The recheck and follow-up rules now ride the tool-mode prompt only once
the conversation holds a table, or from the round after the turn produced
its first one. The clarifying-question rules are shorter and English for
every locale, like the other instructions to the model; the German copy
of them and of the recheck rule is gone. A turn without tables sends
about 1 300 characters less on every round.
The golden transcripts change only in the system prompt's hash and
length (4 509 to 3 187 characters; 4 728 to 3 280 for the German turn).
Ten scenarios, each in German and English, describe what a turn must do
with a question: which tool it calls and with which arguments, whether it
may read anything new, whether it asks back, declines a dose, offers
chips only for what the record holds, answers a chip without a provider,
offers to keep looking after a forced answer, and keeps an earlier
table's title out of the model's context. evaluateScenario grades a run
against them.
The deterministic suite drives POST /api/insights/chat with a scripted
provider; the tool loop, executor, table tool, chips, clarification
filter, method line, reply guards and stream run for real. Every check
also runs against a misbehaving model or a broken record and must fail
there.
route-followups.test.ts pins the front door for chips and answered
questions: validation, the rate limit and capability gate on a reuse
turn, the stored chip winning over anything the request adds, stale
chips and answers degrading to a plain message, and the continue line.
Up to three chips, derived on the server from what the turn read: the
other view of a table with a chart, the period before, a year earlier,
a wider window, and a metric a discovered pattern links to. A chip is
offered only when its read was done this turn and the record holds the
metric; the year-ago and wider-window chips also need the record's
first reading to reach back that far. The model may name chips in a
FOLLOWUPS block, but only to reorder ones the rules already allow; every
label is rendered from the catalog.
A tapped chip resolves against the chips stored on the conversation's
latest reply. "As a chart" and "as a table" are answered from the
stored table without a model call or a budget reservation: the rows are
copied unchanged and the message is tagged providerType "reuse". Other
chips run a normal turn with a server-written context line. A chip that
is no longer current falls back to a plain message.
A setting turns the chips off (absent means on).
The line that resolves an answered clarifying question was already
appended to the system prompt on both paths; nothing proved it for the
no-tools (local model) path. Three route-level cases now do: the local
prompt and the tool-mode prompt both carry the server-resolved line, and
a turn without an answer carries neither. Removing the line from the
local prompt fails the first case.
The golden harness gains a coachMessage.findMany mock for this; it
defaults to no rows, so no existing transcript changes and the golden
snapshots are untouched.
The result-table caption test used the regex dotAll flag, which the
TypeScript target rejects (TS1501); it matches across lines with
[\s\S] instead.
get_metric_table reads one metric as days, weeks or months through the
chart reader, cut in the user's zone and collapsed to one source per day,
with null for a period without a reading. The model reads a bounded
summary; the full table rides the turn, encrypted at rest. show_result
shows a table an earlier reply of the same conversation fetched, resolved
only against that conversation's own messages. Workouts, labs, sleep and
adherence results project into tables, earlier tables are named in the
context by metadata alone, and result:rN marks are stripped from the reply.
The evidence disclosure under a Coach answer now carries one muted
line: which sources and windows the turn read, how many readings, and
how they were aggregated ("Blood pressure, last 90 days: 142 readings,
weekly averages · Sleep, last 30 days: no readings in this window").
The server builds it from the recorded steps and tables in the request
locale; it holds catalog strings and counts only, never a reading, and
a source that turned up nothing says so.
The tool-mode prompt gains a recheck rule: when the person questions a
figure, name the source and window, read it again (or reuse the stored
table), and correct the answer if the two differ.
Each tool call now emits a running step as it starts and its final
status as it settles: found with a server-counted number of readings,
empty with the reason, or failed. The no-tools path shows one step for
the record summary. Steps carry catalog labels rendered in the request
language, closed-enum domains and windows, and integers only; a lab
analyte, a value or a model-written argument never reaches one. At most
twelve per turn. They persist on the provenance, so a reload shows the
same list.
The conversation window reaches the step so a call without a window
names the one it actually read. The no-tools transcripts gain the
summary step; a new transcript pins found, empty and failed steps ahead
of the first token.
A reply may end with a ---CLARIFY--- block (kind metric, window or
context, 512-byte cap). The server keeps only metric choices the record
holds, drops the question when fewer than two remain, takes window
choices from the presets, screens the question text like any reply, and
never offers a second question right after one. Choice labels come from
the catalog, never from the model.
An answer arrives as clarification { messageId, choiceId? }; it resolves
against the stored question on the conversation's latest message and
adds one server-written line to the next turn's context. The card sits
above the composer, announces politely and takes no focus.
A reply that is only the block, with no question before it, is now an
empty reply instead of falling back to the unstripped text.
