Both moved on a re-score: renumbered
segments minted fresh keys, the previous scoring's rows stayed live because
nothing swept them, and the night total counted both copies from then on.
Key the segment on the stable session anchor plus its own start instant
(`<logId>:sleep:<segment-start>`), and add the replace-by-window sweep every
other segment-writing provider already has. The two ship together with the
natural-key rescue that has been sitting in the Fitbit upsert unused: the sweep
tombstones the old-keyed rows first, so the rescue re-keys those very rows in
place instead of inserting a parallel copy or losing the insert to
`skipDuplicates` against the natural unique index.
A one-shot repair queue re-reads the bounded sleep history once per connection
so no user has to do anything; it stamps `sleepRepairedAt` (migration 0270) only
on a clean run, never touches the sync watermark, and reports the rows it
removed. Nights older than the 365-day backfill horizon keep their old keys and
whatever duplicates they already hold — they can no longer gain new ones.
Alongside: `fitbitGet` now honours a `Retry-After` inside a 120 s cap by waiting
once and retrying once, and fails fast above it so a long hold cannot starve the
rest of the poll cohort or outlive a job's expiration. The measurement upsert
skips a write whose payload is unchanged, which takes the hourly overlap off the
delta feed for every Fitbit resource.
A route with no caller is an API detail, not a feature. The Oura, Polar,
Strava and Nightscout cards get the same action WHOOP already had, in the
same row as Test connection, rendered only once connected: post the sync,
show the count or the failure, invalidate the reads that depend on
measurements. No full-history dialog, since these providers have no such
arm. Keys added across the six locales.
These four were the only connected providers with no way to ask for a
sync; they ran on their scheduled pass and nothing else. Each gets a POST
sync route on the hardened Fitbit and Google Health template: session auth,
a 5-per-minute per-user bucket, and a fixed 502 on failure. No body and no
full-sync flag, because none of the four entry points has a full-history
arm to call. Polar goes through the shared legs entry point so both legs
run with the attribution rules built for them.
The Nightscout error arm is the one that matters: a base URL carries the
API token as a query parameter, so the caught message can hold the user's
secret and never reaches the response body. The classified failure is
already on the ledger, recorded with the provider's own redaction.
Connecting also starts the first sync now. The poll handler has accepted a
single-user payload since it was written and never had a producer, so a
fresh connection sat empty for up to an hour waiting for the schedule. The
Oura and Polar callbacks and the Nightscout connect route enqueue one
targeted job each, best effort, with the hourly tick still the safety net.
Strava needs none: its callback already enqueues the history backfill,
which covers the recent window.
The four poll-queue names move into their own module so a request-path
caller can name a queue without dragging the whole worker handler graph
into the request bundle, following the backfill-queue precedent. Creation,
scheduling and handler binding stay in the registrar, and the dead-queue
guard follows the declarations.
The lookback was pinned at seven days, so a sync that stayed broken for
longer than a week came back with a window too narrow to reach the gap.
The quieter case did the same damage with every signal green: a ring left
unsynced holds its nights in Oura's cloud until the app is opened, every
hourly tick succeeds with empty arrays, and when the days finally arrive
they sit outside the window.
Anchor the window on the newest live Oura measurement instead, clamped
between the existing seven-day floor and a new thirty-day ceiling, plus a
day of overlap. Both failure modes leave the same fingerprint, so one
anchor closes both, and the watermark cannot drift out of step with the
data because it is the data. Tombstones are excluded: the sleep sweep
soft-deletes rows a revised hypnogram orphaned, and one of those must not
pass for a live reading.
The ceiling is thirty rather than the ninety days Oura's guidance would
allow, because the whole batch still rides one transaction. With data
flowing normally the clamp floors at seven and the tick is unchanged.
A connection with no rows at all takes the full month.
Cover the adapter end to end: an imperial seed, a guardrail projected
into the display unit, a save inverted back to canonical, and a reopen
that shows the number the user typed, swept across the whole editable
window at the input's own step. Temperature carries the affine case,
where a difference must take the factor without the offset.
Assert the metric path is the exact identity rather than merely close,
so a stored value cannot drift by a rounding pass.
Extend the structural guard to sweep the three target and threshold
surfaces, and add two tripwires: each must call the preference hook and
convert through the shared adapter. Reintroducing a hardcoded unit or
swapping the hook for a literal now fails the guard.
Target ranges stayed canonical while every reading surface moved to the
metric/imperial preference in the previous release, so an imperial
account saw a weight chart in pounds above a target reference in
kilograms. The panel is a coupled display and edit unit: converting only
its display would have corrupted the edit seed and the save round-trip,
which is why it was held back.
Add one adapter that owns every direction of the conversion and route
all three surfaces through it. The reference panel converts the
canonical payload once and hands the converted pair to the range bar,
the status pill, and the editor seed. The editor seeds from the stored
value converted to the display unit, validates against guardrails
projected into that same unit, and inverts the typed number back to
canonical before the PUT.
Guardrails round inward, so the 30 kg floor projects to 66.2 lb rather
than 66.1 lb, which would have inverted below the floor the server
enforces. Absolute values carry the affine offset, differences take the
factor alone. Input step comes from the transform's own precision.
Storage is untouched: thresholds stay canonical SI, and for a metric
account the adapter is the exact identity in both directions, so a
stored value is unchanged bit for bit.
Surface a self-hoster's report that the unit preference was stored but read by
almost no display. Weight, body composition, waist, and temperature now follow
metric or imperial at entry and everywhere they render, storage stays canonical
SI, and the MCP logging tool converts a known unit hint or refuses an unknown
one. OpenAPI diff is version-only; the wire stays canonical.
The crosstab, vitals, and cycle-summary renders are static server renders with
no QueryClient; the new preferred-unit hook reads through the auth query, so
mock it to a real metric-preference implementation (identity values, canonical
symbols) that matches the render those suites assert. Register the new unit
resolver on the MCP write-tools module mock so the preview path resolves.
Add affine-correctness tests for the transform registry (°C→°F absolute vs a
delta that must stay offset-free, lb/in factors, inverse round-trips), a
resolver test for the inbound unit aliases, a preferred-unit entry-form label
test, and MCP cases for converting a lb hint to canonical kg and refusing an
unknown unit. Add the structural guard that trips if any dashboard or insight
surface reintroduces a hardcoded transformed-type unit or value scale. Ship the
diagnose/convert script pair for legacy non-canonical measurement rows.
Route the weight and temperature displays in the mood-factor and cycle-phase
crosstabs and the GLP-1 plateau note through the unit transform: absolute phase
averages take the affine conversion, deltas stay factor-only, and the unit
symbol follows the preference. Thread a unit token into the plateau evidence
copy across the six locales so the sentence reads in the user's unit.
Wire the display transform into the chart scaffold, the chart itself (with an
affine offset for the temperature conversion), the dashboard trend tiles and
weight chart, the vitals grid, the coach read strip, and the device-score
tile. Each surface resolves its unit and scale from the user's preference at a
single choke point and stops hardcoding the canonical unit, so an imperial user
sees a consistent value, axis, band, tooltip, and range hint while a metric
user's render stays byte-identical. Retire the local celsius display shim now
that the transform owns the symbol.
Add the affine display-transform registry (kg↔lb, cm↔in, °C↔°F with the
temperature offset carried as an absolute-only shift), a single client hook
that reads the preference, and a shared inbound unit-alias resolver. Wire the
manual entry form to label, prefill, and convert at the boundary so an imperial
reading persists in canonical SI, and route the record list value cells and
edit dialog through the same choke point. Close the one write path that could
persist a non-canonical unit: the MCP log_measurement tool now resolves a unit
hint to canonical or refuses it, reusing the resolver the CSV importer already
relied on.
A 31st new mirror create at the cap returns 422 with the limit_exceeded
errorCode; an idempotent re-post of an existing mirror at the cap still returns
200; a native create at the cap is unaffected and never runs the count. The
list read exposes externalSource as APPLE_HEALTH for a mirror and null for a
native row.
A client that mints a fresh externalId for the same source concept on every
sync sweep never matches the (userId, externalSource, externalId) idempotency
triple, so each sweep would otherwise create a new mirror row without bound.
Add a growth guard on POST /api/medications: once an account already holds 30
medications mirrored from an external source, a genuinely new mirror create is
refused with 422 and the errorCode medications.mirror.limit_exceeded. The guard
sits after the idempotency lookup, so an idempotent re-post of a medication the
account already mirrors still returns 200, and native creates never run the
count.
Echo externalSource on the medications list read so the web UI and an operator
can tell a mirrored row from a native one, and document both in the OpenAPI
contract.
The diagnose script reports the provenance, external-id values, and intake/schedule
counts of medications by name. The cleanup script deletes only name-matched
Apple-Health mirror rows with zero intakes and zero schedules, dry-run by default,
replicating the delete route (category cleanup, token revoke, audit, cache bust).
Per-collection isolation for the Polar vitals sync: one unavailable data type
no longer blanks the rest, a failing collection is named on the integration
status, and an unexpected response shape is logged instead of reading as no
data. Bump version, regenerate the OpenAPI version string, add the changelog
entry.
The collection fetchers degrade an unexpected 200 body to an empty list, so a
Polar-side envelope rename would read as "the user has no data" and stamp a
clean sync while importing nothing. Emit a fixed-shape polar.envelope.drift
breadcrumb (verb, HTTP status, top-level key names only, never values) at each
silent-empty site in the collection fetcher and the exercises fetcher, keeping
the log-and-degrade behavior so a live account stays alive while the gap becomes
visible in the event stream.
Truth up the client docstrings against the verified AccessLink contract: each
vitals list is a bare array or a named wrapper over a server-fixed 28-day window
with no pagination, and exercises is a bare array over the last 30 days of
uploads. Every hourly tick re-reads the full window, so a late-synced device or
a short outage self-heals with no cursor or watermark; history beyond the window
is an upstream cap. No request parameters added.
Restructure the Polar vitals sync from a single Promise.all over the five
collections into a Promise.allSettled over a per-collection spec, each closure
fetching and mapping so a throwing mapper is contained exactly like a failing
fetch. The healthy collections import; a failing one is recorded once as a
partial failure that names it, without stamping success and without rethrowing,
so a single unavailable data type — SpO2 on a watch without the sensor, a
higher-tier cardio-load read an account lacks — stops blanking recharge, sleep,
and activity on every tick.
A 401 is now a per-collection failure classified reauth_required, not a
whole-leg throw; the write-path escape still surfaces unmarked for the poll
boundary to record once. Mirrors the shipped Oura isolation, minus the
token-refresh leg Polar has no path for.
A failed sync no longer reads as success across WHOOP, Withings, Oura, Polar,
Nightscout, and Strava: a dead WHOOP token stays parked instead of flipping back
to connected, an interrupted history backfill is retried rather than marked
complete, Withings ECG syncs on its own hourly schedule, activity and sleep
write failures surface on the status card, poll-cohort sync errors that
previously vanished are recorded, and Strava history import starts at connect
time.
Add an hourly ECG catch-net cron so a watch-only account with no scale pulls its
recordings without a webhook: the queue was registered and bound but never
scheduled, and the handler now grows a cohort arm that walks every connection on
a payload-less tick, mirroring the activity and sleep handlers.
Make the activity and sleep write paths honest. A swallowed activity batch-write
fell through to the success stamp — a green pill over lost data, green forever on
a persistent write bug; it now records a transient failure and refuses the
stamp. A reconcile throw out of either sleep transaction propagated past the
stamp and was swallowed by both callers, leaving neither success nor failure
recorded; the source now records a transient failure and rethrows so the cron
and webhook paths stay honest without changing their warn-and-continue
behaviour.
Port the hard-fail ledger the Fitbit transport already carries. A revoked WHOOP
grant makes every resource short-circuit and return zero rows without throwing,
so the cycle read as clean: it stamped success, un-parked the reconnect state,
reset the failure streak, and the hourly cohort hammered the dead refresh token
on every tick. Record the three dead-token null-return paths on the ledger and
fold that verdict into both drivers — the full orchestrator and the per-resource
cron path that runs most often — so a dead-token cycle refuses the success
stamp and the park holds.
Return an imported-plus-failed verdict from the orchestrator and gate the
history backfill on it: an interrupted full-history walk now throws for a
pg-boss retry instead of being stamped complete, and a connection parked at
reauth returns without burning the retry budget. Thread the same verdict into
the one-shot sleep-timeline backfill.
Also catch a write-path throw on both drivers so a persistent measurement-write
failure surfaces on the status ledger instead of leaving a stale-green pill, and
widen the cycle overlap to the seven-day re-score window the other collections
use so a late-revised day's strain and energy are re-fetched rather than
stepped past.
Introduce a recorded-marker primitive on the integration ledger so a provider
that records-then-rethrows a sync failure is not recorded a second time at the
poll-cohort boundary, while an unrecorded escape (an Oura or Polar write-path
throw, or any future path) lands on the ledger exactly once, correctly
classified and redacted by the source that owns it. Without the marker a
boundary-blind recorder would double-count the failure streak and could persist
a Nightscout token-bearing URL into the stored error.
Give the poll-cohort factory a required per-provider failure recorder and wire
one for Oura, Polar, Nightscout, and Strava.
Nightscout now counts only freshly inserted readings instead of insert plus
re-confirm (the inflated count fired the reminder trigger every tick), records a
transient failure and skips the success stamp when some rows fail to write, and
emits the measurement-arrival spine for its inserts. Correct the docstring that
promised a far-history backfill that does not exist: the hourly window imports
about 48 hours and nothing deeper.
Strava now enqueues its history backfill at connect time, mirroring the WHOOP
and Fitbit callbacks, so deep history imports within the hour instead of waiting
for the next worker restart.
The GET spread the optimistic-concurrency `updatedAt` into the response even
when the user had never saved preferences, which broke the strict-equality
contract that a never-saved read is exactly `DEFAULT_COACH_PREFS`. Mirror the
insights-layout never-saved GET: a null `coachPrefsJson` returns the pure
default with no token (the first save is the tokenless unconditional write), and
only a saved row and the PUT response carry `updatedAt`. Update the integration
round-trip to expect the token on the saved GET and PUT while keeping the
`coachPrefsJson` column assertion pure.
Add interleaved-writer, tokenless-compat, and malformed-token coverage for the
notification-prefs, modules, coach-prefs, and mood-tag-layout write guards,
including the notification-prefs case where a stale web write must not revert
the app-managed medication flag. Pin the inventory PATCH and DELETE routing
through the advisory lock, and cover the insight-generation privacy re-check on
both the main commit and the reroll arm.
Wire the notification-prefs cards, the module toggles, the coach-preferences
save, and the two mood-tag manage cards to echo the optimistic-concurrency
token they last read and to reload on a 409 instead of resaving over a newer
change. Document the token and the 409 on the four write endpoints in the
OpenAPI registry via the shared base-token field and conflict-response
fragments.
Apply the shared optimistic-concurrency guard (takeBaseToken + guardedUserUpdate)
to the notification-prefs, mood-tag-layout, modules, and coach-prefs write
endpoints so an interleaved edit can no longer silently overwrite a category a
fresher writer just set; a tokenless request keeps the prior unconditional
write. Route the inventory stock-correction PATCH and DELETE through the same
per-medication advisory lock intake consumption takes, so an absolute stock
write can no longer clobber a concurrent dose decrement. Guard every insight
cache commit on the privacy mode still being what the generation read, so a
privacy-mode change is respected by a generation that was already running.
Add per-endpoint unit tests for the insights layout, medications layout, and
coach about-me write guards: an interleaved stale-base writer gets a 409 and
changes nothing, a tokenless request still writes (older-client compatibility),
the fresh token is echoed, and a malformed token 422s. Cover the client token
echo, the two 409 dispositions, and the insights read that must keep the token
through the resolver. Refresh the CHANGELOG and the generated OpenAPI spec.
Extract the dashboard-layout optimistic-concurrency mechanism (issue #581)
into a shared server helper (takeBaseToken plus guardedUserUpdate /
guardedRowUpdate) and a client helper (readUpdatedAtToken, withBaseToken,
isConflict). Retrofit the dashboard-widgets route onto the helper and apply the
same guard to the insights layout, medications layout, and coach about-me write
endpoints.
A request that omits the base token keeps the byte-identical unconditional
write, so older clients are unaffected. Document the contract in the OpenAPI
registry, including the dashboard-widgets path that was previously absent.
Compliance card: a dose-change imperative and a fabricated risk figure are
withheld to the deterministic fallback (no model text persisted, updatedAt
stays null), while a clean grounded summary still renders and persists.
Hero lines: a well-formed, truncated, or fenced JSON envelope resolves to the
deterministic lead/template for both the arrival reaction and the coach nudge;
a plain-text line still ships verbatim.
Route the medication-compliance status card through finalizeStatusSummary so
it enforces the same dose / risk / causal outbound contracts its seven sibling
status cards do, not only the unparseable-envelope withhold it already had. A
dose-change imperative or a fabricated figure is now withheld to the controlled
fallback on this surface too.
Add the status cards' structured-envelope detection to the two background hero
lines (arrival reaction, coach nudge): a brace-led or fenced JSON reply, or a
truncated envelope under the char cap, now resolves to the deterministic lead
rather than rendering raw.
Fold in the residuals from the same sweep: strip a truncated SUGGEST-ACTION
marker from the Coach prose tail, drop an unreachable zero-rate branch in the
adherence fold, and document why the internal conversation summary stays off
the render-boundary screen.
A dose logged from another screen marked the Today hero and the dashboard
stale without re-reading them, so returning to the home page kept showing the
old state until the 120 s poll or a hard reload. The daily reads run with
refetchOnMount off, so a bundle invalidation with the default active
refetch-type never touched them while they were unmounted.
Route every intake-affecting medication write through one blessed
invalidateMedicationReads (bundle invalidation plus a forced inactive refetch
of the digest and snapshot): Take all due, the intake edit dialog, the history
bulk/row delete, the dose-history ledger and its add dialog, the lifecycle and
danger-zone actions, and the intake import. Give the Vorsorge reminder
mutations the same daily-rail refresh, bust the biomarker assessment and the
labs-changes read after a lab write, and refetch the daily reads after a
briefing regenerate so the Today lead updates without a reload.
Restore the dashboard-layout concurrency belt: the settings layout query now
always loads its own tokened copy, so a home-page visit can no longer strand a
save without its base token and silently drop the v1.32.16 conflict guard.
Add a structural guard that fails CI when a new site invalidates a daily-reads
bundle without the paired inactive refetch, so this class cannot recur a
fourth time, plus behavioural coverage for the batch take and the Vorsorge
refresh.
Refs #581
Hold the Fitbit sync verdict when the token is dead or a write fails, and
verdict-gate the history backfill so a partial first-time import retries
instead of marking itself complete. No schema or contract change; the OpenAPI
version string tracks the bump.
New sync-dead-token suite mirrors the Google Health parity test: the three
getValidToken null-return paths each register on the hard-fail ledger, a
successful refresh leaves it empty, and a dead-token cycle returns
{ imported: 0, failed: true } without stamping markSynced or recordSyncSuccess.
Extends the upsert suite with the two write-catch labels (measurements:create,
measurements:update) and the sync orchestrator suite with a write-label
watermark hold. Rewrites the backfill suite to mirror google-health-backfill:
a clean run stamps, a failed verdict throws without stamping, a parked
connection returns without running the sync.
A revoked or expired Fitbit refresh token used to fail the cycle's status
ledger and then stamp the same cycle a success: the connection flipped back
to connected, the watermark advanced, and after a reconnect the 24h overlap
could no longer reach the missed days. getValidToken's three null-return
paths (credentials missing, refresh failure, vanished connection) now register
on the per-cycle hard-fail ledger, so the cycle's verdict fails, the park at
error_reauth holds, and lastSyncedAt does not move. The two measurement write
catches do the same, so fetched-but-unwritten rows are re-fetched next tick
instead of aging out of the overlap.
syncUserFitbit now returns { imported, failed }; the manual-sync route, the
poll cohort adapter, and the backfill consume the new shape. The backfill
stamps backfillCompletedAt only after a clean full-history walk: a parked
connection returns without stamping, and a failed verdict throws so pg-boss
retries on the already-configured retry budget.
Berlin boundary pairs (previously dropped), the gone New York lag-1
offset, all three metric partners moving, UTC-user invariance, and a
metric-by-metric guard proving the weight x BP and BP-medication joins
stay byte-identical across timezones. Update the mocked aggregates for
the additive weightRawRows field and pin the widened raw pull in the
aggregator test.
The mood series has been local-day-keyed since v1.32.12, but the metric
partners it joins against (blood pressure, weight, resting pulse) were
still keyed on the UTC calendar day. The join is an exact day-string
match, so for anyone outside UTC an evening reading landed on the next
UTC day and paired with the wrong mood entry — a one-day misalignment
sold as a same-day correlation.
Derive the three mood pairing partners as exact local-day means, keyed in
the user's own timezone via userDayKey (Intl-based, DST-safe). Blood
pressure and resting pulse reuse raw rows already on the request path;
weight rides one additive column on the existing raw-rows query. The
UTC-consistent weight x BP and BP-medication joins are left untouched —
both their sides share one UTC key space and they migrate with the
rollup tier later.
Server: a write with the current base token succeeds and returns the advanced
token; a stale base returns 409 and writes nothing (no unconditional fallback);
a malformed base 422s; an omitted base keeps the unconditional write. Client:
the ring payload never carries widgets and threads the base token, the shared
busy gate covers either write, and the Save/ring wiring is source-pinned.
Refs #581
A dashboard layout change could be silently reverted when a hero score-ring
toggle was made around the same moment. Both writes PUT the whole layout to
/api/dashboard/widgets, and whichever reached the server last won, so a ring
write built from a pre-save snapshot overwrote a fresher Save.
Guard the write with optimistic concurrency. The client sends the updatedAt it
based its edit on; the route writes conditionally on the stored row still
carrying that token (updateMany where updatedAt = base) and returns 409 with a
typed body on a stale base, writing nothing. The advanced token rides back on
every GET/PUT response so the client can rebase. No migration: the User row
already has an @updatedAt column.
On the client, the ring PUT carries only its own fields plus the base token,
the Save and ring writes share one mutation scope so they serialise, the Save
button and ring controls disable while either write is in flight, and a 409
refetches the winning layout and asks the user to redo the change.
Refs #581
Parity suite proves a Polar exercise and an equivalent Strava activity map to a
downstream-indistinguishable Workout row (sport bucket, start instant, duration,
distance, calories, HR, null-vs-zero). Regression guard asserts the exercises
request hits GET /v3/exercises with no userId path segment. Dedup ladder tests
mirror the Strava suite (overwrite-in-place, Apple+Polar twin collapses, Polar
above Strava). Composite test proves a failure in one Polar leg leaves the
other leg's result intact on the shared ledger. Defensive-shape tests cover the
four flagged assumptions: numeric-vs-string id, UTC-offset composition (positive
and negative), bare-array-vs-paged body, and a 403 scope error as a soft error.
Refs #568
Pull the athlete's exercises from the stateless GET /v3/exercises list (last
30 days of uploads, no transaction, no userId in the path) and upsert each into
the Workout table as source POLAR, keyed on (userId, source, externalId) so a
re-fetch or Polar-side re-analysis overwrites in place. A new table-driven
sport map normalises the raw Polar sport labels into the canonical
WorkoutSportType bucket at ingest, mirroring the Strava/WHOOP mappers; the raw
labels and training load ride metadata for provenance.
The exercise leg rides the existing hourly polar-sync cohort tick alongside the
vitals leg, with independent attribution on the shared polar ledger: each leg
is attempted and records its own failure, and neither can mask the other's
result. Polar takes its place in the workout source ladder between Withings and
Strava. No migration and no OpenAPI change: MeasurementSource.POLAR and the
full workouts source enum already exist.
Refs #568
Replace the internal "[unverified]" token the grounding guard left in a Coach
reply with the editorial elision mark […], and add a quiet muted caption
under the bubble explaining that some figures could not be checked against the
user's data. The mark holds the slot grammar-safely mid-sentence, before a
surviving unit, and inside a range; the notice carries the meaning. The withheld
count rides the existing provenance envelope (unverifiedFigures) through the
stream, persistence, and reload, and the fenced document-chat path carries a
minimal envelope so its notice survives too. New i18n key across all six locales.
