Background insights can no longer use up the day before somebody asks a question; a fifth of the daily limit is held back for what a person does by hand.
The delivery ledger is written without awaiting the result, so the row
appears some time after the dispatch returns. The test slept a flat fifty
milliseconds and then read, which holds on a warm machine and loses on a
loaded runner: the assertion reported an empty ledger as though the
dispatch had not recorded the skip.
It polls for the rows now, up to five seconds, and fails with a message
naming how many it found. The claim is that the row is written, which is
what the code promises; that it is written within fifty milliseconds is
not.
Brings in the truncate retry, the socket fix for the record-fence spec that
failed this branch's second e2e shard, and the corrected absence assertions.
The background ceiling applied only when the operator funded the chain, so on a
chain the operator does not pay for the automatic generators and the chat shared
one undivided ceiling and the first could still take all of it. That is the
configuration the recommended remedy for the original lockout produces — put the
personal Codex account first — so the guarantee was missing on exactly the setup
it was written for.
The ceiling is now expressed as what it is meant to protect. Background work may
run until the day's spend would leave less than a fixed reserve for whoever is
waiting on an answer, on both cost owners. A reserve and a quota look alike at
the boundary and mean opposite things in the middle: a quota stops useful
generation on a day nobody was ever going to interrupt, while a reserve costs it
nothing until the moment it would take the last of the day. An account
generating 1.45 million tokens of insights on its own plan keeps every one of
them and the chat still holds 400 000.
INTERACTIVE_RESERVE_SHARE replaces JOB_SURFACE_SHARE. The old name described a
quota over the jobs rather than a floor under the person waiting, and a name
that states the opposite of the intent is what put the wrong comparison in this
ledger twice already.
The fraction is a default an operator will want to change. The setting and its
documentation are separate work.
The suite wipes the database between files with one TRUNCATE, which needs
the strictest lock on every table at the same time. The server writes in
thirty-nine places without awaiting the result, so one of those can land
after the test that caused it has returned, and Postgres then picks the
truncate as the deadlock victim. It failed an unrelated file twice in two
days and passed on the rerun both times.
The truncate is retried up to four times, on that one error and no other,
with a short back-off. Three cases hold it there: a deadlock that clears,
an error that is not a deadlock and must not be retried at all, and a
deadlock that never clears and must still fail. All three go red without
the retry.
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.
Allowlisting `PUT /api/auth/profile` so the setup flow's baseline step could
finish also handed every visitor the shared demo account's contact email, its
display name, its full name and its insurer fields: the allowlist admits a
path and a method, and every field the handler behind it writes comes with
them. The justification beside the entry named three fields; the entry
admitted eleven.
The write is narrowed on the server, in `applyProfileUpdate`, to height, date
of birth and sex while `DEMO_MODE` is set — not in the client, which is the
party the allowlist exists to distrust. It happens ahead of the email conflict
check, which would otherwise answer "is this address registered here?" to
anybody who can open the demo. The display-name box goes read-only there, as
the anamnesis card already is, so nothing accepts input it will discard.
The entry's comment now lists what the handler writes and what is admitted.
The arrival-reaction claim carried its own copy of the reservation upsert so
the write could ride the claim transaction. The copy drifted: it learned to
write operator_tokens and kept gating on the day's mixed total, at the halved
background ceiling. On the row the wave was written for, 1.2 million tokens
served by the user's own plan, the reaction was refused all day at half the old
ceiling, reproducing the defect the wave exists to remove on the one surface it
forgot to convert. reserveBudget now takes the transaction client, so both
writes still roll back together and there is no second copy left to drift.
The row also remembers who paid for its reservation. The supersede refund in
data-arrival has no chain to ask and could only reverse the total, so the
operator share of every superseded reservation stayed on the row forever,
walking the operator's ceiling upward against spend that never happened. A
supersede is routine, not an edge case: any second measurement of the same kind
on the same local date before the worker runs. The resume path re-derived the
owner from a chain the health ledger may have reordered since, which leaks the
same way or drains another request's balance.
A structural guard makes the invariant executable. Every INSERT column list,
SET clause and RETURNING list that names one counter must name the other, and
the set of files holding raw ledger SQL is pinned. Both defects above were a
grep away and both would have failed this test.
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 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.
"Finish without the profile" completed the flow with no record id, and
the route then derived the child's answers onto the guardian's own
record. The server now decides: answers given for somebody else with no
record to apply them to complete and latch the flow without deriving or
seeding anything, whatever the client sent. The checklist carries the
profile instead and reads none of those answers as the guardian's own.