GET /api/insights/chat/{id}/messages/{messageId}/results returns the
tables one assistant message read, narrowed to the conversation's owner
like the conversation read, with a table withheld as module_disabled when
its domain's module is off and as unavailable when it cannot be read. A
data read: no AI capability.
The published document now describes every stream frame as the
CoachStreamEvent union, including the suggestedAction frame it never
named, the new step, result, followUps and clarification frames, their
components, the provenance fields, the followUp and clarification request
fields, and the reuse provider type.
Each later piece of the dialog gets its own module, answering nothing for
now: the step builder, the table projection and chart choice, the result
ref marks, the follow-up derivation, sentinel, resolution and continue
chip, the reuse turn, the clarification parse and answer, the method line,
and the four prompt addenda. The turn pipeline calls them where they
belong, so the frames go out in the documented order (step, token,
provenance, result, cards, clarification, followUps, done) and a blocked
turn drops its tables, chips and clarification. With every piece empty
the stream and the prompt stay byte for byte as before.
The route keeps the front door: auth, the coach capability, the body
read and schema, the rate limit, the locale and the inbound refusal
screen. Everything after it moves into one module per step:
conversation, context, chain, budget, model, reply-guards, cards,
provenance, persist, sse, errors and pipeline. The capability re-check
at the egress site stays owned by the route and is handed to the
pipeline, so the capability inventory still reads both calls there.
The stream producer now writes through a TurnEmitter created inside
createSseStream and passed to the model step, which is where live
progress frames will come from. Nothing writes to it early yet.
No behaviour change: the golden transcripts pass byte-identical, the
existing route tests pass unmodified, and every annotate, audit and
error-code name appears as often as before.
Guards follow the code. The fence call-site check reads turn/context.ts,
the German-default locale ban covers every file in turn/, and the
fenced module-graph test now asserts that turn/sse.ts and turn/errors.ts
reach neither the tool registry nor the snapshot builder. That test
named lib/ai/coach/tools.ts, which stopped existing when the registry
became a directory, so its negative check could not fail; it points at
tools/index.ts now, and a new sanity check fails if the tool route can
no longer reach the path it names.
Drive POST /api/insights/chat through the real SSE stream and the real
reply guards for every branch a turn can take: tool mode with present,
drifted and missing figures, the no-tools path, inbound and replayed
refusals, the budget refusal, each provider failure class, a cancelled
turn, the outbound block, every sentinel in its valid and malformed
forms, learn-link scrubbing, the three no-provider paths, a workout
scope, the post-call defect path, and the refusals that answer before
the stream opens.
Each transcript records the frames the client receives (heartbeat
comments dropped) and the ordered side effects: persisted messages,
budget reserve and reconcile, audit rows, annotate names with their
meta keys, and the provider inputs as prompt fingerprints.
The status assessments folded their recent days, weeks, months and years on
Berlin days whatever the user's zone, the mood and cycle crosstabs keyed
their metric days in Berlin, the resting-pulse proxy defaulted to Berlin
days, the briefing's opener rotated at Berlin midnight, and every
assessment prompt labelled today's date as Berlin's. They now use the
user's zone. Rollup months and years are labelled by their UTC start,
which is how they are cut; read in a zone west of UTC the label named the
month before.
"Europe/Berlin" was written out in some eighty places. DEFAULT_TIMEZONE in
lib/tz/format is now the one definition: the second copy in mood/date-key,
DISPLAY_TIMEZONE, the Berlin-only day key helpers and the per-route copies
are gone. An in-repo lint rule rejects the literal outside lib/tz and
tests; four areas another change owns are listed as pending.
A per-medication trackIntake flag (migration 0351, default on for every
existing and new row). With it off the dose, dates and schedule stay stored,
but nothing derives a due dose from them: the reminder worker, the manual
reminder check, the today projector, the auto-miss pass, the low-stock pass,
the doses card, the today list, the outstanding-dose badge, compliance and
its heatmap, the doctor report, the Coach, the BP-status gate, insights,
targets, achievements, the efficacy view and the MCP schedule tool all skip
it.
The medication wire serves such a medication with schedules: [] and the
stored rows in an additive recordedSchedules field, so a client that arms
local reminders from schedules stays silent without an update. Switching
tracking off or back on archives a schedule revision at that instant, the
untracked stretch with an empty payload, so it never counts as missed and
the projector does not put a dose from before the switch back on the card.
A structural guard freezes the set of schedule readers: each one filters on
the flag, derives from dueSchedules, shapes the wire with scheduleWireFields,
or is allowlisted with the reason it derives nothing due. The flag travels
through backup, restore and the medication CSV export.