Also fix an adjacent reload gap: provenanceFromJson restored only a subset of the
serialised envelope, so a persisted suggestion card, confirm action card, and the
tool trace were dropped on conversation reload despite the render layer expecting
them to survive. Restore them defensively alongside the new field.
A truncated or malformed { "summary": "..." } envelope fell through the
status-card parser and rendered raw — braces, the summary property name,
quotes, and literal \n — in the Assessment card (resting pulse and the
other per-metric cards). parseSummaryFromContent returned the raw content
on any JSON parse failure, and a truncated envelope with no closing brace
also fell into that path, so partial structured output reached the reader.
Split the read into extractAssessmentSummary, which classifies a completion
as a parsed summary, genuine prose, or an unparseable structured envelope.
A payload that opened like a JSON object but failed to parse (or parsed
without a usable summary) resolves to the controlled fallback rather than
raw text; plain prose that merely contains a brace is still preserved.
finalizeStatusSummary withholds an unparseable envelope so every status
card serves its deterministic stub, and the two direct callers
(medication compliance, derived-score assessment) get the same guard.
Refs #608
The persistent mood-rollup writer bucketed on the UTC truncation of
`mood_logged_at` and ignored the canonical per-row `MoodEntry.date`
label. Every coverage-miss fallback keys on `entry.date`, so the rollup
tier disagreed with the fallback and with the day the user logged the
mood: a late-evening or after-midnight local mood landed on the wrong
calendar day, and the label shifted the moment the boot backfill flipped
a user from the live fallback onto the rollup tier.
Key every bucket off the stored `date` label instead — DAY, WEEK, MONTH
and YEAR — with `bucket_start` set to that label's UTC-midnight instant.
The rollup is now byte-identical to the live `entry.date` fallback and
timezone-independent by construction: DST correctness stays where it
already lives, in `moodDateKey` at write time. No consumer changes, no
schema change; every rollup reader already formats `bucket_start` as a
UTC day label.
Migration 0269 empties `mood_entry_rollups` so the boot-time
full-backfill worker re-mints every user's buckets under the corrected
keying; the worker now deletes-then-refolds so it is self-cleaning
against any row left by an older instance. While the table is empty,
reads fall back to the live `date`-keyed walk, which was already
correct, so there is no window where a wrong day is shown.
Self-hosters can sign in to the iOS app through their own domain, so the
password manager files credentials under the right site and passkeys run in
the correct origin. The app opens the instance's own web login inside an
ASWebAuthenticationSession; after an interactive login the server mints a
single-use, PKCE-locked handoff code on a compiled-in custom scheme, which the
app exchanges for the standard native token bundle. The token pair never rides
a URL — only the opaque code does.
Three endpoints mirror the OIDC native handoff and share its mint/consume core,
now extracted to native-handoff.ts (the OIDC leg keeps working unchanged, as
the default flow):
- GET /api/auth/native/login validates the PKCE challenge, sets a
short-lived encrypted state cookie, and
redirects to /auth/login?flow=native.
- GET /api/auth/native/complete validates the web session, enforces the
freshness binding, mints the code, and
redirects to healthlog://login-callback.
- POST /api/auth/native/token exchanges the code under the cookie-less
native transport gate for the native bundle.
The freshness binding refuses any session older than the flow start, so a
pre-existing browser session can never silently complete a handoff. The start
time is stamped from the database clock (SELECT now()) to stay comparable to
Session.createdAt with no app/DB skew, and createdAt is read through a
dedicated projection so the comparison cannot be silently dropped. A flow
discriminator on the handoff row (migration 0268) makes an OIDC code
non-redeemable at the web-login route and vice versa.
The callback scheme is a compile-time constant with no redirect_uri parameter,
requireAdmin and requireFreshMfa stay cookie-only, and the exchange mints no
wider scope than password login. The state cookie is AES-256-GCM with a fresh
IV. Adds the flow=native login-page banner in all six locales and the three
routes to the OpenAPI contract.
The Coach output guard now tracks the numbers it was shown across the whole
conversation, so a figure carried forward or grounded in a goal, a schedule,
or a reference band stops being marked unverified, while an invented one is
still caught and a fabricated risk number now blocks in six languages.
Promote the per-turn magnitude bag to a per-conversation typed Grounding
Ledger built from every source the model was actually shown: this turn's
tool payloads, prior-turn tool figures, coach-memory goals, the medication
schedule, the reference-grounding block, and guided blocks. Assistant prose
is never a registration source, so a figure that slipped a rung cannot
launder into an authoritative value on the next turn.
Complete the action ladder: a reconciled number passes, a reference or
education sentence is exempt across all six locales, a positively-typed
benign number (date, time, year, ordinal, list marker) passes while an
untyped unreconciled one stays on the strip path, and an unreconciled
health figure is soft-stripped.
Gate the dose continuation exemption on the user's actual schedule: a
"keep taking your N mg" is trusted only when N is a dose the user is on;
an off-schedule maintenance dose is caught. Extend the risk and percent
classifiers to fr/es/it/pl to match the EN/DE depth.
No schema migration: the ledger is rebuilt per turn from persisted tool
traces and re-rendered prompt blocks.
Bumps the version, regenerates the OpenAPI contract and the service-worker
version fallback, and records the release notes for the intake-source and
sync-trigger provenance surface plus the shared hhmm util and migration-number
guard.
Two short-lived branches can each grab the next free migration number and
both merge, leaving the tree with two directories sharing a four-digit
prefix. This structural test fails the build the moment a prefix repeats, so
the collision surfaces at PR CI instead of after merge. One historical
collision (0025) predates the guard and is grandfathered — a migration
directory name is the primary key in the applied-migrations ledger, so it
cannot be renamed without orphaning the record on every deployed database.
Seven near-identical copies of the HH:mm to minutes-since-midnight parser
had accumulated across the scheduling surfaces in two flavours: a strict one
that trusts an already-validated literal and a null-returning one for
unvalidated input. Both now live in src/lib/medications/scheduling/hhmm.ts as
hhmmToMinutes and hhmmToMinutesOrNull, and every call site imports from there.
Behaviour is identical: the per-medication route keeps its malformed-to-zero
fallback as a thin adapter over the shared core, and dose-window re-exports
the strict variant so its consumers and unit suite keep their import site.
Route-level tests that run the real apiHandler + requireAuth so the
cookie-vs-Bearer derivation is exercised end to end, not papered over by a
unit mock:
- a cookie POST records WEB, a Bearer POST records API, and a body that
forges `source` is ignored;
- the dose-history ledger selects and returns `source` per row, null on a
legacy row;
- a batch carrying `syncTrigger:"background"` rides the ingest annotation,
a batch without it annotates null, an invalid value is 422, and the field
leaves the per-row insert/update outcomes byte-identical.
Root-cause the intake source attribution and give the HealthKit batch a
diagnostic sync-trigger, one attribution surface for both.
AuthContext now carries a read-only authMethod ("cookie" | "bearer"),
derived from the same branch that sets the wide event's auth_method. The
medication-intake route maps it to the stored IntakeSource: a Bearer
(native) call records API, a cookie (browser) call records WEB. The other
producer values are untouched — a converge onto a pending reminder row
updates through applyCanonicalSlotWrite, which never rewrites source, so
REMINDER / IMPORT / APPLE_HEALTH provenance is preserved. The value is
never client-asserted, the same posture as userId being narrowed from auth.
The dose-history ledger now selects and returns source per intake row so a
client can label how each dose was recorded. The endpoint gains its first
OpenAPI entry (path plus response schema) so the field lands in the
iOS-consumed contract.
The measurements batch accepts an optional per-request syncTrigger
(foreground | background | push). It is diagnostic-only: recorded on the
measurement.batch.ingest wide event and nothing else — never persisted,
never part of the dedup key, never an input to attribution or how a sample
is stored. Sending it or omitting it produces byte-identical row outcomes.
Regenerated docs/api/openapi.yaml.
Replace the untyped magnitude bag in the Coach numeric verifier with a typed
tokenizer and a normal-form reconciler, and narrow the outbound risk and dose
banks so correct figures and correct refusals stop being blocked while every
fabrication shape still blocks.
Grounding verifier (coach-prose-grounding.ts):
- Classify each prose number before grading — ISO/written dates, clock times,
bare years, ordinals, list markers, hyphenated time spans, ranges, percents,
thousands-separated integers, unit-suffixed measurements. Benign types are
never graded or stripped; ranges decompose into per-endpoint magnitudes.
- Grade against typed authoritative entries (value + kind + aggregation).
Sign-insensitive, canonical rounding, ratio to percent in percent space, and
minutes to hours widen the match only for an aggregate, kinded entry; kind
scoping stops a weight value grounding a blood-pressure claim.
- Thousands/decimal separators follow the reply language.
- Boundary-safe strip so a flagged token never clips a larger grounded number.
Outbound screen (outbound-screen.ts):
- Risk bank requires an asserted figure or a categorical engine result; a bare
engine or horizon mention (education/refusal) passes. Spelled-out percents
count as figures. No refusal-context exemption. EN + DE.
- Dose bank keeps a maintenance-continuation restatement, voided by any change
stem. Taper suffix fixed.
Route + fenced-chat thread the reply locale into the verifier.
Merge the eight reviewed audit fixes that were cut a week ago and never
landed. Main has moved on since, so each fix was re-checked against the
current tree and each of the five conflicts resolved by understanding
both sides.
Kept, still open on main:
- Proxy early exits now carry the transport security header subset via a
shared helper. Reconciled with main's /api/health worker-only carve-out.
- Batch measurement ingest bounds the entry instant so a future-dated row
can no longer win "latest reading" or poison the rollup buckets.
- Web Push dials through the pinned safeFetch dispatcher with
requirePublicHost, closing the last user-controlled-host egress.
- Mood bulk upsert and its tag/factor links run in one transaction, so a
rejected rating rolls the mood row back instead of leaving a tagless row.
- Full backup carries NutrientIntakeDay with its unit and source. Folded
into main's portable-export / disaster-recovery split.
- Live daily-series fallback restores the min/max spread band the tiered
read already emitted.
- Coach stream aborts its in-flight reader on unmount and resets send state
on a conversation switch.
- Manual water writes share the day-bound helper, nutrient add clamps after
the increment, and the thiamin cap drops to a real supplement ceiling.
- Compliance point window widens symmetrically instead of wrapping overnight.
- Oura, Polar and Fitbit token helpers classify an empty or non-JSON 2xx
body as a handled transient error.
- WebAuthn relying-party fallback to localhost is gated to development.
- Chart-token strip list is derived from the allowlist and pinned by a test.
- Regression tests pin the snapshot-cache tenant key, the GlitchTip
path-secret redaction and the Withings credentials delete.
Superseded by main, tests re-pointed at the current shape:
- The intake CSV import was rewritten on main to run through a background
job with a single-active-job admission gate. The future-date bound is
re-applied at request validation and the per-user limiter is kept for
parity with the bulk twin; the old synchronous-write test is replaced.
Integration coverage for `validateSessionFromCookieValue` against real
Postgres: a live secret-backed session resolves to its user, an expired one
and an unknown/forged/empty value return null, and a legacy cuid cookie
resolves only while its row still carries no hashed secret. This is the
hashing + row lookup + expiry the callback's duplicate detection turns on,
which the unit suite stubs out.
Refs #600
A browser SSO sign-in landed on /auth/login?error=oidc_failed even though it
had succeeded. Two layers conspired:
- The service worker enables navigation preload, but its fetch handler returned
early for /api/ paths without calling event.respondWith(). For the OIDC
login/callback navigations that left the preload unconsumed, so the browser
also ran its own default fetch and the callback was requested twice. The
first request redeemed the single-use authorization code and set the session;
the second tried to redeem the now-consumed code, failed at the IdP token
endpoint, and that 307 to the error page is what the user saw.
- Even with the double request gone, a legitimate retry (proxy, double
navigation) hit the same wall on the server.
Fix both layers:
- The worker settles /api/auth/* navigations itself, consuming the preload
response (or fetching once when there is none), so exactly one request
reaches the server. These responses are never cached.
- The callback is idempotent. When the single-use state cookie is already
consumed but the request carries a database-validated, unexpired session, the
callback treats it as a duplicate of a completed login and redirects into the
app instead of surfacing an error. The authorization code is never exchanged
on this branch, and a consumed/forged/expired state from a caller with no
valid session still fails closed with the unchanged oidc_state error.
Refs #600
The previous commit spliced the DATA INVENTORY manifest into the tool-mode
authoritative payload set unconditionally. That made the payload set
non-empty on a turn where the model answered WITHOUT calling any tool
(empty tool results), which flipped the verifier from dormant to active —
grading the reply against an inventory that carries per-domain sample counts
only, never the snapshot's pre-computed averages. A figure the model
correctly narrated from the snapshot (a 30-day systolic average) was then
stripped to "[unverified]" — the exact false positive this work exists to
prevent, and a regression from main's "no tool figures this turn → skip,
prompt-level grounding is the backstop" contract.
The inventory must only WIDEN an already-active authoritative set, never
ACTIVATE it on its own. Gate activation on a real present figure block
(a pinned workout-evidence block or a present tool result) exactly as main
did; when active, the inventory counts still join the set so a plain count
restatement stays grounded.
Adds a route-level regression test (fast suite) proving the verifier stays
dormant when no tool returned figures, plus companion unit tests documenting
that a real tool section grounds the average while a counts-only inventory
does not.
Related to #585. Investigated the intraday pulse chart's read path
end to end: the ZIP-import raw-sample fold, the live-sync uploaded
10-min bucket fold, their merge, and the chart's render logic. All of
it treats the two ingestion shapes consistently — the chart has no
shape awareness at all, it paints whatever bucket series it receives.
No read-side or display-side defect explains a systematic density
difference between the two ingestion paths.
What was missing is visibility into the write and read sides that
would let a future report actually be diagnosed instead of
re-investigated from scratch. The batch endpoint's wide-event trace
never broke a skip down by reason, and never distinguished an
HR-bucket write from any other entry in the same request. The
intraday route's own trace surfaced only the final bucket count, with
no visibility into which shape (raw sample rows vs uploaded bucket
rows vs folded hourly rows) a day's data actually came from. Both
traces now carry that breakdown, so the next report of "the chart
looks sparser after a given date" is answerable by grep instead of by
re-deriving the whole read path again.
Three confirmed false-positive shapes in findUnverifiedCoachNumbers/
stripUnverifiedNumbers were replacing legitimate prose with the literal
"[unverified]" placeholder even though the underlying data was grounded:
- A correctly-computed delta, midpoint average, or percent change between
two of a payload's own headline aggregate figures (e.g. "down 3.5 kg from
your 7-day average") was never itself a raw leaf in any tool result, so it
was flagged as invented. The authoritative set now also carries the
pairwise delta/average/percent-change between a SINGLE payload's own
named aggregate fields (latest, avg7, avg30, avgSys30, allTimeAvg, ...),
scoped per payload so a weight delta cannot ground itself against an
unrelated metric's figure.
- The always-sent DATA INVENTORY manifest (tool-mode base context) carries a
sample count per domain, but that count was never added to the
authoritative set the way the pinned workout-evidence block already is —
so a plain restatement of a domain's count ("you've logged 42 BP
readings") was flagged whenever the model didn't also call that domain's
retrieval tool in the same turn. The inventory entries now splice into
toolResultPayloads exactly like workoutEvidence.
- A calendar-date ordinal ("July 21st") or a numbered-list marker ("1. Cut
back on sodium") tokenised as a bare number and, when ungrounded, got
soft-stripped mid-word into visibly broken text ("July [unverified]st").
Prose is now masked for these two narrow, high-precision shapes before
numeric extraction; the original text used for stripping is untouched.
A genuinely fabricated figure (unrelated to any grounded point, inventory
count, date, or list marker) is still caught and stripped — covered by new
adversarial cases alongside each fix.
Investigated and ruled out: multi-round tool-loop result accumulation (all
present results across every round already reach the verifier); unit
conversion (the app resolves and serves one consistent unit, so there is no
independent conversion step for the model to get wrong). The Daily
Briefing's separate withhold-a-whole-section grounding gate is unrelated to
the reported symptom (a different failure mode, whole-line omission vs.
inline replacement) and was not touched.
Closes#584. `/insights/pulse` computed hasRestingHr from
RESTING_HEART_RATE presence and used it to swap the ENTIRE primary
chart and the Coach read strip to that series whenever any resting
row existed, while the title, stat strip, capture action, and
"show all values" link kept saying PULSE. A user could see a
resting-only chart under a page that claimed to be pulse throughout.
The chart now always includes PULSE as its first series; when resting
data exists, RESTING_HEART_RATE renders as a second, clearly-labelled
series (the same multi-series pattern the blood-pressure page uses for
systolic/diastolic) instead of replacing it. The Coach read strip
stays pinned to PULSE. The page also drops the resting-calibrated
target band it used to paint behind the chart — that band is
calibrated to steady-state resting readings and would misreport an
expected workout spike in raw PULSE data as "out of target"; the
personal target belongs on the dedicated resting-pulse page against
the clean resting series.
PUT /api/settings/ntfy reconstructed the entire stored config from the
request body. GET never returns the write-only authToken (only
hasAuthToken), so clients correctly omit it on any edit that doesn't
touch the credential — but the route treated an omitted or empty token
as "clear it", silently destroying a working notification credential on
routine edits like toggling enabled.
An omitted or empty authToken now preserves the currently stored value;
only a non-empty value replaces it. Mirrors the identical headerValue
preserve-on-empty contract already shipped on the webhook channel route.
PATCH /api/measurement-reminders/{id} persisted an explicit null cadence
clear correctly, but the immediate nextDueAt recomputation merged fields
with `updateValue ?? existingValue` — a JSON null and an omitted field
are both nullish, so the merge silently fell back to the OLD cadence for
one cycle. This hit interval-to-RRULE and RRULE-to-interval transitions
and an explicit anchorDate clear alike.
The merge now keys off Object.hasOwn(updateData, field): updateData
already encodes exactly which fields are being persisted, including the
mutual-exclusivity auto-clear (setting one cadence nulls the other), so
recomputation can never diverge from what the row ends up holding.
Closes#588. The ZIP extractor decompressed export.xml synchronously
into one in-memory buffer, then wrote it with a blocking writeFileSync
— on the same event loop the web server runs in the default
single-container deployment. A real multi-year export could pin the
Node main thread for minutes, freezing every other request, while
briefly holding roughly 2x the decompressed size in heap. A process
that died mid-extraction never reached the failure path, so the
ImportJob row stayed stuck in "unpacking" with "0 rows imported"
forever and no error ever surfaced.
Extraction now streams through zlib's async inflate into a file write
stream, moving the decompression work off the event loop and bounding
memory to the chunk size. The zip-bomb ceiling moves from
inflateRawSync's maxOutputLength to an equivalent byte-counting
transform on the streamed output.
Orphan-ImportJob reconciliation used to run once, at worker boot.
docker-compose's restart:unless-stopped does bring a crashed worker
back, but a fast restart can land before the stuck row's heartbeat
goes stale or pg-boss stops reporting the job as live — and because
the boot-time pass never re-runs on a healthy worker, that row was
never revisited again. A 15-minute cron now re-runs the same reconcile
so a genuinely abandoned import converges to a visible failure instead
of staying silently stuck.
The Settings page's instant score-ring toggle PUT the full dashboard
layout, built from the client's cached `remote` snapshot. That snapshot
can go stale mid-flight: if a normal tile/chart Save committed a newer
layout first, the ring request's explicit stale `widgets` (and
`comparisonBaseline` / `chartOverlayPrefs`) still won on write and
silently reverted the just-saved layout whenever it resolved last.
The ring mutation now PUTs only the two ring fields
(`buildRingMutationPayload`). `widgets` joins the existing
preserve-when-absent merge (`LAYOUT_FIELD_MERGE_DISPOSITION`) so an
omitted field is carried forward from whatever is currently stored
instead of a client-held copy, closing the race regardless of request
order.
Renumber the elevation migration from 0260 to 0267 — main claimed 0260
for an unrelated table while this branch sat unmerged, and 0266 is now
the highest committed number.
Two loose ends the merge itself did not touch: the schema comment on
StepUpElevation.method still called it "audit/telemetry only, never an
authorisation input" — the exact claim the fresh-factor fix (a prior
commit on this branch) made false, since the method is now precisely
what decides whether an elevation reaches the destructive MFA routes.
Reworded to match. And pruneExpiredStepUpElevations in step-up.ts was
dead code: the actual sweep runs in cleanup-handlers.ts against the
worker Prisma client for reasons that function could never satisfy
against the request-scoped one. Removed rather than left to drift.
The named-risk-engine pattern in the outbound safety screen used
score2? (the 2 optional), so it matched bare "score" too — any mention
of one of this app's own computed, documented scores (Sleep Score,
Readiness Score, Health Score, Recovery Score, Stress Score, Strain
Score) tripped the bank meant only to catch a fabricated named
clinical-risk calculator (SCORE2, Framingham, ASCVD, QRISK).
Require the literal "SCORE2" instead. The other risk patterns —
percentage-shaped risk claims and the other three named engines — are
untouched, so a genuinely fabricated risk figure or a named
risk-calculator claim still blocks.
Closes#587.
The empty-path guard in the score fast-path checked raw record presence
(any row at all) rather than each pillar's own eligibility threshold —
weight needs 2 readings, mood needs 5 entries, BP needs a paired
reading, compliance needs an active scheduled medication. An account
with some data but no eligible pillar (a single mood entry, a single
weight reading) fell through to computeHealthScore, which summed no
components and returned a raw 0/red score indistinguishable from a
genuinely poor result.
Check the computed result's own components instead of re-deriving the
thresholds, so the guard can never drift from computeHealthScore's
actual eligibility rules. A pillar legitimately computed as 0 keeps its
0 value and is never swept into the null branch.
Closes#582.
The harness landed in v1.30.35, after the slices that built this
release were already cut — so the reaction line's prompt module was
never registered, and its own docblock's warning against the
two-locale collapse was untested. Registering it surfaced a real gap:
the meaning-first instruction lived only in the system prompt's own
words, never in the user prompt in the family's shared phrasing, which
is where the rest of the surfaces carry it and where the tone contract
checks for it. The German body carried no such instruction at all.
Threads locale through the user-prompt builder so the fixed
instruction line matches the reader's own reply language, and wires
the per-user per-day opener-archetype rotation the milestone spec
called for and the first implementation had not connected.
Nothing warms this queue and no read path triggers it, so an unregistered name
fails open in the worst way: every send resolves, every job vanishes, and the
symptom is indistinguishable from the surface's honest empty state.
The paragraph is model output. Rendering it as markup would be an XSS surface,
and this project ships no markdown library so the question stays closed — the
test is what keeps that from being treated as a formatting decision.
The disable-route suite only ever supplied a cookie-shaped context, which is why
the token-id-as-session-id defect was invisible to it. It now covers both arms
and asserts the credential shape handed to the session revoker, plus that the
elevation is spent before the teardown and not spent when the factor code is
wrong.
The one-paragraph-per-workout claim is a unique index, the cascade is an FK,
and the daily-cap window is a SQL fact — a mocked upsert reports success for
all three regardless.
The push endpoint host is user-controlled — the browser supplies it at
subscribe time. `isPublicUrl` guards it at subscribe time and again before each
dial, but it is a literal input-time check and cannot see a DNS rebind between
the accept and the connect; with a short-TTL record the dial reaches the pod's
private network.
`web-push`'s `sendNotification` dials with its own internal `https.request`,
which bypassed `safeFetch` and therefore the connect-time pin entirely. Both
senders now sign and encrypt through `generateRequestDetails` — identical VAPID
handling and payload encryption — and dial through `safeFetch` with
`requirePublicHost`. Non-2xx is re-shaped as the `{ statusCode }` rejection the
existing classification already understands, so the 410/404 expiry reap and the
transient/permanent split are unchanged.
This is the last user-controlled-host egress that was not behind the pinned
dispatcher (issue #217).
The contract now names which factors satisfy the fresh-factor routes and which
do not, carries `satisfiesFreshFactor` on the mint response so a client can pick
the ceremony before spending a proof, and describes the options endpoint's two
ceremony kinds. The second-factor status read is documented as plain
authentication — it carries no credential material, so it needs no elevation.
The projection's central claim was carried by a comment. It is now walked
before it is returned: any non-numeric leaf at any depth throws, and the only
two string leaves are allowlisted by shape — the sport type against the schema
enum itself, and the day key against its format.