Serving stored text and starting model work are separate questions. A
delegate may not start model work on somebody else's record, but reads
the owner's stored briefing, period narrative, status notes, workout
notes and Coach unread signal exactly when the owner would: those
readers now resolve the record's own capability state, not the
viewer's. Generation, warms and enqueues still ask as the viewer, so a
delegate's read of a stale narrative queues nothing. A hidden
model-written narrative is no longer overwritten by a deterministic
warm while AI is unavailable.
The sharing front door reads the digest's briefing lead again, and an
integration case pins both halves.
Refs #975
Brings in the jobs, stored-text and consent package and the extraction
routes. Conflicts resolved:
- The digest keeps its capability-driven reaction line and check-in,
now through the shared `aiCapabilityForRecord`; the local copy of it
is removed.
- The route inventories classify the extraction routes (documents,
labs OCR, medication text, fenced chat and attachments) as actions;
a capability call counts as the gate for the module inventory.
- With every caller moved, `requireAssistantSurface` and
`AssistantDisabledError` are deleted. The `assistant.disabled.*`
codes stay on the capability refusal, and the guard now asserts the
retired gate is gone.
- `aiCapabilityToServe` answers whether stored model text may be shown
from the record's own state, apart from who is reading.
Refs #975
The correlation scan still paired mood, sleep, glucose, adherence,
symptoms and weather with everything else after their modules were
switched off. The discovery matrix now drops every channel whose module
is off, read through the surface map, and the correlations route passes
the record's module map, skips the lab draws with labs off, and keys its
cache on the switched-off set.
The status family (the five metric cards, medication compliance, the
generic metric status and the biomarker assessment) asked the retired
switch and the `insights` module and refused with 403. The note is model
text and the card around it is data, so they are mixed reads now: while
the `statusText` capability is available the note is served and warmed
as before, with `ai` beside it; otherwise the answer is 200 with the
note null, `preparing: false`, `hasProvider` as provider presence only
and `ai` naming the reason, and neither the cache nor the warm queue is
touched. The generic metric status still reports `insufficient` from the
data. The mood card keeps its mood module gate.
Kept apart so it can be held back while the shipped iPhone app's
rendering of a 200 without a note is confirmed.
Refs #975
Data never depends on AI. The Insights data reads no longer ask an AI
switch or the `insights` module (now only the AI analysis opt-out): ECG
list, detail and upload, rhythm events, derived scores, rule cards,
correlations and patterns, the baseline strip, health status, intraday
pulse, and the Today digest with its dismiss. Breathing screening moves
to the sleep module and lab changes to the labs module. Stored Coach
facts, conversations, message ratings and the seen stamp stay readable
and erasable while the Coach is unavailable. Cards now say
`provider: "rules"`.
Reads that mix data with model text answer 200, null the model text
and publish an `ai` state from the capability resolver, without warming
anything: comprehensive (`ai.briefing`, `ai.statusText`, `hasProvider`
from the presence probe), the derived assessment override, the period
narrative, nudge status, the seeded opener, about-me questions, the
workout paragraph, the pregenerate warm (`queued: false`), the Today
digest (lead, reaction line, check-in) and the dashboard briefing,
which is now applied per read after the snapshot cache. Hiding the
Coach no longer hides the briefing.
`insights/generate` GET and POST ask the `briefing` capability and
`insights/chat` POST the `coach` capability, with the typed refusal
envelope; a missing provider keeps the chat's `coach.provider.none`
frame and adds `reason`. The /coach page, the MCP ECG and intraday
tools and `aiAvailable` on the provider route follow the same answer.
An inventory of AI and data routes replaces the Coach gate inventory:
every route that asks a capability names it, and no data route asks.
Refs #975
Reading a stored document (suggest, summary, extract, index, chat,
reindex), scanning a lab report and turning medication text into a
schedule now answer under documentAi, labsOcr and medicationExtract. The
operator's master switch and the new reading-documents switch stop the
work and the egress; a module off for the record, a delegate inside
somebody else's record, and a missing receipt refuse with the capability
envelope. Existing no-provider codes keep their status.
The document pick and the labs pick re-check the capability for the
provider they picked, immediately before anything is sent, so a job
enqueued before a switch flipped finds no provider and falls back to the
provider-free text layer. Consent follows the document rule for all
three: any provider that leaves the machine needs an ai_extraction or
ai_full receipt, answered for the provider actually picked.
The fenced Coach chat and attachments answer under coach and documentAi.
Uploads are always stored and indexed; the background summary is only
enqueued while documentAi is open. The capability probes report the
capability beside the provider answer and keep a read offered when only
the receipt is missing. medications/extract keeps its 503 for no provider
and adds ai.provider.none, and its budget refusal gains the
coach.budget.exceeded code.
The Coach gate inventory lists the correlations route with the routes
no assistant switch owns. The seeded-question test gives the switch
read a settings row, since a read that fails no longer opens every
switch.
A single account syncing heart rate from a watch took the worker down
within half an hour. The status cards' graded series read the last 91
days of raw rows, bounded in time but not in rows, so a stream sampled
every minute or faster materialised six or seven figures of objects on
every refresh and the read ran into the statement timeout.
Add readDayAggregates, which folds one type into per-day sum, count, min
and max in Postgres (optionally split into age segments and filtered to
the metric's plausibility domain), and move the readers that folded raw
rows into days in memory onto it: the graded series (recent, weekly and
the cold-tier fallback, for every type), the vitals day-mean fallback
behind the personal band, the tiered context's recent days and the plan
review. Recent, weekly and monthly buckets are identical to the raw
fold; the cold-tier yearly slope now runs over day means.
On 91 days of per-minute pulse the graded series now allocates 1.6 MB
where it allocated 62 MB, while the raw read alone holds 22 MB.
A structural guard freezes the remaining untaken measurement reads in
the worker and insight directories: each must name a sparse literal
type or carry a listed reason.
Refs #1023
The operator switches become master, Coach, Daily Briefing, Status
notes and Reading documents. Correlations goes: it gated statistics,
not anything a model writes. Reading documents covers the document
vault, lab scans and medication extraction.
The Coach module's operator availability now follows the Coach
switch, so there is one place to turn the Coach off server-wide. The
migration folds a stored Coach availability of false into the switch,
and gives everyone who hid the Coach the AI analysis opt-out, since
hiding the Coach no longer stops nightly briefing and status notes.
A switch read that fails now fails closed. The insights module is
described as AI analysis in every locale, and the switch guard now
keeps both retired switches out and freezes the callers of the old
operator-only gate.
With the assistant switched off on the server, the Insights overview did
not load. Its main read, /api/insights/comprehensive, was gated on the
Coach surface, and the derived tiles, the ECG list, strip and live ingest,
and the rhythm events were gated on the status-card surface. The master
flag forces every surface off, so all of them answered 403 and the page
replaced the whole overview with an error.
None of these handlers calls a provider or returns assistant prose. They
now gate on the insights module only, like health-status, labs-changes and
patterns already do. comprehensive had no module gate of its own, only the
Coach gate, so it gains the insights module gate the other five already
call. Routes that serve generated text keep their gate.
The route-gate inventory moves the six files onto the not-Coach-owned
list, four OpenAPI descriptions drop the gate they no longer have, and a
comprehensive test pins a 200 with the master flag off.
Refs #975
CLAUDE.md forbids planning shorthand in committed artefacts, and this branch
introduced 74 lines of it across source, tests, a schema field and two
migrations. Someone reading the migration SQL in a year has no way to resolve a
wave letter or a round number, and a comment that points at something outside
the repository names no reason at all.
Three passes. A version followed by a parenthetical keeps the version and loses
the parenthetical, because the version says since when something holds. A
sentence that leaned on the vocabulary to carry its meaning is rewritten to
state the reason directly: the arrival reservation now says it was refused over
money the operator never spent, rather than naming the defect a wave existed to
remove. Three test titles carried the marker in the name itself and are renamed
to what they pin.
Comments only. No behaviour, no renames beyond those three titles.
The owner split left an operator-funded chain with no ceiling at all. Its arm
compared operator_tokens, and that counter returns to roughly zero on every
reconcile the user's own plan settled: on the 2026-09-11 evidence the shared
key answered 500 and codex served, so the counter never grew, every job
reservation was admitted forever, and a runaway client loop could write an
unbounded row. A pre-existing concurrency contract caught it from the other
side, two reservations that must not both pass one cap both passing.
reserveBudget now refuses when the owner's counter is over its cap OR the day's
mixed total is over the abuse ceiling, and a background surface gets its share
of that ceiling too, so background spend is bounded on the total and not only
under a counter the operator's key never fills. The surface is a required
argument, so the type checker enumerated every reservation site.
The hop guard learns the same lesson: an operator-funded fallback hop is now
rationed at the calling surface's ceiling rather than the full operator cap. A
job admitted under the job share could previously spend the whole 200 000 once
its chain fell back onto the shared key, which is the original defect reached
by the fallback path.
The job share applies only where the operator pays for the chain. Multiplying
whichever ceiling the chain resolved halved background generation for a
self-hoster on a local model, who costs the operator nothing.
Refusals now name the ceiling they hit and whose it was. A refusal used to emit
the day's mixed total and nothing else, which on an operator refusal is not the
counter that tripped.
The automatic generators — status and reference cards, the briefing, the
arrival reaction, the nudge, document summaries and the index backfill —
reserved against the same ceiling as the chat. On the operator's account
that meant 150-330 ledger rows a day, 172 metric generations before
06:42Z, and a ceiling already spent when he opened the chat.