This matters because sport_type is a TEXT column whose API write path is
enum-constrained but whose backup-restore path accepts any non-empty string, so
a restored row can carry arbitrary text. It is re-asserted at the projection
boundary rather than trusted from the column. The heart-rate provenance string
is dropped rather than allowlisted, and route geometry stays consumed-not-carried
— it yields one number, and location history may not ride a prompt.
Own-history rows now collapse cross-source twins through the canonical picker,
as the detail page's comparison already did; a session recorded by two paired
watches was inflating the sample size and biasing every median.
The tiered rollup read emits minValue/maxValue; readLiveDaily omitted them, so
on a rollup-cold account or a coverage miss the chart silently dropped the
min/max spread band for the same underlying data — contradicting the file's own
byte-for-byte parity claim. The live aggregate now selects MIN/MAX alongside the
average.
useSendCoachMessage never aborted its in-flight reader on unmount. The drawer
works around that through handleOpenChange, but the full-page /coach surface
passes no registerReset, so navigating away mid-stream left the SSE loop alive:
the provider kept generating and billing, and trailing setState fired on an
unmounted component.
Switching conversations reset every piece of local state except the send state,
so the previous thread's last assistant or error bubble rendered appended to the
newly selected thread until the next send.
The two lifecycle invariants are pinned structurally, not behaviourally — the
repo ships no renderHook. That is weaker, and the header says so; both defects
were a missing call, which is what a structural guard does catch.
Four suites, each mutation-checked against a reverted protection:
- route-workout-scope: the workout row is read exactly once across a ten-turn
conversation that keeps sending the id, the where clause always names
userId, another user's id resolves to nothing and the chat proceeds, and
the section sits inside the health-data fence.
- workout-evidence: the numbers-only closure. An instruction-shaped
sport_type — reachable through the restore path, which types the column as
free text — folds to "other" and never reaches the block; a free-text leaf
at any depth throws.
- page-launch-params: an unknown ?scope= is ignored silently, and ?workout=
yields to both ?c= and the fenced ?doc= path.
- composite-score-anatomy-coach: every score sheet carries an opener, a
scope narrowed to the composite's inputs, and auto-send.
resolveLaunchState's shape assertion gains the workoutId field.
`createTagLinks` validates `ratedFactors` after the mood row is written. On the
bulk path the two ran as separate awaits on the base client, so an out-of-scale
rating threw once the row had already committed: the per-entry catch reported
"skipped" — the one status that promises nothing was saved — while a mood entry
with no tags or factors sat on disk. The single-entry POST has always wrapped
the pair in one transaction; the bulk twin now matches.
The test client for the transaction is a distinct object from the base client
on purpose. Sharing one identity made the atomicity assertion vacuous: a write
that escaped the transaction looked exactly like one that stayed inside it.
Three properties, each with a mutation-verified guard:
The chip and the line run on DIFFERENT clocks and must keep doing so. The chip
expires after three hours; the line stands for the rest of the local day.
Collapsing them onto one clock reads like a simplification and silently deletes
the day's read at lunchtime.
Degradation is the headline. For every reason a line can fail to exist — no
provider, no consent, an exhausted budget, a dead provider, an unusable output
— the row stays line-less, the reservation is reconciled, and the chip and the
provisional-to-final flip still work. The degraded digest is built through the
real composer rather than a hand-shaped fixture, so the assertions cannot agree
with themselves.
The server pass carries no wall-clock time. The hero paints on the SSR pass
because it is the LCP element, so a locale-formatted time inside it would differ
from the browser's — React #418. The tests render through the server snapshot,
which is exactly the pass that must stay clock-free.
The seeded end-to-end spec lands a sleep batch through the real ingest route,
writes the arrival marker, and asserts the chip and the phase flip in place on
a refetch trigger the digest hook already declares — no reload, no new cadence.
One test per gate, each asserting the refusal reason AND that the provider
entry was never reached — a gate that refuses but still resolves a provider
would pass a reason-only assertion while costing what the gate exists to
prevent. The cap and the hash are asserted independently of each other, since
the singleton queue key that normally collapses a burst is best-effort.
The backfill fixture walks the real chain: 750 historical sessions plus 40
re-synced ones plus one fresh, through the live classifier and the live spine
worker, yielding exactly one generation.
The peak detector changed under test: on a flat ride the ninetieth percentile
sits a beat above the mean, so a bpm of sampling noise read as two efforts the
rider did not do. Peaks now need an absolute prominence over the session mean
as well.
The spine dispatches a generator-free enqueue; the worker resolves the
provider. Five gates stand in front of it in ascending cost order — both
modules on, ten-minute duration floor, a hard count of four a day in the
user's own timezone, an input hash over the deterministic evidence, and the
token ledger inside the shared status chokepoint. Each refusal returns a
status rather than throwing, because retrying against a ceiling that does not
move until midnight is an unbounded loop.
The detail route only reads: a workout with no row serves null and renders no
card, which is every workout that predates this and every re-synced one. An
undecryptable paragraph degrades to no card rather than failing the page.
Today gains two additive-nullable digest fields. `justIn` names the day's
newest arrival while it is still news — under three hours old, after which the
record is simply current rather than new. `reactionLine` carries a single
generated sentence about that arrival and replaces the hero's standing lead for
the rest of the local day, never appending a second paragraph, so the hero stays
one lead line tall and nothing below it shifts.
Nothing here polls. The open dashboard picks the change up on the cadence it
already runs — a state change, not a notification.
The generated sentence is bounded before it is written: the arrival spine's
unique row per kind per local day IS the throttle, the worker re-checks the
same claim before spending, and the reservation is reconciled on every exit
including each failure path. Without a provider, without consent, or against an
exhausted budget the row stays line-less and the surface falls back to the
deterministic lead — the chip and the provisional-to-final flip are unaffected.
The chip's time string is formatted client-side after mount; `justIn.at`
crosses the wire as an ISO instant so the server never renders a local
wall-clock time the browser would disagree with.
Status cards and derived-score sheets read one way: they state a verdict and
end there. The metric pages already carried the hand-off; the score sheets
passed neither an opener nor a scope, so they painted no action at all.
- Score sheets (sleep / readiness / recovery / stress / strain) gain the same
auto-sent opener the metric pages use, each scoped to the inputs that drive
the composite rather than to a block it has no equivalent of.
- Workout detail's coach launch carries the workout itself. The chat request
gains an optional workoutId; on the FIRST turn the route narrows by
{ id, userId } and pins ONE bounded evidence section onto the snapshot.
Later turns reuse the pinned snapshot, so per-turn work stays constant in
conversation length.
- /coach?workout=<id> covers the full-page and maximize paths.
The workout block deliberately does not take the fenced transport the stored-
document path takes: a workout row is the user's own sensor record, and the
projection admits no free text. That is enforced, not assumed — sport_type is
a TEXT column, so it is re-asserted against the closed sport union, metadata
is excluded apart from its numeric zone durations read through a Zod slice,
and a recursive guard throws on any non-numeric leaf.
Own-history sport averages move to src/lib/workouts/sport-context.ts so the
narrative and the detail page cannot quote different figures.
The evidence block is a numbers-only server projection: zone seconds, the
workout-window heart-rate shape, route climb and a same-sport 90-day own-history
median. Free-text metadata never enters it, and the sport string is narrowed
against the closed enum first, so the payload carries no text the user did not
generate through a measured field.
The prompt composes the shared assessment base and adds the session contract:
device-attributed, own-history-comparative, effort acknowledged where the
numbers show it, and no training prescriptions of any kind. Registered in the
tone harness beside the assessment cards.
A step-up elevation minted from the account password satisfied the fresh-factor
routes. That is a second-factor bypass: a stolen token plus a known password
could rotate the recovery codes — which take no factor in the body — and then
spend one of them to disable the second factor outright. The web has never
allowed that chain, because a password login does not stamp mfaVerifiedAt.
The method is now an authorisation input. Only totp, webauthn, and passkey
satisfy a fresh-factor gate, which is exactly the set of ceremonies the web
stamps a session for; password reaches what a plain cookie session reaches and
stops. The mint gains TOTP and security-key arms so a TOTP-only account can
still manage its factor natively, and the response says which reach it bought.
The gate also returns a discriminated context. session.id meant a session row on
one arm and a token row on the other, and the disable route passed the token id
to a session-scoped query — excluding nothing, so it deleted every browser
session and revoked the caller's own device login. The transport is now in the
type and the compiler refuses the confusion.
Redemption splits into validate-then-commit so a 429, a 422, or a wrong code no
longer burns a proof against a five-per-fifteen-minute mint ceiling. The claim
is UPDATE ... RETURNING, so the statement that decides is the one that reports.
The mint is transactional behind a partial unique index. Sign-out-everywhere and
both password-reset paths clear elevations. The status read drops to plain
requireAuth — it carries no credential material — and every mutation stays gated.
Each of these was already correct in the tree and had nothing asserting it, so
a refactor would have regressed it silently.
The decrypted-snapshot LRU keys on userId, but only same-user memoisation and
scope-change recomputation were covered — dropping the userId segment would have
served one account's health snapshot to another with a green suite. The project
has shipped one cross-user cache-leak fix already; this is a repeat class.
The GlitchTip forwarder redacts path-segment secrets in the url it sends. The
query-string strip was tested; the path-segment pass was not, and that is the
one that reaches a durable external incident UI.
The Withings credentials DELETE narrows its connection-row delete to P2025 and
rethrows everything else, so a real failure can no longer null the client
credentials and report a successful disconnect while the encrypted OAuth tokens
survive. The route had no test file at all.
Its own table, cascading from the workout row, because workout deletes are
hard deletes. The paragraph is AES-256-GCM at rest and registered for key
rotation; the input hash and prompt version make a re-post of the same
session a free no-op.
The deterministic fallbacks nine metric surfaces render when no provider is
reachable carried German and English bodies only. `getLocalizedText` returned
`locale === "de" ? de : en`, so French, Spanish, Italian and Polish readers got
English — on the no-key path, an outage, a spent budget or a timeout, which is
precisely when this text is the whole assessment.
Every string now has a body per locale: the verdict leads and value sentences
of the signal-grounded composer, the per-metric labels and pointers, and the
eight no-signal floors. Selection reads the reader's locale; English stays as
the structural floor for a locale that is ever genuinely missing a body.
The bodies are written per language, not translated word-for-word. fr/es/it
replace the comparative adjective ("higher than your usual average") with a
preposition — au-dessus de / por encima de / sopra — because a comparative
would have to agree in gender with the metric noun, and the labels do not share
one. Polish restructures further: the sentence takes `wynosi`, which is
invariant for the subject's gender, and the comparison uses powyżej / poniżej
with the genitive; the labels span all three genders, so nothing that inflects
can sit in a shared template.
The tone harness grades the four locales by the same rules rather than passing
them by default: false cheer, clinical-verdict and imperative-opener vocabulary
extended to all six, the no-assessment sentinel keyed per locale instead of a
de/en binary, and a byte-identity guard that fails if any locale silently falls
back to the English text again. The value-led-opener rule was rewritten as
well — it assumed the copula sits against the figure, which under-matched its
own second locale (German "liegt aktuell bei 24") and would have missed all
four new ones.
The batch route now also resolves the user's timezone and module map through
the arrival seam, so a raw call count on user.findUnique no longer isolates
the read this guard is about. Filtering by the SELECT shape keeps the
once-per-batch contract pinned while letting the unrelated cached lookups
through.
The water quick-add validated calendar realism but never the upper day bound the
batch route applies, so a far-future day was accepted and — every read filters
`day >= since` with no ceiling — became the permanent latest day on the settings
card and the dashboard tile. Hoist the bound into one shared helper both routes
use.
`mode: "add"` bounded a single request to the catalog's plausible daily max but
nothing across requests, so repeated quick-adds drove the stored total
arbitrarily high while the batch route enforced the same cap strictly. Clamp
after the increment so the write stays atomic.
Thiamin's cap sat at exactly 1000 mg while its PRI is exactly 1.0 mg, so a
microgram-read-as-milligram sample landed on precisely the cap and the `>`
comparison let it through. 500 still clears any real supplement dose.
NutrientIntakeDay was in no export path, which contradicted the schema's own
reason for denormalising the unit column. Carry it in the full backup with its
unit and source — source is part of the composite key, so without it a restore
cannot tell a manual entry from a synced day total.
The proxy's seven early exits — worker-only 503, two legacy 301s, the demo-mode
403, and the login / onboarding / MFA redirects — returned before the security
header block, so they carried no HSTS, no framing refusal and no nosniff. Factor
the transport subset into a helper and apply it on every exit. The per-route CSP
stays where it is; it is nonce-bound and only meaningful for a response with a
body.
The WebAuthn relying-party candidate list always carried http://localhost:3000,
including in production. The signed rpIdHash still pinned the real domain, so
this was never a reachable bypass, but a configured production origin has no
reason to carry it. Gate the fallback on development, keeping it when nothing is
configured so getRpId() still resolves.
ORPHAN_ENUMS was a hand-maintained literal that drifted every time
ALLOWED_CHART_TOKENS grew — eighteen allowlisted metrics had no strip entry, so
a model writing 'Your PHQ9_SCORE is elevated' leaked the raw enum into rendered
prose. Derive it from the allowlist and pin the coupling with a test.
`classifyIntakeTiming` treated `windowStart === windowEnd` as an overnight
wrap, inflating a degenerate point window into a ~30h on-time band — a dose
taken many hours late still classified `on_time` and fed a perfect-compliance
streak. Widen the point window symmetrically instead, matching the treatment
`resolveLegacyWindowBounds` already applies.
`/api/measurements/batch` never applied the entry-instant plausibility bound
its single-entry twin has carried since v1.17 W1b, so a future-dated row landed
and permanently won 'latest reading' while poisoning the rollup buckets and the
canonical picker. Skip per entry, like the range guard. The past floor stays at
1900 — a historical backfill is a legitimate shape here.
The intake CSV import had neither an upper date bound nor a rate limiter, while
running up to 1000 sequential find + create + consume iterations per call. Add
both, on the bulk twin's bucket shape and budget.
The no-signal floors were the coldest text the product renders. They run when
there is nothing to ground against, and they opened on a bare order — "Measure
blood pressure at rest and under comparable conditions", "Evaluate mood trends
over several weeks" — so the one moment the surface has nothing to say was also
the one moment it told the reader what to do.
They now open on what the surface can honestly say: no assessment is available
right now. That is the meaning-first opening for this case, and it is the only
one available, because a floor with no signal may not imply a read. What
follows describes how the measure behaves rather than ordering the reader
around. The signal-grounded lines above them are unchanged.
The prompt version bump re-fingerprints the assessments cached under the old
biomarker prompt; the per-card input hash folds the constant in, so the
regeneration is one-time and paced by the nightly warm.
Backfill immunity is proved by replay, not by a hand-picked case: a ten-year
Apple Health export (12772 writes) and a 30-day provider re-sync (58 writes)
must both emit zero events, while a fresh sleep night, workout and first
weight of the day emit exactly one each.
The provider-isolation guard walks the worker's static import graph and fails
if any provider client or resolver becomes reachable. It carries two positive
controls, so a broken entry path cannot make it pass against an empty graph.
The integration test covers what unit mocks cannot: the unique index, the
ON CONFLICT DO NOTHING claim, the losing side of ten concurrent claims, the
forward-only marker, ciphertext at rest, the delete cascade, and the
retention window.
The German prompt-injection test asserted only inside `if (decision.refuse)`,
so it stayed green whether or not the filter fired. Un-guard it, drive it from a
case table, and widen the two German patterns to cover the determiners that
walked straight past them ("saemtliche", "die", "deine").
The Oura / Polar / Fitbit token helpers cast a possibly-null `json` to the token
type on a status-only success gate. A 2xx with an empty or non-JSON body handed
callers a null they dereferenced, surfacing an unclassified TypeError instead of
a handled integration error. Narrow explicitly and classify as transient.
Every ingest path in the product can now emit a typed, salience-classified
event into pg-boss, consumed by one worker that claims a durable per-day
reaction marker and fans out to the reaction surfaces.
The recency classifier runs at the seam, before anything is enqueued, so a
mass backfill — a multi-year Apple Health export, a provider re-sync, a
30-day catch-up — emits nothing at all. Non-salient writes are annotated and
dropped.
The worker makes no provider call by construction: it imports no provider
module, writes only markers, and enqueues nothing that is not budget-gated
downstream. Business refusals return a status rather than throwing, so a
refusal is never retried against a ceiling that will not move until the local
day rolls over.
Seams: sleep (one hook, inherited by all seven transports), workouts (the
batch route, the four provider syncs, the export importer), weight and blood
pressure (both measurement routes), labs (manual, OCR commit, document fact
commit).
The provider workout syncs could not previously tell a created row from an
updated one, so a re-sync of a recent session would have re-fired on every
poll; they now recover the distinction from the row the upsert already
returns.
Reaction lines are AES-256-GCM at rest, following the InsightNarrative
precedent, and the column is registered for key rotation. Markers are pruned
after 14 days by a daily maintenance job.
A deterministic grader over the assembled assessment prompts and the
deterministic fallback text. No provider, no network, no database — it runs in
the normal suite in under half a second, which is the point: the live-judge
eval needs a key and can only advise, so the regressions that actually shipped
need a check that can block.
Three classes are blocking. A surface stops leading with meaning (it drops the
shared opening contract, or its user prompt loses the meaning-first
instruction). A locale collapses (fr/es/it/pl fall back to the German body or
lose the reply-language directive). A prompt reintroduces a value-first
instruction — matched with negation handling, since the shared prompts quote
that failure mode in order to ban it.
The deterministic fallbacks are graded on the same axis, including the
asymmetry that matters: a signal with no baseline supports no confident read,
so that one line honestly opens on the value. Pinning it stops a future
"always lead with meaning" edit from pushing the surface into claiming
something the data does not support.
A coverage guard fails when an assessment prompt module has no harness entry,
so the next surface cannot stay uncovered the way the biomarker card did.
The three unit suites that stubbed `requireCookieAuth` / `requireFreshMfa`
directly now stub `requireMfaManagementAuth`, and the password-change suite
stubs the elevation revocation the route gained. Adds the unit checks that need
no database: the elevation TTL is pinned to the cookie path's step-up window so
the two cannot drift apart, the minted value is a prefixed CSPRNG secret whose
plaintext never reaches the row, and a value carrying another credential's
prefix is refused without costing a query.
The per-biomarker assessment was the last surface in the family building its
prompt inline: a five-line scaffold that skipped the base system prompt, took
no opener hint, and instructed the model to "state the current value" — so the
one card where a bare number reads most like a verdict was value-first by
instruction while every sibling had moved to meaning-first.
It now composes exactly as the seven bespoke cards do. The new prompt module
carries a METRIC section that states the one thing this card needs and the
others do not: its snapshot has no signal block, so the read comes from the
pre-computed range placement plus the direction across draws.
Also point both archetype-driven cards at a registered budget. The generic
metric card and the biomarker card each carried temperature/maxTokens inline
at 0.45/1000, so the registry that documents every surface's ceiling was
silently wrong about the two widest assessment surfaces.
The single-marker no-key fallback replaces the multi-metric overview tip the
card used to render — text about watching several metrics at once, on a card
showing one lab value.
Adds the structural guard that freezes the exact route set an elevation can
unlock, plus the behavioural suite that drives each threat case against a real
Postgres: a token alone mints nothing, an elevation minted for one token is
refused for another (same account included), a second redemption fails, a
concurrent redemption lets exactly one winner through, an expired one is
refused, an elevation never reaches requireAdmin, a wrong password is throttled
and audited without saying what was wrong, and the cookie flow is untouched.
The OpenAPI registry carries both new routes and names the accepting set in
prose, so the client implements against the contract rather than against a
description of it.
Signing every other device out, revoking a single session, revoking a
trusted device and revoking them all each fired on a single tap. None is
undoable — the other devices simply have to sign in again, and a revoked
trusted device asks for its second factor next time — while every other
destructive surface in the app confirms first. On a shared screen a
mis-tap is a support call.
A sweep of the rest of the tree found two more hard deletes with neither
a confirmation nor an undo: a Coach reminder and a travel entry. Both now
confirm. Deleting a document deliberately does not: it soft-deletes and
offers an undo, which covers the mis-tap.
The four controls share one ConfirmButton rather than four copies of the
dialog, so the next destructive control gets the confirmation by using
it instead of by someone remembering. Copy is per control and states the
actual consequence — no claim of more finality than is true.
The widget catalogue grew to 42 ids, so the snapshot round-trip count
follows. The mood card's unit key carries "/5" rather than an empty
string — the locale-integrity guard forbids empty values, and the scale
is the honest thing to print next to a 1-to-5 score. It joins the
identical-across-locales allowlist alongside the other unit notations.
Second-factor management has been cookie-only since v1.23. The gate is right:
a Bearer token is a credential whose provenance the server cannot see, and
tearing down a second factor must not be reachable from the weaker of an
interactive login and a value pasted into a script. What was missing was never
Bearer access to those routes — it was a way to say, over a transport with no
session row, that the human holding this device just re-proved a factor.
An elevation is that proof. It is minted only against a re-proved account
password or primary passkey, bound to the exact token that re-proved, single-use
via a conditional UPDATE, five-minute-lived to match the cookie path's step-up
window, and stored hashed. A stolen token mints nothing; a second token cannot
redeem another's elevation; a password change drops every one the account holds.
requireMfaManagementAuth is the one gate the management routes share. On a
cookie session it delegates to the original helpers unchanged, so the web flow
runs the code it always did. requireAdmin stays cookie-only and out of reach.
The native pen list renders these two as its headline and subhead. The
server held neither, so a container registered on the web could not be
shown there without fabricating an empty card, and the client kept its
own copy locally with no back-fill possible.
Both columns are additive and nullable (migration 0255). Every existing
row is truthfully NULL — the values were never collected, and a plain
supply row such as a blister pack or a bottle has no use for either.
Strength stays free text rather than a parsed number + unit: pens state
it per dose, per ml, or per cartridge depending on the product, and
forcing a numeric split would lose which of the three was printed.
On PATCH an omitted field leaves the stored value untouched, the same
rule notes already followed, so a client that does not know the fields
cannot blank what another populated; an explicit null still clears. The
web add-container form collects both for a PEN, so the columns do not
ship empty forever.
Three surfaces where the words no longer matched the code, plus the missing
web control for a preference the server already honours.
Delete confirmations. Four dialogs promised permanence on surfaces that
render an Undo: the measurements and mood bulk bodies, the lab-result body,
and the illness-episode body (which said "cannot be undone" sixty lines
below the code that fires the Undo toast). All four now use the same
sentence the single-row measurement and mood dialogs already used. A sweep
of every other confirmation in the tree found the rest accurate — the
biomarker cascade, the mood-tag purge, and the account purge keep their
warning because those deletes really are unrecoverable.
Sleep debt. v1.19.0 replaced the summed shortfall with a running balance
that credits catch-up sleep; the labels still described the old sum, so the
word contradicted the number for anyone who slept in after a short night.
Corrected in all six locales, plus the places the old semantics had
propagated: the Coach system prompt, the OpenAPI field descriptions the
native client reads, and four module docblocks.
Glucose reference band. `hasDiabetes` selects the tighter reference band and
grounds the Coach, but no web surface set it — a web-only account was pinned
to the non-diabetic scale with no way out. The control goes on Targets,
above the per-metric editor, because it answers the same question that page
answers and the two are ordered: it picks the default band, an override
below beats it. The copy says what it is — a reference band, not a
diagnosis, never inferred from a reading, no clinical action taken on it.
The passkey list was reported as always reading "never used". It does not
reproduce: the column is selected by the list route and stamped on every
verified assertion, both since v1.23.0. Pinned with tests on both halves so
the contract cannot quietly regress.
Both metrics exist in the record but the summary route emitted neither,
so a native client could render every home-screen tile except those two.
Mood reads the canonical daily series (`buildMoodDailySeries`) — the same
rollup-backed aggregation the web mood analytics and the dashboard
snapshot render — and emits the newest day's score with a trailing
seven-day sparkline. BMI comes from `computeBmi`, the one implementation
the web vitals dashboard already renders through the derived-metric
dispatch; the route resolves the figure so no client re-derives weight
over height squared.