Every background surface now reserves through
`resolveDailyCapFor("job", chain)`: half the cost owner's ceiling, all
job surfaces together. The interactive routes keep the whole ceiling, so
a day of background work can no longer take the user's own away.
The reservation compared the day's TOTAL against whichever ceiling the
chain resolved, and the chain walker's hop guard did the same. On an
account whose background jobs run on the user's own ChatGPT plan, both
saw a total the operator had not paid for: the chat refused at 06:42Z
with 513 539 tokens against the 200 000 operator ceiling, and the
fallback hop to the shared key was closed on the same arithmetic.
An operator-funded chain is now measured against `operator_tokens` and a
user-plan chain keeps its abuse ceiling on the total; the hop guard reads
the operator share too. A day of turns the user's own plan paid for can
no longer lock anyone out of the operator's ceiling.
`coach_usage` counted one number per (user, day). A turn served by the
user's own ChatGPT plan after the operator's shared key answered 500 was
therefore charged to the operator-cost cap, and the operator's chat was
locked out every morning by background work he pays for himself.
The ledger gains `operator_tokens`: the share of `total_tokens` served by
an operator-funded provider. `reserveBudget` books a reservation to the
owner of the chain's primary and `reconcileSpend` settles it against the
provider that actually served, both inside the same statements as the
total so concurrent requests can never see one counter without the other.
Every call site names its cost owner and its serving hop; the argument is
required, so a new surface cannot inherit the old blind accounting.
`PUT /api/auth/me/report-selection` threw a bare 422 whose whole `error`
string was `report-selection.leaves.unknown` — a machine token in the
field a client shows a person, with nothing beside it to branch on and
no way to learn which leaves were refused. It now answers a sentence,
with the token in `meta.errorCode` and the refused ids in
`meta.unknownLeaves`, and the operation's description says so.
The site was invisible to the refusal guard because the refusal is
decided by `if (!minted.ok)` rather than by a parse result, so the guard
now reads the refused string as well as the deciding block: every 422
under `src/app/api` whose `error` is a dotted token is inventoried with
the reason it still is one. Eleven remain, each with its reason, and the
inventory may only shrink. Removing the fix above turns it red naming
that route.
The four sibling shape refusals keep their token in `error` on purpose,
and the comment now says why that is not the same decision as the seven
`invalid_json` sites: those moved their string because their status
moved to 400 anyway, while a shape refusal that keeps its 422 would be
changing the wire for nothing. Both halves publish `meta.errorCode`.
Two loose ends in the same guard: a block calling `returnAllZodIssues`
or `apiValidationError` without spelling `422` is now judged (their
default status is 422, so a future `refuse422(err)` helper cannot slip
past), and the site floor moves from 40 to 200 against the 248 the tree
carries — the old floor would have survived losing five matches in six.
Fifty-nine 422s raised off a Zod `safeParse` answered with one opaque
sentence and nothing a client could map onto a field, while the other
166 sent `details.issues` — the multi-issue envelope that exists so a
caller with three bad fields fixes all three in one round-trip. Twelve
of the fifty-nine put a dotted machine token in `error` where the
sentence belongs, so `error` was sometimes text to show a person and
sometimes a token that must not be shown, with nothing in the payload
to tell them apart.
Every one of them now answers through `apiValidationError`, which keeps
the existing `error` string and the existing `meta` byte for byte and
adds the issue list beside them. The dotted tokens keep their place in
`error` for the clients that already match on them and are published in
`meta.errorCode` as well, which is the field a machine code belongs in
and the one a new client should read.
A guard reads the route tree per refusal site rather than per file —
per file is what let this grow, since a route that used the helper once
looked compliant to a `grep -l` sweep — and asserts it still finds the
sites it judges, so a matcher that stops matching fails instead of
reporting a clean zero.
Three consent-surface tests mocked the shared AI-consent bucket at 60 where it
is 20, and the analytics-read and MCP mocks named 60 against real caps of 120.
Nothing asserts the number today, so nothing was wrong on the wire — but the
mocked verdict now carries `limit` all the way to `X-RateLimit-Limit`, and the
first test that reads that header off a mock would pin the fiction.
The response-shape file's header pointed at a document that is not in the
repo; it says what it means directly instead. It also now records the one
thing the gate cannot see: `meta` is a loose object on purpose, so undeclared
keys inside it survive the parse and the strip comparison does not reach them.
A refused client had nothing to back off against. `rateLimitHeaders()`
emitted `X-RateLimit-Remaining` and an ISO `X-RateLimit-Reset`, no
`Retry-After` and no cap, and only about 70 of the 199 route files that
consult a limiter attached even those two. The rest answered with a bare
prose envelope, including every endpoint an offline replay queue hits:
the four batch ingests, `/api/sync/changes` and `/api/auth/refresh`. A
first-run Apple Health import trips the 60-per-minute batch ceiling
reliably, and the only strategy left to the client was a blind backoff.