Mood gates on the mood module through a synthetic `MOOD_ENTRY` key, the
same route `NUTRIENT_WATER` takes, because mood is not a measurement
type; without the join the card would ship past a disabled module. BMI
belongs to no module and rides through with the other core vitals.
The BMI card is omitted rather than emitted empty when the profile has no
height or no weight in the window: the card shape carries no honest
"unknown" state, and a null value would read as a logging gap rather than
a missing profile field.
The layout PUT preserved chartOverlayPrefs, selectedScoreRings and
heroRingOrder when a client omitted them, but the top-level
comparisonBaseline was not in that merge. serializeDashboardLayout clamps
a missing baseline to "none", so a layout save from a client that does
not know the field — the native client treats it as web-only and never
sends it — silently reset the comparison the user had picked on the web.
No error, no audit row.
Replace the hand-maintained three-field merge with an explicit
disposition map over every top-level layout field. The map is declared
`satisfies Record<keyof DashboardLayout, ...>`, so adding a layout field
without deciding replace-or-preserve fails typecheck; the preserve set
and the merge are derived from it rather than restated. The stored-layout
read is skipped when the client sent the full set.
Also catalogue the four clinical signals the native client pins — grip
strength, pain NRS, waist circumference, waist-to-height. They reached
the route as unknown ids, were filtered out ahead of Zod, and vanished
from the persisted layout, so a tile placement was lost on every save.
They join the iOS-only set because the web dashboard has no render path
for them; their charts live on the insights sub-pages. Raise the widget
array cap to 50 to clear the now-42-id catalogue.
Also closes the device list's exposure of the same value, refuses a
cross-device refresh rotation, replaces a vacuous expiry test with one
that fails when the check is removed, and asserts the Nightscout host
guard at save time rather than relying on the probe's placement.
The settings mount fires an idempotent consent heal so an account that
predates the gate does not fall into the no-key state. It looked only for
an active receipt, so a deliberate withdrawal — which sets revokedAt —
read the same as never having consented, and the next visit to the AI
settings granted again, writing an audit row for an act the user never
performed. GDPR Art. 7 (3) puts withdrawal on the same footing as the
grant; one undone by a page load is not a withdrawal.
The mint now distinguishes the two callers. The mount heal grants only to
an account with no consent history at all and leaves a revocation
standing; switching document auto-read on is the consent act itself and
may supersede an earlier revocation. The revocation is re-checked inside
the transaction so a withdrawal landing mid-mint is not overwritten.
The revoke endpoint has existed since v1.4.40 but only the native client
called it: on the web, consent could be given and never taken back. The
AI settings now carry the standing decision and the control that changes
it, in all six languages.
The assertion compared the cookie to the new session row's id, which only
held while the id WAS the cookie. Check the stored hash instead: it
proves the cookie authenticates the surviving row and that the row id
never reaches the client.
The cookie was the Session row's cuid. cuid is built for collision-
resistant identity across distributed writers, not for unguessability —
it encodes a millisecond timestamp and a process-local counter and never
touches a CSPRNG. A session cookie is a bearer credential and has to be
unguessable, so it now carries 32 random bytes and only the HMAC of that
value is stored, matching how api_tokens and refresh_tokens already hold
their secrets.
Compatibility is passive, and deliberately so. A backfill cannot help:
the cookie already sitting in a browser is the cuid, and nothing we
generate server-side can be pushed into it — backfilling would sign
everyone out at deploy. Upgrading a row in place on first use loses to
the parallel requests a single page load fires: each mints its own
secret, one write wins, and the losers cannot re-emit the winner's
secret because only a hash is stored, so the browser can end up holding
a value the row does not honour. Instead, rows minted before the column
keep a NULL token_hash and stay resolvable by id, and are denied the
sliding-expiry extension so they drain within the 30-day session
lifetime. Every login from this release forward is on a secret at once,
and an id stops authenticating the moment its row holds one.
Note for operators: API_TOKEN_HMAC_KEY now also keys session cookies, so
rotating it signs browser sessions out as well as invalidating tokens.
It is already Core-required, so no deployment gains a new prerequisite.
Egress was already correct — the client fetches through safeFetch with
requirePublicHost pinned off the user's opt-in, manual redirects and a
timeout — but the check on the value we STORE was only a side effect of
the live probe the connect route happens to run first. That makes the
guard contingent on the probe's placement rather than on the input.
Run isPublicUrl at input time, gated on the same self-hoster opt-in, so
the URL that gets encrypted onto the row is the URL that was validated.
Cover the alt-notation classes the helper knows about; each case asserts
the probe was never reached, which is what distinguishes the input-time
refusal from the fetch-time one.
Rotation refused a replayed token under a spoofed device id (M-4) but
still rotated a *live* one presented from a different device, and then
inherited the caller's id onto the new row — putting the resulting family
outside the victim device's containment scope. Refuse the mismatch
outright when both ids are present; an absent id on either side stays
unattributable and rotates as before, so a client that sends no
X-Device-Id is not locked out.
The expiry refusal had no coverage at all: the standing test presented a
token that was never issued, so it passed on not_found and stayed green
with the expiry branch deleted. Replace it with the real secret plus a
just-inside-the-window companion, and cover the device binding.
Field-by-field comparison of the handlers against the registered Zod
schemas across measurements, workouts, medications, daily and auth. The
spec was corrected to match the handlers; no runtime behaviour changed.
Measurements:
- GET /api/measurements returns `{ measurements, meta: { total, limit,
offset, … } }` on all seven of its modes. The spec claimed a top-level
`total` that no arm has ever emitted.
- The list rows are stored rows minus the notes ciphertext, or synthesised
day-sum / sleep-segment rows, or pre-folded aggregate buckets. The
resource declared seven fields; the rest are now on the wire contract.
- POST accepts a bare ARRAY body (the combined blood-pressure write) and
answers with an array, and returns 409 on the dedup index.
- GET/PUT/DELETE /api/measurements/{id} carried no path entry at all.
Workouts: both GET operations run the module gate and answer 403
`module.disabled`; the reusable block existed but was never spread. The
batch ingest stays deliberately ungated and undocumented for it.
Medications: `nextDueAt` / `nextDueOverdue` are computed by the read paths
and absent from the create/update results, so they cannot be required on
the shared shape (the list row keeps them required). `stockUnitsRemaining`
sums a Decimal(12,4) and is fractional, not an integer. The cadence chips
object documented four fields that do not exist and missed the five that
do. `IntakeSource` admits APPLE_HEALTH.
Auth: the token bundle had its optionality inverted — `user` is always
returned, the token pair only for native callers. Adds the undocumented
403 `oidc_only`, the passkey 403/404, the TOTP enrolment 409s, the
attestation 400, the no-key-registered 409, the security-key 404s, the
refresh `revoke` response shape, and the path/query parameters three
operations read but never declared.
Daily was checked and is accurate; so are the batch ingest, series,
series-batch, bulk-delete, restore and by-external-ids routes.
persistRollupRows scoped its delete to the [min, max] bucket range of the rows
the aggregate RETURNED. That made the repair self-limiting in exactly the case
that needs it: a recompute could only overwrite a bucket the live aggregate
still produces, so a bucket whose source rows had all gone was never in the
delete range and kept its stale aggregate.
Three shapes of the same hole. An empty aggregate skipped the write entirely
(the caller short-circuited on rows.length === 0). A returned range narrower
than the requested window left the edge buckets untouched. A type that lost
every reading in the window dropped out of the returned rows, so the per-type
delete filter could never reach it.
The DAY write hook masked it — it deletes its own bucket by hand when the day
empties — but nothing covered the WEEK / MONTH / YEAR worker, which recomputes
one bucket span at a time: deleting the last reading of a type in a week
dropped the DAY bucket and left the coarser ones stale forever. The coverage
probe still reported hasBuckets = true, so the read-swap served them.
A recompute now prunes the window it was asked for and writes its rows in the
same transaction. The single-bucket write hook keeps its own scope, which is
correct there because it already deletes its bucket when the aggregate empties.
Still one tier: the rollup replaces the legacy read and falls back to live SQL
only on a coverage miss.
The per-document summary job is enqueued at upload time and no-ops while the
documentsAutoAiRead opt-in is OFF. Nothing ever re-enqueued it, so the switch
only applied to documents uploaded AFTER the flip — a user who filled the vault
first and opted in later had documents that would never be read, and the
setting looked inert.
A genuine OFF->ON transition now schedules one bounded catch-up pass over the
user's un-summarised documents. It grants nothing of its own: it only enqueues
the ordinary summary job, which still re-reads the opt-in, re-asserts the
per-document egress consent and reserves the daily token budget. The pass
itself re-reads the opt-in first, so a toggle flipped back OFF before it runs
egresses nothing.
Bounded by an id-cursor walk capped at 200 documents per pass (a summary is a
provider call, unlike the local-compute thumbnail backfill). Idempotent at four
levels: the trigger fires only on a real transition, the per-user singleton
collapses a double toggle, the candidate set excludes summarised documents, and
the summary job skips one that already has a summary.
Every staged fact carried a sourceText the model echoed back, stored verbatim
as 'the exact verbatim span of the document the fact came from'. Nothing
checked that the span existed. A paraphrase, a normalised rendering, or an
outright invention was persisted and shown to the reviewer as a transcription
of their own clinical record — an anchor pointing at nothing, or at a span the
document never contained.
Treat the echo as a lookup key instead. anchorSourceText resolves it against
the extracted text (exact, then whitespace-collapsed and case-folded to absorb
OCR reflow) and stores the span the document actually carries, with its offset.
A quote that matches nothing is dropped and the fact is flagged unanchored;
vision mode has no extracted text at all, so its facts are unanchored by
construction. Rows written before this resolve to unanchored on read — their
quote is preserved but never presented as checked.
The retry suffix still described the strict insight wrapper that was removed
long ago: it demanded citations[] and warnings[] — fields insightResultSchema
does not have — required metricSource, rationale, id and severity on every
recommendation when all four are optional, and named none of the five fields
the live schema actually requires (classification, findings, correlations,
dataQuality, disclaimer).
So the one corrective pass steered the model away from the contract it was
being corrected towards. A reply that obeyed the correction could not
validate, fell through parseComprehensiveResult's raw-object escape hatch,
and the downstream surfaces read an unvalidated payload.
Describe insightResultSchema and nothing else, and carry the contract as data
in RETRY_CONTRACT_EXAMPLE so a test can assert prompt and validator agree.
The dashboard, Insights, the workouts list, the medications list and the
Coach server-prefetch their data and serialise one account's health record
into the HTML document through a TanStack HydrationBoundary. Those pages
are dynamic today only as a side effect of reading the session cookie —
a property of the page code, not a guarantee. A stray revalidate export, a
force-static segment, or a caching proxy in front of the app would be
enough to retain one account's record and hand it to the next caller.
Stamp `Cache-Control: private, no-store` at the edge instead, driven by the
same predicate that governs the auth redirect, so a page cannot be
session-gated without also being uncacheable. Public legal, auth and
locale-catalog surfaces keep their cacheable posture; API routes keep
managing their own caching.
Two guards: the proxy test pins the header on gated pages and its absence
on the public ones, and a structural test walks the app tree and refuses a
prefetching page that sits on a public path or opts into static /
revalidated rendering.
A summary is enqueued only when a document is uploaded while automatic
reading is already on, and nothing backfills the ones uploaded before
that. The detail view read a null summary as 'being generated' and said
so permanently, while the manual action row was hidden precisely when
automatic reading was on - so the promise could never be met and the
repair was out of reach.
Record what actually happened. Seven outcomes previously collapsed into
the same null: never enqueued, opted out, no vision provider, spent
budget, consent refused, unreadable file, provider error, plus the
withhold. Each now names itself and offers a manual generation, and what
that produces is kept.
The withhold rule is unchanged - refused prose is still never shown or
stored - but recording the refusal as a state makes it re-attemptable
instead of a permanent dead end.
Also removes the read-by-AI footer.
A READY row whose ciphertext will not open reports UNAVAILABLE, so the view
does not render a heading over nothing. Nothing asserted that, and removing the
degrade left the suite green.
Opening a document said "the summary is being prepared" and kept saying it.
The view had one fallback for a null summary column and used it for every
outcome: a document nobody ever enqueued, a missing vision provider, a spent
budget, a withdrawn consent, an unreadable file, a provider error, and a
summary the outbound screen withheld all rendered as work in progress.
The common case is not the screen. The background job is enqueued at exactly
one place — a fresh upload with the auto-read opt-in already on — and no
backfill reaches anything else. Every document uploaded before the toggle went
on has a null column no job will ever fill, and the view promised all of them a
summary that was never coming.
Record the outcome as `summary_state` (migration 0254, additive, defaults to
NONE which is truthful for every existing row) and branch on it. Only PENDING,
set when a job genuinely exists, may claim generation is running. NONE,
WITHHELD and UNAVAILABLE each say what happened and offer to generate one.
The withhold keeps its rule and loses its dead end. Blocked prose is still
never shown and never stored — only the fact of the refusal is kept, and it is
no longer terminal, so asking again re-runs the gauntlet. That was the original
intent; leaving the column null defeated it, because null read as "generating"
and the guard meant nothing would ever overwrite it.
The on-demand summary now persists what it produces. It is the only repair path
for the documents the job never ran for, and re-deriving the same paragraph on
every open costs a provider call for nothing. Generation is still only ever a
click — nothing runs because a view rendered. Transcription stays session-only.
Also drop the "read by AI" line from the detail footer.
The Today hero carried a read-the-full-briefing link to /insights while
the health-score ring above it already opened the same page — two doors
to one room in a single card. The ring stays; the link and its six
locale strings are gone.
The document sheet's close control sat at the plain p-4 right edge, but
an icon button carries its own inner inset, so the glyph landed ~28px
from the frame against a title starting at 16px. Trim the right padding
by that inset when a consumer renders its own trailing control, leaving
the tap target reaching the frame.
Also rewrites the v1.30.15-v1.30.31 changelog entries: they had drifted
into summary lines, internal vocabulary and reasoning asides that do not
belong in a release note.
The Today hero stayed on its all-clear line while a dose was actually due:
the dose-window item only fired once the slot was already overdue, so a
dose the user could take right now produced no rail item at all. Emit the
item from the moment the dose becomes takeable — the on-time window's lead
plus the early grace, the same instant logging it starts crediting the
slot — and give the two faces distinct wording and status: a due dose reads
calm and informational, an overdue one keeps the warning. A settled day
still emits nothing, and a cached block whose anchor has since passed keeps
rendering neither face. Push reminders are untouched; they still fire only
on overdue.
The briefing grounding gate discarded the whole block over one number it
could not verify, and it could not verify plenty of legitimate ones: the
allow-set was a hand-maintained list of feature blocks that had already
drifted from the prompt three times, and still omitted medication
compliance, the context counters, the correlation coefficients, every
coverage object and the comparison snapshot — all of which the prompt hands
the model verbatim. Derive the allow-set by walking the prompt payload
instead, so grounded means exactly what it says: a figure the server gave
the model. A fabricated number still matches nothing.
Fix the disposal to match. A content-failed retry left a hole while a
transport-failed one fell back to the previous cached briefing, though in
both cases the rejected text is discarded and the previous briefing passed
this same gate when it was written. Fall back on both paths, and only
report the briefing as withheld when there is genuinely nothing to show.
The on-demand route had no fallback at all; it now mirrors the background
generator that writes the same cached payload.
With the error notice hoisted to the page, the three cards fell through
to their skeleton on a failed read — a spinner that can never resolve.
Return null instead, so a card rendered outside the page's gate is still
honest about having nothing to show.
The rhythm, average-sleep and chronotype cards on /insights/sleep share
a single ["sleep-rhythm"] read, and each rendered its own copy of the
same failure sentence — so one failed read painted the identical notice
three times, none of them offering a way to try again.
The page owns the shared state now: one notice with a retry in place of
the card row. The cards keep their loading and empty handling and drop
the duplicated error branch.
Importing the plural helper directly from the dashboard trend tile cost
~78 KB gz across the route chunks: the module is tiny, but a new edge
from a widely-instantiated component gets duplicated into every chunk
that contains it rather than hoisted. The provider is already in every
chunk, so expose a plural-aware tCount() there and let components call
that instead. Total client JS goes back under budget.
Also fix the two tests the honest-loading-state change invalidated: the
mental-wellbeing landing tests seeded a never-resolving read and so now
render skeletons rather than instrument cards, and the achievements page
needs a router in scope for its module-gate redirect.
The reminder, coach and plural fixes were pinned; the three formatter
fixes were not. Add the duration abbreviations across all six locales
and a structural check that the cycle calendar, device-score tile and
sleep-stage axis pass a resolved app locale to Intl rather than
undefined, which silently resolves to the browser locale.
"No data yet" is a claim about the user's record. Making it when the
read failed states something false about their health history.
The mental-wellbeing surface was the one that mattered: the assessment
history read dropped isError entirely, so a failure left both
instrument cards showing "not taken yet" and the detail sheet showing
an empty trend. Someone with months of screening results was told, on
a mental-health surface, that the results did not exist. It now tells
loading, failed and genuinely empty apart, with a retry on the failure.
Also swept:
- the workouts insights page read isEmpty but not isError, so a failed
list read fell through every branch to null and painted a titled
shell with nothing in it;
- the micronutrients card (not in the original list) rendered its "no
nutrient data" empty state on a failed read, while its sibling
hydration card already used the shared error card;
- the medication detail page and the custom-metric values list showed
an honest error with no way out. Both keep retry: false — right for
a deleted row, wrong for a transient failure — and gain a
user-initiated retry instead, which cannot storm;
- /achievements rendered a blank page on a direct hit with the module
disabled. It now bounces to /insights like every other gated page.
The achievements query error branch, the custom-metric error branch and
the chronotype notice were already correct and are left alone.
Four locales received English on surfaces whose translations were
already complete, because the resolving code was a de/en binary.
- Medication reminder push and the Telegram inline-keyboard labels
admitted only de/en, so es/fr/it/pl got the English template on every
channel. Route the resolver through a shared coerceLocale().
- The Coach's trailing "reply now in X" clause said English for any
non-German locale, overriding the correct system-prompt directive.
Read the language name from a shared map instead.
- Coach refusal copy was a de/en constant pair, answering four locales
in English at the one turn the Coach is declining to help. Move the
copy into the bundles; the EN/DE constants now derive from them.
- The sleep-stage chart formatted axis ticks as en-US for four locales;
the cycle calendar and the device-score tile followed the *browser*
locale rather than the app locale.
- Three components carried a byte-identical duration formatter that
branched on German. Lift it to one key-driven helper.
Polish needs a third plural tier: 2-4 takes a "few" form a two-form
scheme cannot express, so "2 godzin temu" read as "2 of-hours ago".
Add a pluralKey() helper driven by Intl.PluralRules and a Few tier to
the relative-time and stale-hint keys. The five locales whose rules
have no "few" category resolve exactly as before by construction, and
their Few values mirror Other. Teach the reverse-coverage guard to
expand a pluralKey() base into its three tiers.
generateInsight() parsed against the strict aiInsightResponseSchema,
cross-checked that every recommendation's metricSource appeared in
citations[], and threw InsightSchemaError on a second failure. Its only
caller was runWithFallback(), which had no production caller — every
live surface runs runRawCompletionWithFallback. So none of it executed.
Wiring it into the live path was the alternative and it is not viable:
the live insightResultSchema has no citations[] field at all and treats
metricSource as optional, so the cross-check would have rejected
essentially every generation rather than tightening anything.
Removed: runWithFallback and its param/result types, generateInsight
and its private helpers, findUncitedRecommendations,
findRecommendationsMissingRationale, the never-thrown InsightSchemaError
and the isHardProviderFailure branch that excluded it, and
computeConfidence — a deterministic override that only ran inside the
dead wrapper.
Kept: buildRetryCorrectionMessage, which comprehensive-generate sends on
its own retry leg; the file it lived in is renamed to match what it now
holds. aiInsightResponseSchema stays as the shape that message
describes. Note the retry prompt asks for a stricter payload than the
live validator requires — that mismatch predates this and is left alone.
computeCitationCoverage is untouched and still live: it is a metric, not
a gate, and v1.30.24 wired it into the comprehensive post-parse step.
Chain-runner coverage the deleted tests carried (last-working cache,
empty-input guard) is ported onto runRawCompletionWithFallback, which
shares the machinery.
Three things the write surface claimed but did not hold.
Scope: the run bodies never re-checked ctx.canWrite. Enforcement was
real but lived entirely in the registration decision, and the test
asserted registration only — so a refactor that registered the write
tools unconditionally and meant to refuse at call time would open the
surface with the suite still green. Wrap every run body in a scope
guard at the registry level, so a new write tool inherits it, and drive
the run bodies directly from a read-only context in the test.
Free text: medication names, doses, treatment classes and lab analyte
names were interpolated into tool prose with no delimiter. A lab
analyte name is transcribed out of an uploaded document, so a hostile
document chooses it, and the write tools ride the same wire. Fence the
assembled prose with the shared data-fence markers and scrub short
result titles, which would leak markers into a host's result list.
Confirm: the header called the confirm flag a PROTOCOL-LEVEL CONFIRM
GATE. It is a client-asserted boolean with nothing binding it to a
prior preview. Binding it to a server-issued preview token would prove
only that a preview call happened, which any client can do, so it buys
no human-in-the-loop for a round trip and server state the stateless
transport does not want. Describe it as the convention it is and name
what actually bounds the blast radius.
The changelog merge left conflict markers inside the CACHE_VERSION
assignment. The version generator only rewrites the version string, so
it could not repair the broken structure, and every service-worker test
failed while both branches were green on their own.
enforceBudget and recordSpend have no callers left. Both defects they
carried recurred every time a surface copied them: a TOCTOU window
between the read and the bump, and a cap parameter defaulting to the
operator ceiling so an omitted argument rationed a self-hoster paying
for their own key. Deleting them means a future surface cannot reach
for the older, simpler-looking helper and reintroduce either.
getDailyTokenSpend goes with them — it existed to feed the read half.
The ledger-poisoning clamps recordSpend carried (NaN, negative,
fractional token counts) move to reserveBudget, where the property now
lives. Refresh the doc comments that still narrated the old flow.
The route used the enforceBudget + recordSpend read-then-write pair and
passed no cap, so every call measured against the operator ceiling — a
self-hoster on their own key was rationed by the operator's bill while
paying for their own usage. It also left the TOCTOU window the rest of
the codebase closed: concurrent calls could each read a sub-cap total
and all proceed.
Move the gate after the chain resolves and reserve atomically against
resolveDailyCap(chain), matching runStatusCompletion and the document
paths. Reconcile on every exit: the provider's reported count on a
completed call, a full refund when nothing was generated.
Pin the four guarantees at the new chokepoint and at both call sites: spend
lands on the ledger, over-cap is refused before any egress, a BYOK chain is
measured against the user-plan ceiling rather than the operator's, and a
failed call refunds in full. Also pin the ordering the consent gate depends
on — a consent-blocked user is never charged.
The fixture described a clean resting series but supplied rows for only
twenty of the thirty days it fed PULSE for. Now that the resolver merges
per day, the ten resting-free days are estimated from their own readings
— and on a day whose only readings are a 150 bpm workout the honest
estimate falls outside the band. Match the spans so the case tests what
its name claims; gap-day behaviour is pinned in resting-pulse.test.ts.
The status tier funnels every generation through runStatusCompletion, which
reserves against the day's ledger before egress and reconciles after. The
comprehensive briefing did not: the shared generator and the on-demand
POST /api/insights/generate route both called the provider chain directly, so
none of that spend reached coach_usage and no daily ceiling applied to it.
Route every briefing provider call — first pass, JSON-shape retry,
grounding-correction retry, and the daily paragraph re-roll — through one
chokepoint that reuses reserveBudget / reconcileSpend and resolveDailyCap. The
reservation lands after the consent gate, so a consent-blocked user is never
charged, and a failed or timed-out call is refunded in full.
Over-cap is a typed skipped outcome for the off-request generator (no upstream
fault to retry against, so no failure marker and no retry enqueue) and a 429
for the on-demand route.
Each case fails on the pre-fix code: the rollup tests on an unaligned
recompute window, the batch tests on the older stats: snapshot winning,
and the resting test on a series that stops at the fold horizon.
resolveRestingPulseSeries returned the resting branch as soon as a
single RESTING_HEART_RATE row existed and never consulted the PULSE
proxy again. Since the retention fold began minting derived resting
rows for the days it folds, a proxy account carries derived rows for
old folded days and nothing but raw PULSE for the recent ones — so the
visible series stopped dead at the fold horizon.