Threading headers through 129 handlers is 129 edits the next new route
forgets, so the verdict is carried instead: `checkRateLimit` records it
in a request-scoped cell and `apiHandler` dresses any 429 that leaves
without headers. A route that already built its own is left alone —
`Retry-After` is the presence test, and it comes from the same builder.
`Retry-After` is whole seconds rounding UP (RFC 9110 §10.2.3), floored
at zero. `X-RateLimit-Limit` is new and needs the cap, so the limiter
result carries it. `X-RateLimit-Reset` keeps its ISO form: it is the
wrong convention for the name, but it is on the wire today and changing
it is a decision of its own.
The cell lives in its own leaf module because 177 test files replace
`@/lib/rate-limit` wholesale with a mock factory, and `apiHandler` must
not depend on anything those factories stand in for.
A `z.object()` without `.strict()` strips an unknown key and parses
clean, so a caller that misspells a field name gets a 200 and loses the
value with nothing to tell it. The worst of them was the Coach
question-dismiss endpoint, where an omitted `question` means "dismiss
all": a typo in that key wiped every pending question and answered as
though one had been dismissed.
Twenty-four inline schemas across twenty-two endpoints are strict now —
the settings toggles, the restore and bulk-delete endpoints, message
feedback, the admin user update, the onboarding acknowledgements, the
provider chain, the chart overlay preferences, the medication extractor
and the two client-side reporters.
The strip stays where it carries weight. Sync ingest, device
registration and the layout PUTs are built out of fields that shipped as
additive and optional, on the promise that a server which does not know
one ignores it; an app running ahead of the install it talks to is
routine here rather than exceptional, and refusing would turn a dropped
field into a refused batch — a year of history lost to one row, a
dashboard that will not save, a phone that quietly stops registering for
push. A read filter discards nothing when a stray query parameter rides
along, the OAuth endpoints are required by RFC 6749 and RFC 7591 to
ignore parameters they do not recognise, and an import file comes from
another version or another tool by definition. Each is registered with
its reason in a new inventory test, so a schema cannot join them without
someone making the call. That test parses the tree rather than grepping
it: fourteen of these schemas are written `z\n .object({…})` and a
`z.object(` sweep never saw them.
`sanitiseZodIssues` renders `unrecognized_keys` instead of passing it
through. Zod leaves `path` empty and puts the names in `issue.keys`, so
the sanitised issue named nothing at all and the caller was told only
that its body had failed. The names now ride a bounded `keys` field and
the message is rebuilt from them, capped at ten keys of 64 characters
and restricted to identifier characters, so a caller-supplied key cannot
echo back verbatim through the error path or into the audit ledger. The
names survive `stripValuesFromMessage` because a field name is
structure, not user-typed content.
The blood-pressure medication continuity series joined every day that had a
systolic reading against `takenByDay.get(day) ?? 0`, so a day with no intake
record at all entered the scatter and the coefficient as a measured 0 %
continuity. A week away from the app read as a week off the medication.
Only days carrying an intake record are paired now. A logged skip is a real
zero and stays; a day with no record is absent, and absence is not zero.
The timezone-parity fixture in the route test recorded three logged days out
of five and relied on the other two arriving as zeros, which pinned the
defect rather than the timezone question it asks. Its two remaining days are
recorded as skips, which keeps five paired days and keeps every one of them
something the person actually logged.
`calculateCompliance` answers `rate: 100, totalExpected: 0` for a medication
that expects no dose. Arithmetically honest, and a lie on a surface that
reads the number as adherence — it also inflated the averages the AI prompts
are built from. The doctor report has excluded both the PRN arm and the
empty-schedule arm since it was written, with the reason beside it. The
insights envelope, the blood-pressure gate and the medication-compliance
snapshot filtered only PRN, and the empty schedule is reachable:
`POST /api/medications` accepts `schedules ?? []` on a scheduled medication.
Move both arms into one predicate, `expectsDoses`, and route all four
surfaces through it so they cannot drift apart again.
The comprehensive-briefing cache (User.insightsCachedText) is one slot
per account with no record of the language it was written in. Every
reader served whatever was there, so a German-language account whose
nightly warm had written English kept seeing English on the dashboard
lead and the Today briefing. The request path only replaced it once the
row was a day old and the provider answered, which with a provider at
its usage limit was never.
Add User.insightsCachedLocale (migration 0335). Both writers store the
locale they generated in; the timestamp-only refresh on an unchanged
snapshot tags the row too, since the fingerprint covers the generation
locale. Every reset that nulls the text nulls the tag.