Merge per day instead: a day with a resting row uses it, a day without
falls back to the proxy. which stays a three-value labelling signal and
reports proxy as soon as one point is estimated, so a partly-estimated
series is never labelled resting heart rate.
Widen the intraday baseline read to pull PULSE history when the newest
resting row is stale, not only when the rows are few — the count gate
alone pinned the baseline to the fold horizon.
Two entries carrying the same stats: key in one batch both miss the
existing-row probe, so the overwrite branch never runs, both land in the
bulk insert, and skipDuplicates keeps whichever the unique index saw
first. The older snapshot of a per-day cumulative total or of a filling
ten-minute heart-rate bucket therefore won and the newer, larger figure
was dropped silently — both entries reported inserted, the counters
still summed, and the client checkpointed past both.
stats: rows carry overwrite semantics by contract, so within one batch
the last entry for a key is the authoritative one. Resolve it up front
and report the superseded earlier entries as duplicate with a
superseded_in_batch reason, a status the client already checkpoints
past.
recomputeUserRollups filtered measurements by a caller-supplied from/to
but grouped them by date_trunc, and persistRollupRows then deletes every
rollup row in the bucket range it is about to write. A from that landed
mid-bucket therefore replaced a complete bucket with an aggregate
computed only from the readings after that instant.
Every caller passes a moving instant — now minus ninety days on the
read-path refresh, an arbitrary instant on the coverage fallback, now
minus five years on the boot backfill and the restore path — so a
different oldest bucket was corrupted on each invocation and nothing
repaired it. Snap both edges out to whole buckets via bucketSpan, whose
own contract note already required exactly this of every recompute.
The red-team corpus probed only user messages and model replies. Neither is
the channel that reaches the prompt unguarded: a lab analyte, panel or unit
name, or a medication label, none of which the account holder authors when
the row was committed from an uploaded document.
The free-text guard's file list was hardcoded and had rotted — three of its
four entries were no longer reachable from the snapshot, and it covered one
of the dozen files that actually emit free text, which is why the lab fields
went unnoticed. Derive it from the import graph instead; it now covers 75
files, with a floor assertion so a resolver break cannot silently empty it.
Deriving the list immediately surfaced a second leak: features.ts shipped a
raw medication name and dose into the snapshot medications block.
Also fix sanitizeForPrompt itself. It deleted control characters, which
welded neighbouring tokens together and destroyed the word boundary its own
injection patterns anchor on, so a payload preceded by a newline survived
because of the newline. Replace with a space and collapse before matching.
The Coach SNAPSHOT was spliced into the prompt bare. Its JSON leaves are not
all server-computed: lab analyte, panel and unit names are transcribed by a
model out of an uploaded PDF and stored verbatim as the biomarker name, so a
hostile document chooses those strings. The about-me block has been fenced
since v1.16.1; extend the same pattern rather than adding per-field escapes.
- Move the fence markers and the wrapping primitive into a shared data-fence
module. Content is scrubbed against every known marker, not just its own
pair, so text in one block cannot forge another block's boundary.
- Fence the SNAPSHOT with an explicit data/instruction frame.
- Fence the OCR text handed to the document extractor; the previous
'OCR TEXT:' label was a label, not a boundary.
- Sanitise analyte, panel and unit at the labs source, which feeds both the
snapshot prompt and the labs_read tool result, and the medication label in
the adherence storyline.
- Capture a ---REMEMBER--- sentinel as proposed rather than active. It was
the only model-driven write with no confirm step, and the reminder recall
block fed it back into every later turn. Proposed reminders are already
excluded from that block and already carry a confirm control.
The status-provider suite drives the real budget module against an
in-memory stand-in for the day's usage row, so it exercises the atomic
reserve/reconcile path rather than asserting a mock was called: spend
lands on the ledger, an exhausted day is refused before any egress, a
failed generation is refunded, and a chain on the user's own key is
measured against the user-plan ceiling rather than the operator's.
The snapshot suite walks a 60-turn conversation and pins the full
snapshot to exactly two turns — the first and the elision crossing —
including the case where a lost reply shifts the turn parity.
The provider chokepoint every status and reference generator funnels
through ran with no budget accounting at all. Nine generator families
egressed behind it — the specialised status cards, the generic metric
cards, biomarker cards, the batched assessment, the derived scores, the
period narrative, and the off-request Coach memory workers (rolling
summary, fact extraction, plan proposals) — and none of that spend landed
on the daily ledger or met any ceiling. On an operator-key account it was
unmetered operator spend.
Reserve and reconcile around the provider call, using the same atomic
upsert the Coach and document paths use. The ceiling follows the cost
owner: a chain whose primary is the operator's own credential draws the
operator ceiling, everything else draws the user-plan ceiling, so a
self-hoster on their own key is never rationed by the operator's bill.
The status enqueue keyed its de-dupe singleton on (user, metric, locale).
The locale on a status read is client-chosen, so each of the six
supported locales opened its own slot and one legitimate generation pass
could be multiplied into six. Key on (user, metric) and carry the locale
in the payload instead.
The Coach re-shipped the full snapshot on every turn past the history
cap, not just the turn that crosses it, so a long conversation paid the
per-turn snapshot cost the snapshot-once model exists to remove. Re-ground
on the crossing only, tested as a narrow window rather than an equality
because a turn advances the count by one or two depending on whether the
reply persisted.
Give /api/ai/test a daily ceiling and a ledger entry. Its burst bucket
bounded rate but nothing cumulative, and with an empty body it resolves
the same chain generation uses — which can be the operator's credential.
Correct the memory-refresh worker comment, which asserted a budget check
inside the chokepoint that did not exist.
The coverage check hung off generateInsight, whose only caller runWithFallback
has no production caller left -- every live surface routes through
runRawCompletionWithFallback instead. The admin quality dashboard was reading a
metric nothing emitted. Widen the input to the structural shape both payloads
share and call it from the comprehensive post-parse step, so the annotation
reflects real generations.
Writing the locale tests found three gaps the English-only banks hid: Polish
verb suffixes fell outside the ASCII word class, a trailing word boundary after
an accented character never closed, and German subordinate clauses put the verb
last so the natural phrasing of a dose instruction never matched. The payload
collector also skipped legacy bare-string recommendations entirely.
The briefing gate grades numbers only, and only the paragraph, signals and
key-findings blocks -- so a digit-free risk assertion with a dose suggestion
attached passed it untouched, and recommendations and summary were not graded
at all. Screen every readable prose field against the insights contracts.
A safety violation cannot be surgically stripped the way an ungrounded number
can: the offending sentence may sit in any field, and deleting one would leave
the rest referring to it. So the background generator fails the whole run and
persists nothing, leaving the previous cached payload in place, while the
user-initiated route returns an explicit error for the same reason.
The causal bank lived in the narrative validator as an English and German list,
so the no-causation rule was enforced in code on one surface and for two of the
six locales it ships in. It now comes from the shared screen, which carries all
six, and the narrative also gets the dose and risk-score contracts it never
checked -- persisted prose with a dose imperative in it is as unsafe as a chat
reply with one.
Red-team cases carry their language. The German outbound case was probed as
English, which would have reported a pass the guard never earned.
The summary is model prose about the document and is contractually
descriptive-only, but it was returned to the browser verbatim with no guard.
Screen it against the insights contracts and let each caller apply its own
policy: the on-demand route replaces the text with a short honest statement
because a user is waiting on it, while the background job persists nothing --
its updateMany is guarded on a null summary, so a stored refusal would stamp
the document permanently with no path to regenerate.
Transcription stays unscreened by construction. It reproduces the user's own
document verbatim; a discharge letter that says to increase a dose is the
prescriber's instruction, and censoring it would delete the user's own record
from their own view. Locale is required on the summary input type and absent
from the transcription input so a caller cannot silently fall back to the
English banks.
The outbound safety screen ran on the Coach only. A dose-change imperative or
a fabricated clinical risk score written onto a per-metric, biomarker or
derived-score status card persisted as that day's cached assessment, while the
identical sentence in the Coach was caught and replaced -- the only transform
between provider and persistence was whitespace normalisation.
Move the pattern banks into a shared screen that takes the reader's locale.
The old signature took no locale, so the dose-refusal and risk-score contracts
had no enforcement for fr/es/it/pl even though the ground rules ship in all six
contract files. Banks now cover all six locales and each screen runs the
reader's bank plus English, since a provider often answers in English
regardless of the language directive.
Status cards withhold on a trip: serve the deterministic stub and persist no
model text. Nobody waits on a background-generated card, and persisting a
refusal would pin it to the card for the day.
The reconcile is the only thing that carries a policy onto a queue that
already exists, so whether it ran is the difference between the previous
release working and doing nothing on an upgraded instance. It reported
that through workerLog at info level, which is deliberately silent - so
the intended deploy signal could never appear, and the change was not
verifiable in production.
Emit a boot wide event instead, on every boot including the no-op one:
a reconciled count of zero on a later boot confirms an earlier boot
already migrated everything, whereas silence is indistinguishable from
the reconcile never running.
One suite per gate: the rich reads omit with an honest module_disabled
reason, the correlations reader omits and touches no channel, the
doctor-visit resource and prompt refuse without assembling the record,
and the clinician share collapses to its documents-only state. Each
asserts the operator availability path alongside the user toggle, and
that the gate short-circuits before the engine runs rather than building
the answer and withholding it.
The structural test pins that every metric the rich reads can resolve has
either an owning module or a written review decision, and is explicit in
its own comment about what it cannot prove — it does not show that any
read calls the gate. Writing it surfaced an unowned HRV_RMSSD, the
fallback type an RMSSD-only account resolves to, which would have served
recovery data the primary type refuses.
The MCP surface reached three data families that deliberately bypass the
Coach snapshot and so inherited none of its module narrowing: the
doctor-report aggregate behind healthlog://report/doctor-visit and the
doctor_visit_summary prompt, the rollup-backed rich reads (compare_metric,
get_metric_baseline, detect_changepoints), and the correlations reader.
A disabled module — or one an operator switched off server-wide — still
egressed to a third-party assistant.
Gate at the read rather than per call site. The module-to-data ownership
table moves out of the Coach snapshot builder into
lib/modules/measurement-scope so every read gates off one map; the rich
reads resolve metrics through a module-aware resolver; the correlations
reader gates on insights itself, which also closes the same gap on the
Coach chat tool and the metric-page read strip.
Per-domain reads omit with an honest module_disabled reason; the
whole-record aggregate refuses, mirroring /api/export/health-record.
collectDoctorReportData gains a fail-closed backstop on its own module
key. The clinician share link, found while tracing the aggregate's
callers, had the same gap and now collapses to its documents-only state.
Two layers. The integration suite runs a real pg-boss against a real
Postgres and proves the semantics the policies rest on: that a bare
singletonKey under the default policy de-duplicates nothing, that
exclusive suppresses a duplicate while the first job is queued AND while
it is active, that short suppresses only while queued and admits a send
once the job is active, and that a distinct key always enqueues.
It also pins the two facts that make the fix reach a running instance:
createQueue does not change the policy of a queue that already exists,
and updateQueue refuses a policy change, so the reconcile is required.
The unit guard couples each enqueue site to its registrar entry. The key
and the policy live in different files with nothing in the type system
holding them together, so deleting an entry would leave the call site
looking correct while it silently stopped working.
The projector, the reminder worker and the intake write path key on the slot
instant byte-for-byte, and all three derive it from a DIFFERENT reference —
a cron tick, a later cron tick, and the moment the user pressed take. That
is the whole hazard: the primitive used to sample the zone offset at its
reference, so on a transition day two ticks on opposite sides minted
instants an hour apart for the same slot and the dedupe key stopped
matching.
The existing DST cases compare the resolver to a projector mint computed
from the SAME reference, which cannot see this. Pass deliberately different
references across both Berlin transitions.
The suite was not timezone-pinned. CI runs UTC, so a whole class of
day-bucketing defect was invisible to the gate while a contributor outside
UTC saw a red suite on unmodified main. Verified: a UTC+14 host fails five
tests on an otherwise green tree. Pin TZ=UTC in both vitest configs; tests
that need a different HOST zone now set and restore it explicitly.
The resting-pulse fixtures passed offset-less ISO literals, which parse as
host-local, so they could not distinguish the Berlin day-key the
implementation uses from a UTC one — swapping the day-key's zone left the
file green. Build each instant explicitly in Berlin and add cases whose
local day and UTC day genuinely differ; that swap now fails two of them.
Delete the 'DST boundary (spring-forward)' compliance case. It passed no
timezone, so the slot builder took the legacy UTC path where the date
carries no transition, and it asserted totalExpected === 7 — which holds in
every zone and therefore could never see the double-emit its own comment
described. Threading a real zone does not rescue it: the assertion stays
true in Berlin, UTC, UTC+14 and UTC-11 alike. The property it claimed is
already pinned properly in the scheduling suite, which walks
buildCadenceTimeline with an explicit zone across both the 23 h
spring-forward and the 25 h fall-back day.
The card and the table carried byte-identical copies of the day-label
arithmetic. Two copies is how they drift, and neither was reachable from a
test, which is why the double conversion in both survived review.
Lift it into resolveNextDueDayLabel, which returns a discriminated label the
callers map onto their own strings, and cover the case the old code could
not get right: a browser zone that differs from the profile zone. Those
tests pin the host zone explicitly — the defect was invisible whenever host
and profile agree, which is the only configuration a UTC-only suite runs in.
Also pin that the window-status verdict is identical from every host zone,
including one whose spring-forward gap swallows the profile wall clock.
The per-connection backfills and per-user one-shots take 'exclusive': a
worker restarting during a heavy account's multi-hour pass used to append
another identical full-history job per restart, because the discovery
predicate only stops offering a job once it has completed.
The morning-digest refresh takes 'exclusive' so its user+localDate key
becomes a real at-most-once-per-morning contract; a wall-clock
singletonSeconds window cannot express a local date.
The per-document queues take 'short' -- they have no discovery pass to
re-converge, so only queued duplicates may be collapsed or a re-process
after a content change would be lost.
Leave the environment-fetch queue on 'standard' and say why: its
explicit-range backfill deliberately sends without a key, so any policy
would silently merge two different requested ranges into one.
The restore target came from the encrypted payload's own userId field
while the audit entry recorded the backup row's userId. A file declaring
a different account would therefore have been written into that account,
with the trail naming the one the operator selected.
Both values are in hand at that point, so compare them and refuse on
mismatch, recording both ids. Adds the first test coverage for this
route; the mismatch case fails without the guard.
Also correct the CLAUDE.md claim that every API route wraps apiHandler -
the three MCP OAuth routes deliberately use withBackgroundEvent so they
can return RFC-format error bodies, and carry the same protections by
hand.
chore(release): v1.30.20
Every queue was created under pg-boss's default 'standard' policy, under
which no partial unique index constrains singleton_key. The singletonKey
each enqueue site passed was therefore inert and de-duplicated nothing.
Add a per-registrar policy table that records the decision and the reason
per queue. The recompute queues take 'short' (collapse only while queued,
because the handler re-reads the bucket at run time); the per-user
backfills take 'exclusive' (queued or active, because a second concurrent
run is pure duplicated work and discovery re-enqueues while work remains).
createQueue only applies a policy on first provision, so also reconcile
the stored policy for queues that already exist -- without that the tables
would be correct on a fresh database and inert on every running instance.
toZonedDate rendered an instant as an en-US string in the target zone and
re-parsed it as host-local. Two defects rode on that.
When the fabricated wall clock landed in the HOST zone's spring-forward gap
the engine silently advanced it an hour, so a dose inside its window could
be classified late and a just-taken dose could fail to suppress the take-now
pill — for users whose device zone differs from their profile zone.
The medication card and table then passed that already-shifted value into
dateWithWeekdaySmart, which applies the profile zone itself, so the next-
intake day label converted twice and could name the wrong calendar day.
Invisible whenever browser zone and profile zone agree.
Migrate every consumer to wallClockInTz, which returns the fields directly.
The status math only ever reads year/month/day/hour/minute/weekday, so it
now runs on WallClockParts throughout and reduceCurrentWindowStatus takes
the real instant instead of a pre-shifted Date — caller and helper can no
longer disagree about which zone the comparison runs in. Day labels use a
calendar-day distance (localDayIndex) rather than an hour count, so a 23 h
or 25 h day cannot round tomorrow into today, and the formatter receives
the raw instant.
Fixtures name the zone they mean instead of building host-local Dates.
localHmAsUtc sampled the zone offset at the reference instant and applied
it to the target wall clock, so on a DST-transition day the answer depended
on which side of the transition the reference sat: a 20:00 slot resolved to
19:00 local when the caller ticked before the Berlin autumn change and
correctly when it ticked after.
That broke the slot-dedupe contract. The projector, the reminder worker and
the intake write path each pass their own reference instant for the same
slot and key on the resulting instant byte-for-byte, so on that day they
disagreed and a duplicate pending row became reachable.
Rebuild it on zonedWallClockToUtc: resolve the local calendar day from the
reference, then convert that day's hh:mm. The offset now settles at the
target local time, and because zonedWallClockToUtc runs the same two-pass
converge as startOfLocalDayInTz, localHmAsUtc(x, tz, 0, 0) equals
startOfLocalDayInTz(x, tz) by construction — the two used to disagree by an
hour on transition days, which opened the one-shot on-time band an hour into
the previous local day.
Cover both Berlin transitions from references on either side, plus a
southern-hemisphere zone whose transitions invert.
The inventory had a BUILDER-GATED bucket that accepted the mere presence
of an aggregator's name -- loadFhirContext, collectDoctorReportData -- as
proof a route was gated. That proved nothing. loadFhirContext filtered
per-domain modules but never evaluated doctorReport, so all five
/api/fhir data routes sat in a green bucket while serving the whole
record to an account that had the module off.
Drop the bucket. A route now counts as gated only when it calls a gate,
and the orphan message says plainly that importing a module-aware builder
is not one.
Add two assertions the old shape could not make. REQUIRED_TREE_MODULE
pins a tree to the key its routes must name, so /api/fhir cannot satisfy
the inventory by gating on some other, more permissive module.
BEHAVIOURAL_GATE_TESTS lists the tests that actually exercise a disabled
module end to end and fails if one goes missing, since this file reads
source text and can never prove a refusal itself.
Verified by removing the gate from the Patient route with the builder
import left in place: the old inventory passed, this one names the file.
Both consent gaps had the same shape: a module resolved a live provider and
nothing connected that to "check a receipt first". Behavioural tests cover
the two surfaces; this covers the class.
A module importing resolveProvider, resolveProviderChain or
resolveProviderForTest holds something it can call generateCompletion on, so
it must also import a consent helper — or sit on an allowlist with a written
reason. Four entries today, all of them probes or the credential connection
test, none of which send health data.
The guard is an import-level tripwire and its header says so: it cannot show
the gate is reached on every path, called before the egress, or passed the
right surface. What it does is stop a new ungated call site landing without a
reviewer seeing this file change.
The self-context clarifying questions and the AI-composed proactive nudge
both reached a provider without a consent check of any kind. Either could
resolve to the operator's server-managed credential (admin-openai or the
shared central Codex), so a user with no AI credentials and no receipt on
file had health data leave the server on the operator's key.
The questions path is the sharper of the two: its prompt carries the
complete Coach snapshot. Gate both with the chain predicate pair the status
path already uses, on the coach surface. The gate is skip-shaped on both —
the questions fall back to the deterministic hints, the nudge keeps its
deterministic template — so a missing receipt costs the user nothing but the
model's phrasing. BYOK, local and ChatGPT-OAuth chains stay ungated; that
egress is the user's own act.
Move the questions path off the read-then-write budget check onto the atomic
reserve/reconcile pair, so two concurrent saves can no longer both observe a
sub-cap spend and both call the provider. The cap now follows the chain's
cost owner instead of always charging the operator ceiling.
GET /api/dashboard/summary builds its own payload instead of going
through readDashboardSnapshotCached, so it never picked up the module
gate the shared snapshot builder applies. It emitted glucose and sleep
cards on data presence alone while /api/dashboard/snapshot correctly
withheld them for the same account.
Lift the drop decision out of gateSummariesByModules into
disabledSummaryTypes in the client-safe widget-modules module, so the two
dashboard aggregates decide from one map. The snapshot keys its summaries
by MeasurementType and the summary route emits cards keyed by an iOS
MetricKind, so what is shared is the decision, not the filtering;
SUMMARY_METRIC_TYPE_BY_KIND is the join between the two.
Omit rather than refuse, matching the snapshot sibling: the payload spans
ten cards across several domains, and blanking the dashboard because one
module is off would be the wrong trade. Core vitals always pass through.
The filter runs outside the analytics cache on purpose. That LRU is keyed
by user with a 60 s fresh / 1 h stale window and a module toggle does not
evict it, so gating inside the builder would keep serving a disabled tile
for up to an hour after the user turned it off.
Wire shape unchanged -- a gated card is absent, not null.
/api/fhir/* serves the same whole-record aggregate as
/api/export/health-record -- right down to the decrypted insurance number
on the Patient resource -- but never consulted the module gate. The
export refuses with a 403 module.disabled envelope when the account turns
the doctor-report module off; the FHIR face happily returned the full
Bundle to the same token.
Gate all five data routes (Patient, Observation, MedicationStatement,
MedicationAdministration, $everything). Refuse rather than omit: a
whole-record export has no truthful partial answer, and unlike the sync
delta feed no background client depends on it draining to stay
consistent. The refusal reuses the shared module.disabled envelope so the
errorCode clients already branch on stays the same.
loadFhirContext gains a fail-closed assertion as a second layer. It is
the one door to the aggregate, so it refuses outright instead of trusting
that every future caller remembers its gate; reaching it means a route is
missing one, which is a bug to surface rather than a flow to serve.
The static CapabilityStatement at /api/fhir/metadata stays open -- server
metadata, no user data.
The multi-domain feed at GET /api/sync/changes read cycleDayLog and
menstrualCycle without consulting the module gate, so an account with
cycle tracking off still received period days and per-day symptom rows
over the same token that every /api/cycle/* route refuses with a 403.
Resolve the cycle module once per request and skip both queries when it
is off, rather than filtering after the read. The blocks come back empty
and the request still succeeds: the feed multiplexes five domains over
one cursor, so refusing it outright would wedge measurement, mood and
intake catch-up for every account that has cycle tracking off. The two
cycle watermarks stay untouched while the module is off, so re-enabling
resumes where the client left off.
A client that synced cycle rows before the module was turned off keeps
its stale local mirror. Tombstones would be the wrong signal there --
they mean deleted, and disabling a module must leave the rows intact --
so purging a paired mirror needs an explicit signal in the feed contract
and is left for a coordinated contract cycle.
Wire shape unchanged.
Three leaks of the same class - data leaving the process unprotected.
addMeta and addExternalCall wrote their values verbatim while setError
and setHttp scrubbed, so the builder's redaction contract held for two
entry points out of four. The finished event is serialised whole to
stdout and the log store, which made meta a trusted sink by accident: a
Telegram request timeout put the bot token there, and the AI provider
chain fed upstream error bodies through it. Both now scrub, at any depth.
Scrubbing cannot help with the second half. A gateway that echoes the
request it rejected echoes the prompt, and on that path the prompt
carries the health snapshot; no pattern separates that from any other
text. Upstream bodies are therefore classified rather than quoted - the
diagnostic question is which rejection occurred and whether it repeats,
which a classification answers without the content.
The idempotency replay cache stored response bodies unencrypted. The
create paths echo their own decrypted payload, so cycle notes, mood text
and allergy reactions sat in cleartext for 24 hours in a column that
lands in every backup; the existing guard missed it because it matches
secret shapes and health data is not secret-shaped. Bodies are encrypted
now, and when encryption is unavailable the response is not cached at
all rather than written in the clear.
chore(release): v1.30.18
Four structural guards freeze the invariants the change rests on: the set of
files that can resolve a Bearer token to a user, the single deliberate
any-valid-token posture, the mint sites with their scope sets, and the rule
that a declared scope is an exported constant rather than a literal.
Thirteen behavioural cases run against a real Postgres and the real resolver.
Four of them go red on the pre-fix tree — a narrow token reading the
full-backup export and the labs surface, and an MCP token on a REST read.