The snapshot briefing lift and GET /api/insights/generate treat a tag
that differs from the reader's locale as an empty cache: no prose, no
timestamp, the preparing / no-provider state, and the out-of-band warm
enqueued in the reader's locale through the existing path. The
generator's own short-circuit and the POST short-circuit refuse to call
a foreign-language row fresh. A NULL tag (rows written before this
change) is served as before until the next write tags it. The Today
digest derives from the snapshot and follows.
Covered by unit tests on the lift, the route and the generator, and an
integration test that writes the row through Prisma and reads the
snapshot for a reader in another language.
`GET /api/insights/derived` accepts `windowDays`, bounded to 1..90, and
threads it to the dispatcher. Omitting it changes nothing: every engine
keeps its own default, so a caller that asks for nothing reads exactly
what it read before.
Ninety is where the tier this route promises runs out. The rollup router
opens its WEEK floor at 91 days, and the baseline engines cannot compose
a spread from WEEK buckets, so a 91-day request falls through to a raw
`measurement.findMany` with no row cap — for a densely sampled type that
is a full history scan, and the analytics-read budget allows 120 of them
a minute per account. The trailing `series` is capped at 30 points
regardless, so past the ceiling a wider window would only broaden the
mean behind the value. Out of range is a 422 rather than a clamp.
An unhonoured window is visible rather than silent. `provenance.windowDays`
is the window the engine actually used and `coverage.historyDays` the days
that actually backed it, so a three-week record answering a thirty-day
request reports both numbers instead of implying coverage it does not
have. `HEALTH_SCORE` composes fixed per-pillar windows and reports the
widest; a new suite freezes that it is the only metric that does, so an
arm added later cannot quietly drop the parameter.
Recovery keeps its canonical resolution under the wider window. A worn
band stamps the wake morning and the computed proxy the night before, and
only `resolveCanonicalRecovery` pairs them; a test seeds forty such
nights, asks for sixty days, and holds the answer to forty nights with a
flat trend — a bypass leaves eighty rows and a false climb.
Contract: model `WellnessScoreValue` field for field and publish it. The
envelope's `value` has to stay an open record because one route serves
eighteen differently-shaped metrics, so nothing can `$ref` the shape and
it went out undocumented — `series`, `daysInWindow` and `asOf` were on
the wire with no way to find them short of inspecting a live response. It
is registered in the forced-components slot instead, held against the
runtime interface by a test, and the fields it does not cover are named
in the schema comment so the remaining gap is known rather than invisible.
Four surfaces assembled the correlation-discovery matrix and three carried a
different channel set. The Coach `get_correlations` tool had neither the
environmental-exposure channels nor the custom-metric ones, so asking it
whether air pressure relates to your sleep met a matrix the pair had never
been in. Its own file header said the opposite - "the SAME full-matrix scan
the route runs" - and had said it since the weather channels arrived in
v1.25. The period narrative was missing those two families and the
compliance / symptom pair on top.
`src/lib/insights/discovery-matrix.ts` is now the one place a matrix is
built. v1.30.3 hoisted the fetch discipline into a shared module for exactly
this reason and left the assembly duplicated behind; this finishes it. The
route, the per-metric card, the Coach tool and the period narrative all call
`assembleDiscoveryMatrix` and get the same families. What still differs is
declared as an option with its reason at the declaration: the measurement
read mode (a read-cost choice, left per surface), the extra metric types the
narrative's delta beat needs, and the RATED mood factors.
Mood factors stay on the narrative alone. That capture kind is frozen, the
two surfaces that persist a discovered pair write its channel key into a
plaintext column, and a custom factor's key is an opaque id - so admitting
them elsewhere is a decision about what to ask, not a parity fix.
Restore the per-metric filter on the assessment card's relations. It was
written in place of the pattern-dismissal filter rather than chained with it,
so a weight card's prompt had been receiving every surviving pair in the
record while the function name, the docstring and the early-return guard all
still said one metric.
`discovery-matrix-guard.test.ts` freezes who may fetch a channel, who may
construct one, and who may scan. Each matcher asserts a non-zero match count,
so a matcher that has stopped matching fails rather than agreeing with the
allowlist.
The statistics are untouched - same Pearson, same exact p-value, same
Benjamini-Hochberg, same floors. The Coach's family is larger now, which is
the point: it is corrected over the same family the insights page is.
The GLP-1 side-effect chips in the mood form write the localised label into
the flat `MoodEntry.tags` column, and three readers matched that label against
a hand-written list of English and German words: the therapy timeline, the
Coach snapshot, and the doctor report's GLP-1 block. The app ships six locales.
Tap the nausea chip with the UI in French, Spanish, Italian or Polish and the
day never reached the timeline, the Coach grounded its answers on an empty
side-effect list, and a clinician read an empty table over a record that had
three months of symptoms. Nothing on any of those surfaces said the table was a
filter rather than the truth. The three lists had also drifted apart from each
other.