The rest pin the flows that must NOT change: the per-medication ingest pair
still works, the token without the per-medication grant is still refused, a
wildcard token still reaches the export and the batch ingest, an MCP token
still resolves on /mcp with its write scope intact, a declared fhir:read
still opens the FHIR face and nothing else, cookie sessions are unaffected,
and requireAdmin still refuses a wildcard Bearer.
The workout-batch case that asserted narrow tokens are admitted claimed the
iOS sync depended on it. It does not: workouts:ingest has no mint site
anywhere, and iOS holds a ["*"] token. Replaced with a case that asserts both
halves of the real contract.
Documentation updated for the enforcement rule and the retired mint.
POST /api/tokens minted ["medication:ingest"], and that token could never do
the job the settings copy advertised: /api/ingest/medication gates on the
per-medication medication:<id>:ingest grant as well, which this endpoint never
issued. It was refused at the ingest surface and admitted everywhere else.
With Bearer scope now fail-closed the token is inert rather than dangerous,
which is a worse thing to hand a user than the bug it replaces. Remove the
mint. The working credential comes from the per-medication API-endpoint
toggle, which issues both grants scoped to one medication.
List and revoke stay so existing tokens remain visible and revocable, and the
settings card now points at the per-medication toggle instead of offering a
name field. Six locales updated.
resolveBearerToken took an optional permission, and an unset one admitted
any valid token. 324 of the 330 route files pass no permission, so a token
minted with a narrow scope reached the full-backup export, the labs and
measurement surfaces, the coach, and the bulk deletes — the settings copy
tells users to hand exactly that token to third-party automations.
Invert the default. The scope argument becomes a required ScopeRequirement
union: a route that names no scope now accepts cookie sessions and
cookie-equivalent ["*"] tokens only, and refuses a narrow token with 403 and
an undeclared_scope audit row. A route that means to accept a narrow token
names it, as the FHIR routes already do.
No route file changes. The absence of an argument becomes a positive
statement rather than an omission, so a newly added route denies a newly
minted narrow scope by default — the safe outcome is the one you get by
doing nothing.
/mcp keeps its own posture through an explicit any-valid-token requirement:
it authenticates here and authorises downstream on audience and write scope.
That is the single deliberate fail-open call site in the tree, and making the
parameter required is what keeps it greppable.
An MCP-audience token loses its REST read leg as a result, narrowing it to
/mcp only. The isMcpAudienceToken guard stays as defence in depth.
The dense-intraday fold tombstones the raw heart-rate readings a proxy
account's resting figure is derived from, and mints one derived
RESTING_HEART_RATE row per folded day to carry the figure forward. The
mint is skipped for accounts that already hold native resting rows,
because the read resolver ignores the proxy for them.
The probe behind that skip matched every source, including the COMPUTED
rows the mint itself writes. From the second run onward it therefore
reported native resting data for every proxy account, skipped the mint,
and let the fold delete that day's raw readings anyway - losing the
resting figure permanently, once per run.
Scope the probe to non-computed sources. Adds a regression test that
reproduces the second-run state and fails on the previous code.
chore(release): v1.30.16
getBaseSystemPrompt appended the directive, then each metric module
appended its own section after it, so the directive was no longer the
last instruction the model reads - the one property that makes it
authoritative. The base now exposes its body, and every composing module
wraps the finished prompt so the directive ends it.
Also drops a duplicate import that survived the merge of the two prompt
branches.
The deterministic fact-pattern pass fell back to German when no locale was
passed, the same de-default polarity the narrative surface just lost. Fact
text is model-facing memory, so English is the fallback for every locale
without a reviewed body; German stays for German readers.
Also correct the InsightNarrative.locale doc comment — the column holds any
shipped UI locale, not just de/en. Comment only, no SQL change.
The derived batch grid carries only the deterministic assessment, whose
templates ship de/en bodies. Its `locale === "de" ? "de" : "en"` was
correct but indistinguishable from the accidental binaries that were just
removed; route it through instructionLocale so the intent is explicit.
Behaviour unchanged.
Three layers. The German and English narrative prompts are frozen by content
hash, normalised for the prompt-version string that a prompt change is
required to move — any other drift fails. The four riding locales must name
their language, carry the verbatim reply-language directive from their own
safety-contract file as the last instruction, and contain no German-body
sentinel. A generator-level test drives a French run end to end and asserts
the row is keyed fr, the cache action is locale-scoped, and the system prompt
reaches the provider in English with the French directive.
Alongside it a source-level ban on the exact shape the bug took
(`=== "en" ? "en" : "de"`, `?? "de"`) across the coach and narrative
surfaces, with a self-check so a typo in the pattern cannot turn the guard
into a silent pass, and coach-memory recall cases for all four locales.
Covers the assertion that would have caught the narrowing: normalizeLocale
passes all six locales through and defaults an unknown value to English, the
de/en cache keys stay byte-identical while the four widened locales get their
own, and a fr locale entering a status generator reaches the prompt builder
as fr and writes the fr cache row. The forced-warm case pins the same for the
comprehensive and per-status warms.
Holds every sibling prompt module to one invariant: strip the directive
and the interpolated language name out of a French, Spanish, Italian or
Polish prompt and what remains is the English prompt, byte for byte. A
German instruction body leaking back into a non-German prompt, an
English body drifting away from what the four locales receive, and a
module appending a second directive on top of the one the base prompt
already carries all fail here.
Also drops the now-unused narrow locale alias from the two modules whose
prompt builders take the full locale.
The narrative read route narrowed every non-English locale to German, so a
French, Spanish, Italian or Polish account read the German narrative row and
warmed it from the German prompt body. Carry the resolved locale through
instead: the row is keyed (user, period, locale) across the full UI locale
union, and the four locales now compose the English instruction body with
their language named and the locale's own reply-language directive appended
last. The German and English prompts are unchanged apart from the version
string.
The nightly warm keeps the user's stored locale for the same reason — it has
to write the key the read route reads — and the coach and dashboard narrative
recalls read that same key. Coach memory stays de/en: it is model-facing
context, not user-facing prose, so the four locales compose and key English
rather than German.
NARRATIVE_PROMPT_VERSION 1.13.0 -> 1.14.0.
The per-metric assessment pipeline typed every locale as `de | en` and
narrowed on the way in, so a French, Spanish, Italian or Polish account
never reached a prompt with its own locale — the output-language directive
could not fire and the reader got English, or German where the binary
defaulted that way.
Widen `SupportedLocale` to the six locales the UI ships and replace the
narrowing binaries with one validating `normalizeLocale` that passes a
recognised locale through unchanged and defaults an unknown value to
English. The status-generate payload, status batch, status cache,
comprehensive generation, the pre-generate warm passes and the forced-warm
handlers all carry the full locale now.
Fix the collapse sites that discarded it: the warm enqueue on the
comprehensive route and the pregenerate route, the nightly batch cron, and
the forced-warm job handler — the last two defaulted to German outright.
The deterministic derived-score templates and the previous-context block
ship de/en bodies only, so they route through `instructionLocale` (German
for German readers, English otherwise) while the prompt names the reader's
own language.
The derived-score archetype and the Coach preference overrides branched
on the raw locale, so fr/es/it/pl composed German instruction text. Both
now route through `instructionLocale`. The score archetype rides on
`getBaseSystemPrompt` and therefore appends no directive of its own —
the base prompt already carries it.
The biomarker assessment is the one prompt in the family that does not
compose the base scaffold, so it appends the reader's directive itself;
the append collapses into the shared path once that prompt moves onto
the base scaffold.
The two prompt builders in each of those modules now take the full
`Locale` rather than the narrowed de/en alias, so the routing is right
the moment the pipeline stops collapsing the four locales on the way in.
Also targets the output-language contract test's regex at the compile
target instead of the dotAll flag, which does not compile.
German and English prompts are unchanged, byte for byte.
The per-metric system and user prompt builders branched
`locale === "en" ? EN : DE`, so a French, Spanish, Italian or Polish
reader composed the German instruction body — whose output clause asks
for German prose. Route them through `instructionLocale` instead, so
German is the only locale that takes the German body and everyone else
takes the English one plus the reader's own output-language directive
already carried by the base prompt.
The German and English prompts are unchanged, byte for byte.
The parity test joined raw fragments, which no longer matches now that
the English output clause carries a language token; it substitutes the
expected name from a literal map instead of from the module under test.
Adds the contract test: which body each locale composes, that its own
language is named, that its own translated directive is present and
last, that no German reaches a non-German prompt, and that no two
locales collapse to the same prompt. What it cannot show - whether the
prose reads well in the four languages - is stated in the file.
getBaseSystemPrompt collapsed every non-English locale to the German
instruction body, whose output clause asks for a German assessment. A
French, Spanish, Italian or Polish reader therefore received German
prose on any surface that reached this prompt.
Those four locales now compose the English instruction body with their
language named in the output clause, and the locale's own reply-language
directive - already carried in production by the native insights and
briefing prompts - appended last. German and English compose exactly as
before, byte for byte.
Adds the shared output-language helper the remaining prompt modules
adopt, replacing the scattered en-or-else-German binaries.
The voice rule states that commits are authored by the maintainer's git
identity alone. Editors and CLIs append co-author trailers and session
links by default, and once such a line is in the object store only a
history rewrite removes it — so enforce it before the commit exists.
The guard rejects tooling co-author trailers, session links, and
generated-with lines; human co-authors and ordinary messages pass.
Move the workouts list read into a shared cached module so the RSC
wrapper and GET /api/workouts share one deduplicated projection pass,
and dehydrate it under the client cell's exact query key so the page
paints its rows instead of a loading skeleton.
Correct the aggregated heart-rate externalId documentation: the example
and the min/max field notes still described the original hourly bucket,
while the accepted contract has been 10-minute buckets since v1.30.7.
Extend the dashboard's RSC server-prefetch pattern to three more route roots
so their above-the-fold content paints populated on first load instead of
flashing a shell that fills in on mount.
Each route gains a thin RSC wrapper that runs the same cached read its API
route uses, dehydrates the result under the exact client query key, and hands
it to the client leaf through a HydrationBoundary. Every wrapper fails soft to
today's client-fetch path on any error and honours the DASHBOARD_SSR_PREFETCH
kill-switch the e2e server sets.
- /insights: prefetch the dashboard snapshot — the same locale-keyed cell the
dashboard warms, resolved once server-side so the key matches by
construction. The AI briefing and the heavy analytics aggregations stay
client-lazy (the briefing must never trigger generation; a cold analytics
read would trade FCP for TTFB).
- /medications: prefetch the medications list through the shared
caches.medications cell (extracted to src/lib/medications/list-read.ts). The
list cell keeps refetchOnMount: "always" — a cross-device take/skip produces
no client mutation event, so the mount refetch is a correctness gate; the
prefetch removes the initial skeleton flash without dropping that guarantee.
- /coach: prefetch the nudge status (extracted to
src/lib/ai/coach/nudge-status.ts) so the auto-open-most-recent decision is
ready at hydrate, collapsing the one sequential waterfall. Only runs when the
Coach surface is reachable; the streaming conversation is never warmed.
Add per-route tests pinning server-key === client-key, JSON wire-shape
round-trip, and the fail-soft / module-gate paths. Repoint the source-reading
insights, take-all-due, and Coach gate-cascade tests at the client leaves.
Two additive indexes on medication_intake_events: (user_id, scheduled_for)
for the sibling reads that filter without medication_id, and
(user_id, medication_id, taken_at) for the per-medication last-taken
group-by that scanned the whole intake history on every dashboard load.
Additive migration 0253, no data change, no behaviour change.
Adversarial review found the transport gate (isCookielessNativeCaller) was
broader than finishLogin's bearer predicate: a blank-UA caller with no
X-Client-Type could pass it and fall into the cookie branch, receiving a
Set-Cookie session instead of the token bundle (inert, not exploitable, but
the endpoint's own invariant was false). Also require isNativeClientRequest
so the cookie path is structurally unreachable here, and report the actual
revoke outcome on a replay so the wide event never overclaims.
Unit-test the handoff library (mint stores only the hash + binds the
challenge + 90s TTL; consume covers not-found / replay-with-revoke /
expiry-burn / PKCE-burn / atomic consume / lost-race), the
token-exchange route (rate limit, transport gate, one identical generic
401 across every invalid-code class, success with issued-hash stamp +
audit, and OIDC_ONLY does not block it), the callback native branches
(handoff mint + scheme redirect with no session, MFA ticket on the
scheme with no session/cookie, post-identity + IdP errors to the
scheme, CSRF web-fallback), the login native-param handling, and the
by-hash refresh-token revoke used for replay containment.
The web OIDC flow ends in a session cookie the Bearer-transport native
client cannot consume, so on an OIDC_ONLY instance the iOS app had no
sign-in. Add a native leg: the app opens the login URL with
client=native and its own PKCE S256 challenge inside an in-app browser;
the callback resolves identity exactly as before, then mints a one-time
handoff code and 302s to the fixed compiled-in scheme
healthlog://oidc-callback (or hands back the existing MFA ticket when
the account has a second factor). The app exchanges the code plus its
PKCE verifier at POST /api/auth/oidc/native/token for the standard
native token bundle.
The return address is a compile-time constant — no redirect_uri
parameter and no env allowlist, so the open-redirect / scheme-hijack
class is removed by construction. The native flag and app challenge
travel only inside the AES-256-GCM state blob, so a network attacker or
the IdP cannot flip a web login into a native one. The handoff code is
256-bit CSPRNG, HMAC-hashed at rest, 90-second TTL, single-use via a
guarded consume; a failed PKCE check burns it. A replay of a consumed
code revokes exactly the pair the first exchange issued. The token pair
never rides a URL; only the opaque code or the MFA ticket do. The SSO
login can never satisfy requireFreshMfa step-up — the transport is
Bearer and step-up is cookie-only.
Adds the OidcNativeHandoff model (migration 0252), the token-exchange
route, a daily cleanup sweep, and the OpenAPI entry.
The correlation-narrative localisation replaced the env-channel narration
labels (which read through getEnvironmentField().narrationLabel) with the
localised insights.correlation.channel.ENV_* keys, leaving these two helpers
dead.
The detail surface at /insights/workouts/[id] gains a real heart-rate
curve, effort zones, an aspect-true GPS route, and per-kilometre splits
— all read from data already stored, no new integration, no migration.
- HR curve: one server path with strict provenance — stored
WorkoutSamples series first, a raw-PULSE window reconstruction second
(quality-gated so sparse data hides rather than lies), nothing third.
Adaptive bucketing keeps on-screen density constant across a 20-minute
HIIT and a 6-hour ride; a min/max envelope surfaces intervals.
- Effort zones: five %HRmax bands via the Tanaka HRmax, with a device's
own reported time-in-zone winning when present. Descriptive only.
- GPS route: self-rendered inline SVG with an equirectangular
cos(midLat) longitude correction (fixes the ~40% horizontal stretch),
aspect-clamped viewBox, RDP downsampling, start/end markers, and an
optional elevation profile. Nothing leaves the origin — no map tiles.
- Per-km splits and a GPX export, both derived client/server-side from
the geometry already on the payload.
- The endpoint extends additively (hrSeries, zones, splits, sportContext,
a reserved insight seam) behind an opt-in compact=1 param that drops
the raw sample/timestamp blobs for the web client; without the param
the response is unchanged.
- Every data-less section now hides instead of rendering an empty card;
the growing detail component is split into a workout-detail/ directory.
- List rows flag which sessions open into a rich detail.
The generated-insight narrative layer built its user-facing sentences by
string interpolation and never routed through the i18n bundle, so a
non-English UI rendered English text spliced into localised copy — e.g.
"Higher activity steps tends to go with lower next-day weight" surfacing
mid-German on the Weight insight. Because those strings bypassed t(), the
locale-integrity guard could not catch them.
Route every correlation narrative through the server translator:
- Correlation discovery (daily lagged pairs + labs-vs-outcome) now takes
the reader's locale, resolved from the correlations route the same way
other insight generators resolve it. Sentence templates, the direction
word, and every channel label live under a new insights.correlation.*
namespace, translated in de/en/es/fr/it/pl. The three fixed hypotheses
(BP-compliance, mood-pulse, weight-weekday) and their weekday labels are
localised the same way; the analytics cache key now includes the locale
so a language switch cannot serve another locale's cached text.
- Metric labels resolve to stable localised channel labels; the never-
causal framing ("not a cause" / "an association, never a cause") and the
faint/moderate/high hedging tiers are preserved in every language.
English output is byte-for-byte unchanged. Adds a regression test pinning
German narration and asserting it is not the English string.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018V8V9qapgHi7fhgtMJ8SkM
Legacy mode (snapshot flag off) keeps its browser-local batch window
unchanged — the profile-tz window + series prefetch run only in snapshot
mode where a byte-identical dehydration key exists.
Kill the above-the-fold first-load flash on `/`: the hero digest and the
batched chart series still fired client-side after JS download + hydrate,
so the hero and charts painted empty and filled in seconds later (LCP
~5.5 s mobile-throttled while the snapshot already dehydrated).
Extend the existing `page.tsx` snapshot prefetch to also seed the daily
digest and the chart series batch into the same dehydrated TanStack cache,
under keys byte-identical to the client's:
- Extract the series-batch route core into `readSeriesBatch(...)` so the
RSC and the route share one code path (no HTTP-to-self); the route thins
to auth + Zod + the shared reader, behaviour unchanged.
- New client-safe pure module `batch-chart-types.ts`: `deriveBatchChartTypes`
(the single implementation of the visible-with-data type-set derivation,
order + resting-HR-vs-pulse pick preserved) and `computeBatchWindow`
(profile-tz end-of-day minus the coverage window). Server and client run
the same function over the same snapshot payload, and the window is
computed once server-side and threaded to the client as a prop, so the
`chartSeriesBatch` key matches on both sides by construction.
- Render the hero data-first so the SSR HTML carries the populated hero
(the LCP element) — hydration-safe because the digest is dehydrated and
never restored from the IndexedDB persist allowlist; bound that invariant
with comments on both the hero gate and the allowlist.
- Warm the shared chart-runtime chunk during hydration.
Both prefetches honour the `DASHBOARD_SSR_PREFETCH=false` escape and fail
soft to the client-fetch path on any error; a ~400 ms soft budget bounds
the cold-day digest cost. No contract change, no visual change.
Two data-integrity gaps found in review of the v1.30.7 contract:
- The bucket externalId guard accepted a non-canonical timestamp that V8
silently rolls over (T24:00, 04-31, non-leap 02-29). Such a row would key
under a literal string the canonical client never emits, so the two would
never collapse on the overwrite key, and it would plot on the wrong local
day. Require canonical round-trip (toISOString() === suffix) in both the
guard and the bucket-start parser.
- The per-bucket min/max bypassed the plausibility-range guard that value
passes; now that they drive the DAY min/max band and the intraday envelope,
an out-of-range or mis-ordered spread poisoned the aggregate. Persist the
spread only when both extremes are in range and ordered around value;
otherwise drop to null and keep the average.
Refine the go-forward aggregated HR wire contract (iOS #34) from the never-
adopted hourly bucket to a 10-minute UTC bucket, so one contract serves both
the measurements-list de-explosion and the intraday-pulse 10-min curve.
- Batch route accepts stats:HKQuantityTypeIdentifierHeartRate:<10-min ISO-Z>
with avg + min/max, overwrite semantics, malformed-suffix rejection; the
validator is identifier-generic (allowlist) so respiratoryRate/SpO2 can join
later without a grammar change.
- The intraday reader classifies a day's PULSE rows three ways by externalId
shape (raw / uploaded 10-min bucket / server-folded hourly), builds the
10-min curve directly from uploaded buckets with the min/max envelope and
tension detection enabled, and overlays buckets over raw on the one
cutover-straddling day.
- DAY rollup min/max read COALESCE(value_min, value) so a bucket-native day's
band reflects the true intra-bucket extremes.
No schema change (value_min/value_max exist since v1.19.2). Reconciliation is
go-forward cutover; the nightly retention fold handles the raw backlog.
The target-summary and coach-read strip tiles sit directly under a
metric sub-page's h1, alongside the stat strip that already renders its
title as an h2. Render their titles as real h2 headings too via the
existing titleAs prop, so heading navigation lands on each section. A
test pins that the dashboard exposes its page-level h1 through the
shared page header (no duplicate sr-only heading needed).
The mood and medication-compliance heatmaps exposed per-day values only
through pointer tooltips. The cells now carry a visually-hidden day list
beside the grid — one entry per day that holds a value — so a screen
reader reaches the same figures without a tab stop per cell. A cell
subtree nested under the SVG's aggregate role=img is pruned from the
accessibility tree anyway, and a year-long grid would otherwise flood
the keyboard order; the rects stay a pure pointer affordance and the
aggregate summary label is unchanged. Tests pin the list and that the
cells carry no tab stops.
Bump text-[10px]/text-[11px] to text-xs on muted description, hint, and
status-label surfaces that had drifted below the type-scale floor —
coach conversation captions, settings/admin status badges, mood
confidence pills, side-effect and cycle meta lines. Tabular-nums
badges, chart-plot legend/axis chrome, ring geometry, and
width-constrained tag tiles keep their compact size by design.
Repoint EcgCrossLink from the overview `#ecg` fragment to `/insights/ecg`.
The fragment is absent until the overview section's own fetch resolves, so
the deep link no-oped on navigation (UX-flows finding F1-1); the routed
sub-page always mounts.
ECG recordings had no pill in the tab strip, no entry in the sub-page
manager, and no routed page — the only inbound pointer was a cross-link
card on the spot-pulse page. Hiding the overview ECG teaser left no route
back to a user's recordings.
- Add a routed `/insights/ecg` sub-page that reuses the existing ECG
section (heading suppressed so the page shell owns the title); the
overview teaser stays.
- Add an ECG entry to the Heart group of the tab strip, gated on a
dedicated `hasEcgRecordings` signal since ECG carries no measurement
type. The Heart group surfaces for an ECG-only device too. No ECG pill
appears without recordings.
- Mount the ECG cross-link on the resting-pulse and HRV pages, matching
its placement on the pulse page.
- List ECG in the sub-page manager under Heart, its eye toggling the ECG
overview section so hiding it is reversible.
- Point the metric catalog's ECG row at the new page.
- Add `navEcg`, the sub-page description, and the empty-state copy across
all locales.
The webhook-driven per-resource sync advanced the resource cursor but never
touched IntegrationStatus, so lastSuccessAt stayed frozen at the last full
poll — the Settings 'last synced' pill and the integration-status read
reported stale for up to an hour after a webhook had already landed last
night's data. Stamp recordSyncSuccess per resource, reusing the full sync's
exact success guard (a genuine import clears the error state; a 403 soft-skip
leaves the looks-healthy window closed).
The signal registry's `mcp:false` flag vetoes a metric-status resolution
so an unreviewed non-exposed signal cannot leak through the id/display-name
path. The metric-status discovery allowlist is a deliberate, human-named
exposure — co-equal to `mcp:true` — so its ids must stay resolvable, or
compare_metric / get_metric_baseline would advertise a metric in discovery
that then refuses to resolve. Exempt the allowlist from the veto; the guard
still fires for genuinely-unreviewed mcp:false ids (workout HR stats).
Wrist temperature, cardio recovery, sleep score, breathing
disturbances, strain/load, and the rest of the v1.10/v1.18.1
metric-status-registry ids were already resolvable through
compare_metric / get_metric_baseline / detect_changepoints (an exact
registry-id or display-name match), but sat off the Coach snapshot
inventory entirely — so list_metrics, the measurements-inventory
resource, and search never advertised them. A discover-before-fetch
assistant, which the server's own instructions direct, could never
learn they existed.
Adds a fixed, reviewed allowlist (MCP_METRIC_STATUS_DISCOVERY) plus a
metricStatusDiscoveryRows() presence probe, mirroring the existing
MCP_CLINICAL_SIGNALS pattern: one grouped presence query, appended as
supplement rows to list_metrics and the inventory resource, and merged
into the same probe search already runs for the v1.25 clinical
signals. fetch(id) now hydrates these ids via the baseline read the
same way it already does for clinical signals.
Adds MCP coverage for ECG recording metadata (recorded time, duration,
sampling rate, sample count, average heart rate, lead, and the
device's own rhythm classification) — previously unreachable because
ECG has no MeasurementType and so cannot appear in list_metrics or the
metric resource templates.
get_ecg_recordings mirrors GET /api/insights/ecg exactly, down to the
select clause: waveformEncrypted is never selected, so a decrypt can
never happen on this path. Non-diagnostic framing carries through:
classificationSource is a fixed "device" literal, and the description
states plainly that HealthLog never re-classifies. Gated on both the
insights module and the insightStatus assistant-surface flag, mirroring
the route's own two-layer gate.
Adds MCP coverage for the intraday pulse / "shape of the day" —
previously unreachable, since get_metric_series only serves daily
aggregates and cannot answer "what happened this afternoon?" or "was I
tense during the meeting?".
get_intraday_pulse is a thin wrapper over loadIntradayPulse, returning
the same DTO GET /api/insights/pulse/intraday serves verbatim: the
10-minute mean shape (dense-90d, hourly-grain fallback outside that
window per the resolution field), the resting baseline, and the
cautious elevated-at-rest tension window when every confidence gate
holds. Gated on the insights module; never fabricates a reading for a
day with no data.
Adds MCP coverage for the NutrientIntakeDay pipeline (water, caffeine,
24 micronutrients) — previously fully unreachable over MCP, and
search("water") / search("vitamin") silently miscollided or read as
honest absence even for a daily logger.
get_nutrients is a thin facade over the same two reads GET
/api/nutrients and GET /api/nutrients/daily run: the tz-anchored
day-key math, the sum-across-sources fold, and the sex-aware EFSA
reference resolution. Gated on the opt-in nutrients module, mirroring
both backing routes. Discoverable via search (a grouped presence
probe) and fetch, plus a healthlog://nutrient/{code} resource template
with completion + listing scoped to the user's own logged codes.
The Coach snapshot read raw prisma.workout rows while GET /api/workouts
runs pickCanonicalWorkoutRows, so a dual-source user (e.g. Apple Watch +
Withings recording the same run) got the session double-counted in the
Coach prompt and the get_workouts MCP tool. Route the snapshot's workout
block through the same picker, keyed off the user's own sourcePriorityJson.
HRV resolves SDNN only on every MCP/coach path, so an Oura/Polar/WHOOP-only
account (RMSSD, not SDNN) saw a false empty HRV block even though the app's
HRV sub-page already unions both flavours. The coach snapshot now queries
both types; the MCP rich reads (compare_metric, get_metric_baseline,
detect_changepoints) fall back to RMSSD when the user has zero SDNN rows,
mirroring the app's primary-then-fallback rule.
Also aligns surfaces.mcp as a real gate: a metric-status id the signal
registry marks mcp:false (cardio recovery, wrist temperature, sleep score,
...) no longer resolves through the rich-read registry path, closing a
contract gap between the two registries (no data was actually exposed
before this, but a future mcp:false signal without this guard would have
leaked silently).
The remaining day-edge tz mismatch between the rollup-backed rich reads
and the user-tz coach timelines is left documented rather than patched:
the underlying rollup buckets are themselves written on UTC calendar
boundaries, so shifting only the read-side window would trade one
one-sided inconsistency for another without fixing the root cause.
Visceral-fat explainer was missing its unitless-scale caveat in every
non-English locale, so those readers got the heart-disease/diabetes framing
without the "not interchangeable with imaging thresholds" caveat. Translate
the full sentence into de/es/fr/it/pl, and mirror the study citations
(fat mass, visceral fat, daylight, falls) that had only been added to the
English strings.
Condition the "automatic screening" watchdog copy on a source actually being
connected — silence with nothing connected means nothing is being watched,
not that everything looks fine — across all six locales.
Disclose in the full-backup scope note that document metadata and any
AI-generated summary are included in the plaintext export; only the
original files and workout GPS/sample data are excluded.
Metric catalog page painted every metric as "Not tracked yet" (with
Connect CTAs) while its four availability probes were still in flight,
and stayed there forever on a failed batch — a load failure indistinguishable
from an honest absence. Gate rows on the probes' aggregate loading/error
state: row skeletons while loading, a QueryErrorCard with retry on failure.
Repoint ComplianceInfoTip onto InfoPopover instead of carrying a verbatim
structural copy (adds icon/align/data-slot override props to the shared
primitive); migrate the cycle log-sheet's FieldInfo off a hover Tooltip
onto the same click-opened InfoPopover, since a hover trigger never opens
reliably on the sheet's thumb-first touch surface.
Lift ViewToggle/SegmentedToggle off their stale 40px tap floor onto the
44px/36px floor the sibling intake and history-rail controls already
carry, and correct the docstrings to match. Restore the dropped
safe-area-inset-top term in /coach/conversations' height calc so it
matches /coach on notched devices. Bring the coach history rail's
off-scale spacing and sub-floor timestamp text onto the standard scale.
POST /api/nutrients/batch performed zero cache invalidation, so an
Apple-Health-synced water/nutrient total kept the dashboard's tile
serving the pre-sync figure until the analytics fresh TTL lapsed
(~3 min) — the manual water POST already hard-evicts on its own write.
Calls invalidateUserDashboardSnapshot(user.id) once at least one row
inserted or updated, mirroring the manual route's own posture so a
synced total reflects on the very next dashboard read.
computeBpInTargetPct's same-calendar-day pairing fallback (for readings
whose sys/dia timestamps drift past the 5-minute session window) keyed
on toBerlinDayKey regardless of the caller's actual user. A non-Berlin
user's two same-local-evening readings straddling Berlin midnight (but
not their own midnight) were rejected as a pair, running the in-target
percentage on fewer readings than it should have — precisely the
population the cross-tz runtime guard in bp-in-target-fast-path.ts
force-routes to this same pairing fallback to protect.
Threads an optional tz through computeBpInTargetPct / collectBpPairs /
computeBpInTargetWindows (default Berlin for untouched legacy
callers), wires it through the fast-path's live fallback and the
comprehensive-insights route's own computeBpInTargetPct call — the
other caller with the user's tz already in scope.
Investigated but left the comprehensive route's resting-pulse-proxy
day-key and BP-medication-continuity day-key on their current UTC
convention: both are joined by an exact YYYY-MM-DD key against a
UTC-anchored partner (the mood rollup's bucketStart, and the
aggregator's date_trunc'd BLOOD_PRESSURE_SYS daily series). Switching
only one side to the user's tz would trade "evening readings shift a
day" for "evening readings mismatch their pairing partner" — a
different bug, not a fix. Documented in place; a real fix needs the
mood-rollup / aggregate tiers themselves to bucket per-user-tz.
The raw-row read backing the day-collapsed list is desc + take(limit).
On a dense account whose in-window row count crosses the cap, the
oldest returned row sits mid-day — every earlier sample on that same
calendar day was cut off, so its bucket's SUM silently understates the
real day total with no marker.
Flags exactly that one bucket partial: true when the cap was actually
hit, so a page that came in under the cap (the whole window covered)
carries no flag on any bucket.
The eleven per-metric/derived-score status helpers (pulse, weight,
bmi, mood, blood-pressure, medication-compliance, general, the generic
metric card, biomarker, the shared feature cache, the derived-score AI
warm layer) computed todayKey from toBerlinDayKey(new Date()) before
resolving the user's own timezone — several fetched the profile only
for downstream day-bucketing, after the cache read had already run on
the Berlin key. A non-Berlin user's card regenerated at their local
evening (Berlin midnight) and, for biomarker series and the
derived-score prompt label, mis-keyed data or displayed a hardcoded
"(Europe/Berlin)" regardless of their actual zone.
Threads resolveUserTimezone()/userDayKey() ahead of every cache read
so the day-key and the data bucketing agree, and reuses the one
resolved value everywhere downstream instead of re-deriving it.
Before the source-collapse fix, a day with both an Apple Health AND a
Withings write for the same cumulative metric (steps, active energy,
flights, distance, daylight, falls) summed both sources into one
inflated total and locked it in as the all-time PersonalRecord. The
worker's improvement gate only ever compares a freshly-computed day
against the stored best, so the false record silently outranked every
honest day forever and the affected metric could never set a genuine
record again.
Adds a boot-time one-shot repair (queue: cumulative-pr-rederive),
mirroring the existing step-consolidation-repair pattern: deletes the
suspect measurement-driven records created before the fix landed, then
re-runs detection with silent:true so the honest re-derived best
replaces it without firing a "new record" push for what is data
hygiene, not an achievement. Self-converging by a fixed cutoff instant
rather than a marker column or reused SQL migration number.
The asc+take(20000) window-cap bug (fixed on the correlations route
alone) and the mixed-grain toDailyMeans bug (fixed on three of four
readers) both survived on the Coach get_correlations tool, the
per-metric assessment card, and the period narrative. Dense accounts
saw the cap drop the newest rows on the coach/card readers and drop
the CURRENT period on the narrative — backwards for a recent-window
and a current-vs-prior surface respectively; the narrative's own local
toDailyMeans twin also averaged sleep stage-segments and blended
per-sample step chunks with drained daily totals instead of summing.
Hoists the fetch + grain resolution into
fetchMeasurementWindowSeries/fetchMoodWindowSeries
(correlation-channel-series.ts) — desc+cap+resort, full column select
for buildMeasurementDailySeries's per-type grain — and routes the
route, the coach tool, the per-metric card, and the period narrative
through it, deleting every local copy.
2026-07-18 06:47:32 +02:00
2105 changed files with 220772 additions and 44773 deletions
# Opt-in for a plain-HTTP IP_GEO_LOOKUP_URL provider (e.g. free ip-api.com); off by default — looked-up IPs then travel unencrypted over your server's egress.
# IP_GEO_ALLOW_INSECURE="true"
#
# Directory the offline GeoLite2 databases are read from; default /opt/geolite2 (the path the image bakes its own copy into). Point it at a directory you mount yourself to bring your own MMDBs — see docs/self-hosting/geolite2.md.
if [ "$GITHUB_EVENT_NAME" = "pull_request" ]; then
echo "::notice::Pull-request image validation does not publish release tags."
exit 0
fi
if [ "$GITHUB_REF_TYPE" != "tag" ] || [[ ! "$GITHUB_REF_NAME" =~ ^v(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)\.(0|[1-9][0-9]*)$ ]]; then
echo "::error::Release ref '$GITHUB_REF_NAME' is not a strict SemVer tag (expected vMAJOR.MINOR.PATCH with no leading zeroes or prerelease suffix)." >&2
echo "::warning::COOLIFY_AUTO_DEPLOY is not 'on' or 'off' — skipping auto-deploy. Set an explicit mode in repository variables."
exit 0
fi
if [ -z "$COOLIFY_WEBHOOK" ] || [ -z "$COOLIFY_TOKEN" ]; then
if [ "$mode" = "on" ]; then
echo "::error::COOLIFY_AUTO_DEPLOY=on but COOLIFY_WEBHOOK or COOLIFY_TOKEN secret is missing. Set both under repo Settings → Secrets and variables → Actions, or set COOLIFY_AUTO_DEPLOY=off to silence."
exit 1
fi
echo "::warning::COOLIFY_WEBHOOK or COOLIFY_TOKEN secret missing — skipping auto-deploy. Configure in repo Settings → Secrets to enable, or set the COOLIFY_AUTO_DEPLOY variable to 'off' to silence this warning."
exit 0
echo "::error::COOLIFY_AUTO_DEPLOY=on but COOLIFY_WEBHOOK or COOLIFY_TOKEN secret is missing. Set both credentials or disable promotion explicitly." >&2
exit 1
fi
# v1.4.31 — wait for GHCR cache propagation before firing
# the Coolify webhook. Five releases in a row (v1.4.27
# through v1.4.30.1) the webhook reported "finished" but
# Coolify still pulled the prior `:latest` digest. The
if [ "$status" -lt 200 ] || [ "$status" -ge 300 ]; then
echo "::error::Coolify deploy webhook returned HTTP $status — image is published, but auto-deploy did not fire. Maintainer: deploy manually on the deployment host (SSH force-pull recipe, see docs/ops/deploy.md)."
echo "::error::Coolify deploy webhook returned HTTP $status for verified image ${IMAGE_TAG}@${IMAGE_DIGEST}." >&2
exit 1
fi
echo "::notice::Coolify auto-deploy webhook fired at $(date -u +%FT%TZ) for ref ${{ github.ref_name }} (sha ${GITHUB_SHA::7}). If /api/version doesn't reach the new tag within ~60s, re-trigger the deploy from the Coolify UI or deploy manually on the deployment host (docs/ops/deploy.md)."
echo "Coolify deploy queued — awaiting outgoing notification webhook to /api/internal/deploy-webhook for final status."
echo "Coolify deploy queued for verified image ${IMAGE_TAG}@${IMAGE_DIGEST}."
- **Timezone**: `Europe/Berlin` for display, UTC in database.
- **Encryption**: Sensitive data (Withings tokens, API keys, VAPID private keys) encrypted with AES-256-GCM via `src/lib/crypto.ts`.
- **Dracula colors**: Use CSS variables `var(--dracula-purple)`, `var(--dracula-cyan)`, `var(--dracula-green)`, `var(--dracula-orange)`, `var(--dracula-pink)`, `var(--dracula-red)`, `var(--dracula-yellow)`, `var(--dracula-fg)`, `var(--dracula-comment)` for chart/graph elements.
- **Proxy route protection**: `src/proxy.ts` checks `healthlog_session` cookie on all non-public paths. Unauthenticated page requests → redirect `/auth/login`. API routes keep their own `getSession()` auth. Public paths defined in `PUBLIC_PATHS` array.
- **PDF generation**: Client-side via `src/lib/doctor-report-pdf.ts` using jsPDF. Data fetched from `POST /api/doctor-report`. European medical format with German terminology.
- **Achievements**: Persistent in `UserAchievement` table. API at `/api/gamification/achievements` computes current state, compares with DB, persists new unlocks. `unlockedAt` timestamp is stable once set.
## Critical Gotchas
These are hard-won lessons. Ignoring them will cause errors:
### Prisma 7
- **No `url` in schema.prisma** — The database URL is configured in `prisma.config.ts`, not in the `datasource` block.
- **Import path**: `import { ... } from "@/generated/prisma/client"` (with `/client` suffix, not `@/generated/prisma`).
- **Adapter required**: PrismaClient needs `PrismaPg` adapter from `@prisma/adapter-pg`. See `src/lib/db.ts`.
- **No `earlyAccess`** in defineConfig — causes TypeScript errors.
### Next.js 16
- **Proxy, not middleware**: Next.js 16 renamed middleware to proxy. The file is `src/proxy.ts` (not `middleware.ts`). Having both causes a build error: "Both middleware file and proxy file detected".
### Libraries
- **SimpleWebAuthn v13**: No `@simplewebauthn/server/script/deps` — define Transport type inline.
- **zxcvbn-typescript**: Default export only. `zxcvbnAsync` and `zxcvbnOptions` do not exist as named exports.
- **pg-boss v12**: `{ PgBoss }` named import (not default), use `localConcurrency` (not `teamSize`), handler receives `Job<T>[]` array.
- **Zod v4**: Import from `"zod/v4"`, not `"zod"`.
- **jsPDF**: Client-side only. Import dynamically in browser context. Used with `jspdf-autotable` plugin.
### Settings & Admin (per-route layout, v1.4)
- **Settings**: 8 routes under `/settings/[section]` — `account`, `integrations`, `notifications`, `dashboard`, `ai`, `api`, `advanced`, `about`. The legacy `/settings` 308-redirects to `/settings/account`. Sidebar deep-links and the `<a href="/settings#anchor">` patterns from 1.3 still resolve via the redirect.
- **Admin**: `src/app/admin/page.tsx` is now a 77-LOC shell that mounts the per-section components in `src/components/admin/`. Status-card grid lives in `status-card-grid.tsx`. The aggregator endpoint `/api/admin/status-overview` returns the six-card summaries in one batched query.
- ESLint `react-hooks/set-state-in-effect` is **strict** now (was non-blocking when the settings monolith carried inline-effect state-setters). Use lazy `useState(() => …)` for localStorage reads, TanStack Query for data fetches.
### Insights Page
- Large file (~1750 lines), 7 content sections with AI summaries.
- Sticky horizontal chip navigation (`InsightsSectionNav` component) with IntersectionObserver for active state tracking.
- Returned in `/api/auth/me` response as `gravatarUrl` field. Used by `AvatarImage` component with `AvatarFallback` for initials.
### Notification System
- **Channels**: `telegram`, `ntfy`, `web_push` — stored as plain `String` on `NotificationChannel.type` (no Prisma enum; the canonical list lives in `src/lib/notifications/types.ts`).
- **Event types**: `medication_reminder`, `measurement_anomaly`, `compliance_low`, `withings_sync_failed`, `system_alert` — also `String` on `NotificationPreference.eventType`. Source of truth is the discriminated union in `src/lib/notifications/types.ts`.
- **Opt-out model**: Preferences default to ON (enabled) when no `NotificationPreference` row exists.
Working notes for anyone (human or otherwise) editing this repository. Skim before the first PR; refer back when a convention question comes up. Pair with `README.md` (the public face) and `CHANGELOG.md` (the release history); this file is the maintainer's operating manual.
## What HealthLog is
Self-hosted personal-health-tracking PWA: weight, blood pressure, pulse, body composition, glucose, sleep, mood, medication compliance. Withings + Apple Health sync, multi-provider AI Insights (BYOK or local), doctor-report PDF, native iOS client in public beta. Single `docker compose up`, Postgres-backed, AES-256-GCM at rest. PolyForm Noncommercial 1.0.0 (AGPL-3.0 through v1.15.18). Current line: v1.20.x.
## Voice and privacy
Hard rules for every committed artefact — commit messages, CHANGELOG entries, PR bodies, code comments, GitHub release notes, docs site copy. Treat these as the project's public face.
- **Voice.** English. Maintainer prose: present-tense, imperative for commits, factual for prose. No marketing tone, no apology tone, no "wow what a cool feature" framing. The voice is one human writing notes for the next.
- **No assistant / tooling references.** Do not surface AI, agent, assistant, copilot, marathon, phase, wave, round, or session vocabulary into anything that lands on GitHub — including commit messages and trailers. Commits carry no `Co-Authored-By` tooling trailer; they are authored by the maintainer's git identity alone.
- **No personal data.** No real names (the maintainer's, a reporter's, or anyone else's). No live health figures, no live tenant identifiers, no real-account anchors. Generalise to "a self-hoster", "a user", "an operator", "an authenticated session", etc. The internal `.planning/*` files are gitignored and exempt; everywhere else in the repo, no.
If a draft commit message or release note would violate one of the three, rewrite it before committing. The `git log` is more durable than today's task.
## Branch and release model
`main` is the single trunk and is always releasable. There is no long-lived `develop` branch — the former one was retired and archived as the tag `archive/develop-pre-v1.5` (it carried no work absent from `main`; a long-lived integration branch has no role in a single-maintainer shop and only accrued back-merge debt). This is trunk-based development with release tags: short-lived branches off `main`, work always converges back onto `main`, and the immutable `vX.Y.Z` image built from a tag on `main` is the one source of truth — including for the OpenAPI contract the iOS client consumes.
- **Trunk.** `main` always builds, passes CI, and is deployable. Every commit on it is a release commit (`chore(release): vX.Y.Z`), a release-branch merge, or a maintenance commit (docs, gitignore, audit findings) applied directly.
- **A release** is built on a short-lived `release/vX.Y.Z` branch cut off `main` — built wave-by-wave, then opened as a PR `release/vX.Y.Z` → `main` with green CI, merged (merge-commit), and tagged `vX.Y.Z`. The tag push triggers `docker-publish.yml` (GHCR multi-arch + `:latest`). Deploy is manual (see `docs/ops/deploy.md`); after a deploy, verify `/api/version` (it returns `version` + `buildSha` + `builtAt`) on every target, since the `:latest` pull is the source of truth, not the queued deploy status. Delete the release branch after merge.
- **Small standalone changes** (a single fix, a docs touch) can go through a short-lived `fix/X` / `feat/X` branch + PR, or directly onto `main` when the maintainer chooses — `main` is the target either way.
- **Hotfixes** branch from `main` and merge straight back to `main`. There is no second long-lived line to forward-port to, so the old back-merge debt cannot recur by construction.
- Never force-push `main` or any release branch. Never skip hooks (`--no-verify`). Never `--no-gpg-sign` unless explicitly asked.
## Semver
Prefer patch bumps. A release dominated by bug fixes plus additive features without breaking changes ships as `vX.Y.Z+1`, not `vX.Y+1.0`. Bumping minor is reserved for breaking changes, removed flags, deprecated routes, or multi-feature milestones framed as such (e.g. v1.5 = native iOS client + Apple Health integration). Pre-existing example: a `v1.5.0` cut was rerouted to `v1.4.14` mid-flight when the diff turned out to be bugfix-dominated.
| Forms / validation | `react-hook-form` 7 + `zod` 4 + `@hookform/resolvers` + `zod-openapi` 5 | The Zod registry is the source of truth for `docs/api/openapi.yaml`. |
| Container | Multi-stage Alpine Dockerfile, GHCR multi-arch (`amd64` + `arm64`) | `pull_policy: always` in `docker-compose.yml` is load-bearing — without it Docker re-uses the stale `:latest` digest. |
More detail in `.planning/codebase/tech.md`, generated as part of the v1.5.2 codebase audit.
## How a request flows
```
client
│ HTTP
▼
src/proxy.ts renamed from middleware.ts. Public-path allowlist,
The architecture map in `.planning/codebase/arch.md` walks each layer with file:line citations.
## Code conventions
- **UI work follows the design standards.** `.planning/audits/2026-07-05-fable/UI-STANDARDS.md` is the binding design rulebook (card anatomy via TileHeader, spacing scale, text-colour rules — foreground for content, muted for meta only —, token discipline, primitive mapping). Read it before touching any surface; deviations need a written reason in the PR. Where the standards conflict with older in-file comments, the standards win.
- **RSC by default.** Server components are the default; add `"use client"` only when a hook, state, or browser API actually needs it.
- **Every API route wraps in `apiHandler`.** Verified — zero exceptions in the current tree. Every body-accepting route runs Zod `safeParse` and returns 422 via `returnAllZodIssues` (multi-issue envelope sanitised against echoed input through `sanitiseZodIssues`).
- **`userId` is always narrowed from session or Bearer.** No route accepts `userId` as a body field — it comes from `requireAuth()` and feeds the Prisma `where`. The batch endpoints (`/api/measurements/batch`, `/api/mood-entries/bulk`, `/api/medications/intake/bulk`, `/api/workouts/batch`) don't even declare a `userId` field in their Zod schemas.
- **No mass assignment.** Every `prisma.X.{create,update}({ data: ... })` builds its `data` object field-by-field from `parsed.data`, never by spreading the parsed object whole. `tokens.create` hardcodes the narrow scope so the user-facing endpoint can never mint a wildcard token.
- **`requireAdmin()` is cookie-only.** Bearer tokens — even with `["*"]` scope — cannot reach admin endpoints. This is a structural boundary in `requireAdmin()` (`src/lib/api-handler.ts`), not a runtime check that future code can soften.
- **Raw SQL is parameter-bound or whitelist-spliced.** Every `$queryRaw` uses tagged-template parameters. Every `$queryRawUnsafe` / `$executeRawUnsafe` either passes positional `$N` placeholders with an argument array OR splices a value already asserted against a closed enum + regex (see the whitelist-splice block in `src/lib/rollups/measurement-rollups.ts`, around the `$queryRawUnsafe` calls). Inline comments document the whitelist at the splice point.
- **TanStack Query keys live in the centralised factory.** `src/lib/query-keys/` (16 modules, `index.ts` barrel) is the only legal source of `queryKey` / `mutationKey` arrays. The in-repo ESLint rule `healthlog/queryKey-factory` (set to `error`) flags any bare array. Same-key + different `queryFn` shape silently poisons the cache; the factory prevents it. Every read unwraps `(await res.json()).data` from the envelope.
- **`annotate()` on every interesting code path.** Wide-event observability. Action names follow `<surface>.<noun>.<verb>` (`coach.budget.exceeded`, `measurement.batch.ingest`). Don't free-text in `meta` — pin the shape so dashboards stay stable.
- **i18n keys exist for every `t()` call.** `i18n-call-site-coverage.test.ts` walks every `.ts(x)` under `src/`, extracts every `t("ns.key")` literal, and asserts each key resolves in `messages/en.json`. `i18n-locale-integrity.test.ts` propagates the EN guarantee across `de / en / es / fr / it / pl`. If a guard fails, fix the bundle, don't suppress the test.