The chips are a closed set with stable keys, so matching the label was the
defect and the key is the fix. `glp1-side-effect-tags.ts` holds the catalogue —
seven keys, each with the i18n key its chip renders — and the mood form now
draws its strip from it, so the write side and the read side cannot diverge
again. `glp1-side-effect-tag-match.ts` builds the label-to-key index by reading
`messages/<locale>.json` for every entry in `locales`, rather than transcribing
the labels into a second list. Transcribing would have fixed today's four
locales and reopened the same hole on the day a seventh is added; deriving
means a new bundle is matchable as soon as it lands, with no edit in the
matcher. Comparison folds case, accents and separators, so the English and
German tags already in the database keep resolving and a straight apostrophe
matches the typographic one the French bundle uses.
All three readers now report the catalogue key instead of the stored string.
The same recorded symptom yields one key whatever language it was tagged in,
which also fixes the mixed-language output an account gets after switching UI
locale; the doctor-report PDF resolves the key through the label map so the
clinician reads it in the report's language.
A tag outside the catalogue is still never counted — a free-text mood tag must
not reach a doctor as an invented symptom, and the rest of somebody's day does
not belong on a therapy timeline. It is no longer dropped without trace: the
timeline route annotates the unresolved count on its wide event, so the
classification rate stays answerable while the tag text stays in the database.
The Coach's copy of the list also carried `vomiting`, `reflux` and `erbrechen`.
Those are removed. No chip writes them, so they only ever matched a hand-typed
tag in two of six languages; keeping them would have rebuilt the defect inside
its own fix. Restoring either symptom means giving it a chip and a label in all
six bundles.
The response schema for `GET /api/insights/glp1-timeline` now enumerates the
keys, and its description no longer documents the bilingual match as if it were
the contract.
The numeric grounding guard keyed its activation off `loop.toolResults` alone.
A tool that finds nothing resolves to `{ present: false, reason: "none" }` with
no `available` payload, `runCoachToolLoop` drops that shape, and the no-tools
snapshot fallback is populated only in the non-tool branch — so the guard sat
out every turn where the record was empty, which is precisely the turn where a
fabricated figure is likeliest and does the most harm. Ask about a metric you
have never logged and an invented number was persisted and streamed verbatim,
with no withheld-figure marker to say it had never been checked.
`toolTrace` is the discriminant, because it records the pure misses the results
array drops. Tools ran and every one missed: grade the reply — an empty ledger
there is not missing information, it is the positive statement that nothing can
be cited, so `findUnverifiedCoachNumbersInLedger` takes an explicit opt-in to
grade against it rather than no-opping. The model answered without calling a
tool: still dormant, unchanged, because the tool-mode base prompt carries no
pre-computed figures and flagging a legitimately recalled one was the v1.32.1
regression. The inventory counts ride along as a widener so a plain count
restatement stays grounded.
Eliding the digits is not enough on an empty record. "Your sleep averaged […]
minutes, up from […]" still asserts a series, an average and a direction for a
metric with no readings; only the precision goes, the false claim stays. So a
turn that missed every tool and lost a figure to the guard is replaced with the
honest answer, in all six locales. Scoped so a reply whose figure reconciles
against a prior turn or an inventory count is left alone.
Three characterisation tests for grounding gaps found while verifying that
every figure the Coach states comes from the record.
The numeric grounding guard rewrites any figure the turn's Grounding Ledger
cannot account for, and it holds: removing the strip reddens the existing
cross-turn ledger test. What is not covered is its activation. On the tool
path the guard runs only when `loop.toolResults` is non-empty, and a tool that
found nothing resolves to `{ present: false, reason: "none" }` with no
`available` payload, which the loop drops. So the guard disarms itself exactly
when the record is empty: ask about a metric with no readings, and an invented
figure is persisted and streamed verbatim, with no withheld-figure marker to
tell the reader it was never checked. The no-tools snapshot fallback is
populated only in the non-tool-mode branch, so the tool path has nothing to
fall back to, and every cloud provider runs the tool path.
Separately, a rollup bucket outlives the rows it was folded from. The read-swap
falls back to live SQL only on an empty band, `ensureUserRollupsFresh` repairs
the DAY tier over the trailing 90 days only, and it keys off
`Measurement.updatedAt`, which a hard delete does not bump. The integration
test reproduces the hard `deleteMany` in the WHOOP body sync against a
MONTH-band bucket: the deleted value keeps reaching the Coach's coarse tail
while the all-time extremes beside it report only the surviving value, and the
block is stamped current because `asOf` is derived from raw reading age and
knows nothing about rollup age.
All three assert current behaviour and carry a docblock saying so. Closing
either gap will redden them, which is the point.
Marc-André Bombeck