- **OpenAPI registry stays in sync.** Zod schemas carry `.meta()` annotations under `src/lib/openapi/`; `pnpm openapi:generate` emits `docs/api/openapi.yaml`. CI fails on drift. Re-run the generator after touching a request / response schema and commit the YAML alongside the Zod change.
-`pnpm typecheck` — `tsc --noEmit`. Required before every commit.
-`pnpm lint` — ESLint flat config with `eslint-config-next` + the project's `healthlog/queryKey-factory` rule. One harmless warning in `withings/resume/__tests__/route.test.ts` is allowed; everything else should be clean.
-`pnpm test` — Vitest 4 unit + component suite. Required before pushing a fix; full suite is fast (~30 s).
-`pnpm test:integration` — Vitest against a `testcontainers/postgresql` Postgres 16. Needs Docker / OrbStack running. Slower but pins the contracts the unit-mocks can't.
-`pnpm e2e` — Playwright against `pnpm exec next build` + Postgres. Heavy; CI runs it, locally only when touching UI flows.
-`pnpm openapi:generate` — regenerates `docs/api/openapi.yaml` from the Zod registry. Run after any request / response schema change. CI's `openapi:check` fails the build on drift.
-`pnpm check-env` — validates that `.env.production.example` covers every required variable. Wired into CI as `env-check.yml`.
-`pnpm bundle-report` / `ANALYZE=1 pnpm build` — bundle inspection. Use when a feature adds material weight.
- One-shot scripts live in `scripts/`. Run them via `pnpm dlx tsx scripts/<file>.ts` — the production standalone image strips `tsx` and the bare `pnpm tsx ...` invocation fails inside the container. Recurring tasks belong on pg-boss, not on a CLI script.
| Session | `src/lib/auth/session.ts` + `src/lib/auth/secure-cookie.ts` | Postgres-backed sessions, sliding 30-day expiry, `shouldEmitSecureCookie()` the one source of truth for the `Secure` flag (every cookie setter routes through it). |
| Refresh tokens | `src/lib/auth/refresh-token.ts` | Per-device one-time-use rotation with reuse-detection that revokes the device's token family. |
| Bearer tokens | `src/lib/auth/hmac.ts` + `src/lib/auth/issue-token.ts` | HMAC-SHA256 hashing under `API_TOKEN_HMAC_KEY`; no plaintext path; `lastUsedAt` updated fire-and-forget. |
| Rollup tier | `src/lib/rollups/` | DAY / WEEK / MONTH / YEAR pre-aggregations. Read-swap pattern: try the rollup, fall back to live SQL only on coverage miss. Boot-time `rollup-full-backfill` queue handles new accounts. |
| Compliance | `src/lib/analytics/compliance.ts` | Cadence-aware medication compliance — daysOfWeek + intervalWeeks honoured across the eight call sites that surface a rate (Coach prompt, BP-status gate, dashboard pillar, …). |
| AI providers | `src/lib/ai/provider.ts` (`resolveProvider`) + `src/lib/ai/provider-chain.ts` (`PROVIDER_CHAIN_TYPES`) + `src/lib/ai/{openai,anthropic,local,codex,mock}-client.ts` | Five providers, hand-rolled fetch over the documented wire (no vendor SDKs). Mock is excluded structurally by the `PROVIDER_CHAIN_TYPES` allowlist — production cannot reach it. |
| OpenAPI | `src/lib/openapi/registry.ts` + `src/lib/openapi/routes/` (23 modules, `index.ts` barrel) | Source of truth for `docs/api/openapi.yaml`. |
| Schema | `prisma/schema.prisma` (80 models) | `cuid()` PKs, `snake_case` columns via `@map`, encrypted columns mostly under `*Encrypted` (search the file). |
| Compose | `docker-compose.yml` | `app` + `db`. Env-var whitelist under `environment:` — vars not listed never reach the container. `pull_policy: always` is load-bearing. |
`.planning/codebase/arch.md` carries the full annotated walk for every section.
## Security-relevant patterns
The v1.5.2 audit produced `.planning/security-audit-v1.5.2.md` and a stack of per-domain findings under `.planning/security-audit-findings/`. The patterns worth preserving on future churn:
- **Encryption: fail closed everywhere.** Missing keys, malformed JSON, unknown key ids — every path throws rather than silently writing plaintext. Operators rotating keys must keep the legacy entry in `ENCRYPTION_KEYS` until the rotation script reports zero remaining legacy rows. The script covers every `*Encrypted` column + `CoachMessage.encryptedContent` (Bytes) + `IntegrationStatus.lastError`.
- **SSRF: input-time `isPublicUrl` is the floor.** The helper `isPublicUrl()` in `src/lib/validations/notifications.ts` catches every IPv4 / IPv6 alt-notation class (octal / hex / decimal IPv4, IPv4-mapped IPv6, IPv6 ULA, CGNAT, link-local, loopback, metadata). Use it for every outbound URL that comes from a user / admin input. DNS rebinding and the redirect-follow gap remain as architectural follow-ups (issues #217 + #218). Until those land, every NEW outbound `fetch()` that touches a user-controlled host must pin `redirect: "manual"` and an `AbortSignal.timeout`.
- **Outbound egress goes through `safeFetch`.** `src/lib/safe-fetch.ts` is the one documented egress entry (`redirect: "manual"` + `AbortSignal.timeout(15_000)` defaults; opt into the connect-time DNS-rebinding pin with `requirePublicHost: true` for any operator- or user-supplied host). The in-repo ESLint rule `healthlog/safe-fetch-required` (set to `error`) bans raw `fetch(` under `src/lib/` + `src/app/` outside the wrapper internals and test files; same-origin relative-path (`/api/…`) client fetches are exempt by construction.
- **Rate limits live in Postgres.** `src/lib/rate-limit.ts` uses a single atomic SQL upsert returning `{count, reset_at}`. Multi-instance correctness is structural. Bucket-key convention: `<surface>:${identifier}` where identifier is `userId` for authenticated routes and the trusted-proxy IP for anonymous. Anonymous auth surfaces use `checkAuthSurfaceRateLimit(...)` so a trust-violation collapses every anonymous caller into a single tight bucket rather than a free-for-all under the unknown-IP bucket.
- **Webhooks: shared-secret + `timingSafeEqual` + per-source rate limit BEFORE secret verification.** Withings, Telegram, moodLog, the Coolify deploy webhook, and the CSP report endpoint all follow this exact pattern. The Withings path-segment secret is scrubbed from `http.path` / `http.route` by `redactSecrets()` via the `PATH_SECRET_PATHS` registry in `src/lib/logging/redact.ts:40`.
- **Idempotency keys are user-scoped, length-bounded, and refuse to cache secret-shaped responses.** `src/lib/idempotency.ts`. A client cannot replay another user's idempotency key (composite unique index by `userId_key_method_path`). The cache helper refuses to persist a response body containing `hlk_` / `hlr_` / `sk-` patterns even if a future caller wraps an auth endpoint by mistake.
- **CSP: strict, per-request nonce, surgical `connect-src`.** Production CSP gates third-party hosts to the routes that need them — AI hosts only on `/settings/ai/*`, `wbsapi.withings.net` only on `/settings/integrations/withings/*` + `/api/withings/*`. Exactly one `dangerouslySetInnerHTML` in the tree (the theme bootstrap in `src/app/layout.tsx`, nonce-bound, file-literal content). No markdown library installed anywhere; Coach / briefing / insights render as plain React text children. Don't add one without a security review of the renderer's HTML-allow defaults.
- **No CORS headers anywhere.** Same-origin-only by construction. Adding `Access-Control-Allow-Origin: *` anywhere would silently open every authenticated endpoint to cross-origin POST; webhook routes accept cross-origin POSTs but they authenticate by header secret, not cookie.
- **Redaction at the egress boundary, not at the call site.** `redactSecrets()` + `redactSensitiveFields()` + `redactOptional()` are applied centrally in `WideEventBuilder` and `reportToGlitchtip`. Don't add new redactor calls in handler bodies — extend the central denylists in `src/lib/observability/redact-payload.ts` and `src/lib/logging/redact.ts`.
## Self-hosting gotchas
These keep tripping operators; they're worth knowing before you push something that breaks them.
- **`SESSION_COOKIE_SECURE`** controls the `Secure` flag on every cookie the app issues. Unset → flag tracks `NODE_ENV === "production"`. Set to `false` for LAN / Tailscale / VPN-only HTTP self-hosts where the operator deliberately serves plain HTTP. Set to `true` to force the flag even in dev. The compose `environment:` block must list it explicitly — vars not on the whitelist don't reach the container even when `.env` has them. v1.5.2 closed the missing-whitelist regression.
- **`pull_policy: always`** in `docker-compose.yml` is load-bearing. Without it Docker keeps a stale `:latest` digest cached and silently skips the registry round-trip on `compose up`.
- **`ENCRYPTION_KEYS` map vs legacy `ENCRYPTION_KEY`.** The map (`{"v1":"<hex>","v2":"<hex>"}` + `ENCRYPTION_ACTIVE_KEY_ID=v2`) is the modern path. The legacy single-key env var stays accepted as `v1`. Rotation playbook lives at `docs/ops/encryption-key-rotation.md`; do NOT drop a key from the map before the rotation script reports zero remaining rows on that key.
- **GeoLite2 databases.** Operators run `scripts/fetch-geolite2.sh` with a MaxMind licence key before `docker build`; the image bakes the MMDBs into `/opt/geolite2/`. Without them the resolver falls back to online `ipwho.is` and the admin login overview surfaces "unknown" for offline lookups.
- **APNs `.p8` key.** v1.4.47.x established `APNS_KEY_B64` as the env-var contract — pasting the PEM body verbatim through Coolify's `env_file` mangled the literal newlines. The key must be scoped "Sandbox & Production" in the Apple Developer Portal; a Sandbox-only key fails with `BadEnvironmentKeyInToken` on production builds.
## iOS handoff
The native SwiftUI client lives in a separate repository and rides public beta via TestFlight. Backend contract the iOS app speaks against is locked in `docs/api/openapi.yaml` and has been continuously validated since v1.4.23. Coordination notes worth knowing on the backend side:
-`POST /api/measurements/batch` recognises `externalId` values starting with `stats:` (e.g. `stats:HKQuantityTypeIdentifierStepCount:YYYY-MM-DD`) as **overwrites** — a re-post replaces the row's value / unit / measuredAt / externalSourceVersion / deviceType / sleepStage. Every other externalId prefix (`uuid-*`, opaque HK identifiers) keeps the first-write-wins immutable contract because each sample is a canonical reading. Per-entry status field carries `"updated"` for an overwrite vs `"inserted"` for a fresh row.
- Refresh-token rotation is per-device; reuse-detection revokes the family scoped to that device.
-`requireAdmin()` is cookie-only by construction — a Bearer token, even with `["*"]` scope, cannot reach `/api/admin/*`. The iOS app is not an admin client.
- The `clientManaged` opt-in on `PATCH /api/auth/me/notification-prefs` suppresses server-side `MEDICATION_REMINDER` APNs for clients that manage their own local reminders. The cron skips dose-due APNs when the flag is `true` and emits a `medication_reminder.suppressed_client_managed` wide-event annotation per skip.
-`push_attempts` table holds the last 90 days of APNS / Web-Push / Telegram / NTFY delivery attempts (channel + eventType + result + reason + createdAt). The admin diagnostic endpoint at `/api/admin/notifications/diagnostic` surfaces masked device tokens + channel state + last 20 attempts for the calling user.
- Any request / response schema change must regen `docs/api/openapi.yaml` and commit the result. CI fails on drift.
- The iOS app authenticates via password/passkey only (`src/lib/auth/native-client.ts`) — there is no OIDC flow for native clients yet. `OIDC_ONLY=true` is enforced server-side on those same routes, so enabling it locks every iOS user out of sign-in until a native SSO flow ships. Don't recommend it to an operator running the iOS app.
## DO-NOTs
- **No `--no-verify` on commits.** If a pre-commit hook fails, fix the cause; never bypass.
- **No `--no-gpg-sign` unless explicitly asked.**
- **No force-push to `main`** (or to release branches, by extension).
- **No destructive git operations as shortcuts.** `git reset --hard`, `git clean -fd`, `git checkout .` etc. only when the maintainer asked or the state is unambiguously local-only.
- **No assistant / planning vocabulary in any committed artefact** — see Voice and privacy above.
- **No personal names** — the maintainer's, a reporter's, or anyone else's — in any committed artefact, including code comments referring to "the user who reported issue #X".
- **No secret-shaped strings in `.env.example` or git history.** Pre-existing scan shows clean.
- **No markdown library in dependencies.** Coach + briefing + insights render as React text children for XSS reasons; adding one is a security decision, not a UX one.
- **No `Access-Control-Allow-Origin` headers** anywhere except the documented webhook surfaces — and those authenticate via header secret, not cookie.
- **No `pnpm tsx scripts/foo.ts` in production.** The standalone image strips `tsx`. One-shot scripts run via `pnpm dlx tsx`; recurring tasks belong on pg-boss.
- **No backwards-compatibility shims for hypothetical callers.** When a change is clean, make it cleanly; do not leave dead branches "in case" something we don't ship needs them.
| Maintainer working notes | `.planning/PROJECT.md`, `.planning/ROADMAP.md`, `.planning/STATE.md` (gitignored release scratch lives alongside; the patterns are in `.gitignore`) |
| Personal preferences + collaboration memory | `~/.claude/projects/-Users-marc-Projects-HealthLog/memory/MEMORY.md` (local to the maintainer's working copy; not in the repo) |
- **NEVER read `src/generated/**`** — Prisma-generated client (~9 MB, 83 files incl. a 224 KB+ inline schema). Reading/grepping it explodes the request context → 10-min "cogitations" + `ECONNRESET`on large turns. Regenerate via`pnpm prisma generate`if needed; never open it. (Enforced via`.claude/settings.local.json` Read-deny.)
- Avoid bulk-reading other large artifacts: `pnpm-lock.yaml`, `CHANGELOG.md`, `messages/*.json`, `docs/api/openapi.yaml` — grep narrowly instead of full reads.
- For very large multi-agent actions: cap concurrent sub-agents (many simultaneous streaming requests + big context is the worst combo for connection resets). When on a VPN (Tailscale MTU 1280 / iCloud Private Relay), large requests can hit MTU black-holes — disable the tunnel for heavy sessions.
@@ -4,7 +4,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
## Project
HealthLog — a personal health-tracking PWA (weight, blood pressure, pulse, mood, medication compliance) with Withings integration, moodLog.app sync, Dracula-themed UI, mobile-first design, offline caching, and doctor report PDF export.
HealthLog — a personal health-tracking PWA (weight, blood pressure, pulse, mood, medication compliance) with Withings integration, Dracula-themed UI, mobile-first design, offline caching, and doctor report PDF export.
"$comment":"Client-bundle budgets, enforced by scripts/check-bundle-budget.mjs (--check) after `next build` in CI. Values are KB gzip and include the shared rootMainFiles baseline. Measured at v1.27.15 + the mobile/PWA fix wave: / 417, /insights 402, /measurements 405, /insights/mood 419, total 2683, baseline 129 — budgets carry ~10-15% headroom so the gate catches step changes (a statically imported ~100 KB catalog, a ~90 KB duplicated library chunk), not noise. Raise a number consciously, in the PR that pays the cost, with the reason in the PR body.",
"$comment":"Client-bundle budgets, enforced by scripts/check-bundle-budget.mjs (--check) after `next build` in CI. Values are KB gzip and include every emitted lazy route and locale chunk, while route budgets include the shared rootMainFiles baseline. Re-measured at v1.34.1: / 431, /insights 420, /measurements 417, /insights/mood 438, total 3238, baseline 130, lazy catalogs 792. The 108 KB aggregate rise from v1.34.0's 3130 is not one user load: the explicit assessment error/timeout/poll-exhaustion contract adds 2.7-4.4 KB to each of roughly forty mutually exclusive metric route chunks, which the whole-build sum counts independently. Catalogs account for only about 2 KB of the delta, no catalog is statically route-referenced, and Recharts remains exactly one fingerprinted chunk. A measured single next/dynamic assessment boundary was reverted because it raised the artifact further to 3243 KB while saving only about 1 KB per route. The aggregate cap therefore moves narrowly from 3140 to 3260 (22 KB headroom); per-route ceilings, the static-catalog guard, and the one-Recharts-chunk guard remain unchanged.",
<titleid="title">HealthLog data flow — device to insight</title>
<descid="desc">Five-stage pipeline: sources (Withings, Apple Health export, iPhone HealthKit, manual entry, moodLog.app) feed ingest endpoints, which write to Postgres Measurement and MeasurementRollup tables, then surface on Dashboard tiles, Insights cards, the AI Coach, and the Doctor PDF.</desc>
<descid="desc">Five-stage pipeline: sources (Withings, Apple Health export, iPhone HealthKit, manual entry, Nightscout) feed ingest endpoints, which write to Postgres Measurement and MeasurementRollup tables, then surface on Dashboard tiles, Insights cards, the AI Coach, and the Doctor PDF.</desc>
| `04-source-priority.svg` | Three sources for the same metric on the same day → resolver picks one canonical row per (cumulative vs point) ladder. Losing rows stay in the audit trail. |
| `OPENAI` | <https://platform.openai.com/api-keys> | `gpt-4o` | `https://api.openai.com/v1` | Measurement context sent to OpenAI |
| `ANTHROPIC` | <https://console.anthropic.com/settings/keys> | `claude-sonnet-4-6` | Anthropic SDK default | Measurement context sent to Anthropic |
| `LOCAL` | None (your endpoint) | `local-model` | OpenAI-compatible URL you control | Stays on your network |
| `CHATGPT_OAUTH` | Sign in with your ChatGPT account | Codex-routed | `chatgpt.com/backend-api/codex/responses` | Routed via your ChatGPT subscription |
| `OPENAI`| <https://platform.openai.com/api-keys> | `gpt-4o` | `https://api.openai.com/v1` | Measurement context sent to OpenAI |
| `ANTHROPIC`| <https://console.anthropic.com/settings/keys> | `claude-sonnet-4-6` | Anthropic SDK default | Measurement context sent to Anthropic |
| `LOCAL`| None (your endpoint) | `local-model` | OpenAI-compatible URL you control | Stays on your network |
| `OPENAI_COMPATIBLE` | Your gateway's key, or none | none (you name it) | OpenAI-compatible URL you configure | Reaches whatever your gateway routes to |
| `CHATGPT_OAUTH` | Sign in with your ChatGPT account | Codex-routed | `chatgpt.com/backend-api/codex/responses` | Routed via your ChatGPT subscription |
The four spellings above are the canonical `User.aiProvider` enum
The five spellings above are the canonical `User.aiProvider` enum
values. Anything else falls through to the admin-shared OpenAI key
(if configured) or the `NoProvider` stub that surfaces "No AI
provider configured" to the UI.
@@ -54,6 +55,11 @@ order. The default is encoded in
4.`local` — self-hosted Ollama / LM Studio / vLLM endpoint.
| `<Workout>` | `Workout` rows | Activity type + duration + energy + distance. |
| `<Correlation>` | `Measurement` rows | Compound samples like blood pressure (systolic + diastolic) explode into their per-metric children. |
| `<ClinicalRecord>` | `Measurement` rows where the type maps | Health-record FHIR snapshots; only the metric-typed fields are ingested. |
| ECG CSV | `EcgRecording` rows | The source-provided waveform is encrypted before persistence; descriptors and the device's classification are stored without diagnostic inference. |
The HK quantity types that map to HealthLog metric types live in
`src/lib/measurements/apple-health-mapping.ts`. Anything outside the
mapped set (workout routes, ECG voltage traces) is counted in
`progress.records_skipped` but not stored.
mapped set (for example workout routes) is counted as unsupported and
not stored.
### ECG archive safety, privacy, and idempotency
ECG support is source-agnostic: if HealthKit includes a valid
`HKElectrocardiogram` CSV, HealthLog can import it regardless of which
compatible device originally wrote it. Archive member names are never
used as filesystem paths and ECGs are never extracted to plaintext
temporary files.
The auxiliary reader enforces limits before and during decompression:
- at most 20,000 total archive members and 2,000 ECG CSV members;
- at most 16 MiB per decompressed ECG CSV and 512 MiB across all ECG CSVs;
- at most a 200:1 declared compression ratio; and
- at most 2,000,000 waveform samples per recording.
Encrypted ZIP members, unsupported compression, unsafe or duplicate
paths, forged size metadata, malformed timestamps or samples, and
limit overruns are rejected. An ECG failure is fail-soft: valid
`export.xml` measurements and workouts still commit, while the
terminal result reports only
`discovered/imported/updated/skipped/failed` ECG counts. It never
returns filenames, source labels, sample values, or raw parser errors.
Waveforms are converted to integer microvolts and encrypted with the
existing AES-256-GCM ECG codec before the transaction writes an
`EcgRecording` with `source=APPLE_HEALTH`. Stable identity is derived
from normalized recording content rather than the CSV filename, so a
renamed re-export updates the same user-scoped row while distinct
recordings and different users remain independent. Known Apple
device classifications are preserved as source-reported facts;
unknown or localized classifications are stored as `null`. HealthLog
does not inspect the waveform to infer AFib or make a diagnosis.
### Cumulative HK type → daily aggregate
Cumulative quantity types — steps, active energy, walking/running
distance, flights climbed — get **collapsed into one row per
user-local day** to match the iOS daily-aggregation convention.
distance, flights climbed, time in daylight, and fall count — get
**collapsed into one estimated row per user-local day**. `export.xml`
contains source-specific records, not HealthKit's canonical daily
statistics, so the importer does not sum different devices or apps
into a source-blind total.
- Records are grouped by type, local day, and a SHA-256 hash of the
available source name, source version, and device attributes.
Values sum within each identified source-day; the largest
source-day subtotal is selected deterministically. If every source
attribute is absent, records remain separate rather than being
trusted as one source. Raw source and device labels are not
persisted.
- The row records `EXPORT_XML_SOURCE_MAX` provenance, its distinct
contributor count, and the selected non-identifying source hash.
- External ID format: `stats:<HKType>:<YYYY-MM-DD>` (e.g.
| `POSTGRES_PASSWORD` | Bundled Postgres password — used by both the `db` service and `DATABASE_URL` | Database restart + connection-string update |
| `ENCRYPTION_KEY` | AES-256-GCM key for at-rest secrets (Withings tokens, AI provider keys, VAPID secrets) | Coordinated; see `docs/ops/encryption-key-rotation.md` |
| `API_TOKEN_HMAC_KEY` | HMAC-SHA256 key for hashing Bearer API tokens and web session cookies before storage | Invalidates every issued `hlk_*` token and signs out every browser session |
You do **not** edit `DATABASE_URL` on the bundled `docker-compose.yml`
stack: compose constructs the connection string inside the `app`
- **`utils/`** — fixtures such as `mock-dashboard-snapshot.ts`.
- **`*.spec.ts`** — one spec per flow: auth/login/redirect, dashboard, charts, the medication-wizard cadences (`medications-wizard-*`), insights generation + scroll restoration, doctor report, settings, mobile-viewport consistency, locale switch, accessibility (`a11y.spec.ts`), and the `v1427-*` regression set.
## Convention
## Conventions
Import `test` and `expect` from `./setup/test`, never from `@playwright/test` — types (`Page`, `Route`, …) still come from the package. The entry point extends Playwright's `test` with an automatic fixture that drops route handlers and page listeners after every test, so a handler left in flight cannot fail the test it belonged to once the context closes. `src/__tests__/e2e-route-cleanup-guard.test.ts` enforces the import boundary.
Assert against stable `data-*` attributes (`data-slot`, `data-display-step`, root display attrs), not viewport-dependent visible text. Responsive `sm:hidden` UI changes break `getByText` on desktop while the underlying element is still present — anchoring on data attributes keeps the specs stable across breakpoints. See [`../CLAUDE.md`](../CLAUDE.md).
`an intentionally 1500px-wide block did not widen main's scroll area — main overflow-x:${state.overflowX} overflow-y:${state.overflowY}, canary width ${state.canaryWidth}, inside main: ${state.canaryInMain}`,
Loaded 100 of 2105 files, more files were not shown because too many files have changed in this diff.
Show more
Reference in new issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Marc-André Bombeck