A record id the caller does not guard answers 404 like every
id-addressed managed-profile route, one answer for missing, not managed
and not theirs alike; the contract documents the field, the 404 and the
422 text. Pinned by two integration cases and a second child journey.
"Someone I look after" derived the child's answers onto the guardian's
own record. The completion now takes the managed record the profile
step just created, checks the same access rule the guardian's modules
route applies, and lands the same derivation and dashboard seed on that
record through the same record-keyed write; the guardian's own flow is
stamped complete and latched without touching its map. Three
integration cases pin it, and the child journey reads the child's map
through a real switch — it asserted the guardian's map before, which
pinned the wrong record.
Also from the review: the checklist's visit row read a meta the visits
list never publishes and now counts its body; the first-result tile
read the measurements list as an array and now shows the reading; the
demo allowlist admits the disclaimer and tour writes, with a structural
test over every onboarding route file; restart clears the old first
result, and the screen trusts the ledger before painting a result as
done; the headroom sweep's raw-key check matches the flow's own key
shapes; the demo migration script stops copying the dropped column; the
questions-to-pass rule and the shell's onboarding-path rule are pure and
pinned; the baseline form composes the shared action row; the schema
diff loses the unrelated formatting churn.
Run against the standalone build, the four journeys and the axe sweep
surfaced five things the unit tier could not see. The completion route
derives only once every question is settled, and Q6 stays pending for
somebody with no unit-bearing area — the confirm screen now passes the
questions the flow never showed before it completes. The dashboard seed
matched a JSON null where the unset column is a SQL NULL, so it never
landed; the retired step route carried the same filter. The auth shell's
onboarding branch matched only the front door, so every question screen
rendered inside the full app chrome with the sidebar. Inline links
relied on colour alone, the step list sat outside any landmark, and the
baseline form brought a second h1 onto the confirm screen. The two
sweeps run serially, since each drives one account through the API.
The integration file gains the seed case that would have caught the
null filter.
The guardian's modules branch no longer invalidates the health score in the
route; the record-keyed write it shares with the setup flow does. The guard
pinned the old call site and went red for a behaviour that had not changed.
It now asserts the route goes through the shared write with the record id
and that the helper is still the place that invalidates.
The needs-based flow is the one setup flow now. POST /api/onboarding/step
and the six goal slugs go, and with them the two account columns the
wizard kept — its checkpoint and its goal list — through a migration
that drops them; onboarding_completed_at stays as the first-run
redirect's gate. The one durable effect the goals had, the dashboard
tile order, moves to the completion route: the answers now carry both
halves of Q2, which modules are on and what sits on top, and the seed
keeps the wizard's one contract of writing only while the layout column
is still unset.
The OpenAPI table loses the step path and the completion's description
says what the route does today; the demo-allowlist test names a
mutation that still exists.
The write-level row listed "a vaccination dose and the booster reminder it
suggests" three paragraphs under a section table that puts the vaccination
record in Health background. Read together, that tells a self-hoster a
health-background WRITE grant can plan a booster — which the route beside it
now refuses, because the reminder it mints lives under Readings. The row lists
the dose alone, and the exceptions paragraph, which is where a reader looks for
exactly this, carries the two-section case and says the dose logs either way.
With it, the small things the same pass turned up:
The cycle-insights fixture asserted a property it could not show. Its docblock
claims the cycle-native answer survives the fence, and it seeded no symptom at
all, so `symptomPatterns` was empty in every leg and a regression that emptied
it along with the foreign columns would have passed green. One catalogue
symptom on the luteal days, asserted on the owner leg and on the cycle-only
one.
The immunization rows matched on a bare `role="button"`, which passes vacuously
the day the card becomes a real button and fails for the wrong reason if
anything else in the subtree grows one. They carry the same stable marker the
visit card took.
`/api/cycle/calendar` reads the same wrist-temperature channel the crosstab
does and is deliberately not fenced, and it was the only one of the pair with
nothing saying why: no value from it reaches the response.
The encrypted export's new 403 documented the switch refusal and not the
narrow-scope one it answers first, unlike the two settings operations beside
it. Both sheets say at the top that their Save and Delete are gated at the
caller, so a third mount reads the constraint before it violates it. And one
docblock line ran past the wrap.
Two cases in the integration file answered one or two questions and posted
the confirm, which the gate now refuses; they answer or pass all six. The
audit assertion reads the details column as the JSON string it is.
Four of them. The confirm re-stamped its completion instant on every call, so
a replay quietly moved the moment the setup finished at while the derivation
beside it was carefully latched. The two list answers were stored exactly as
sent, duplicates included, so the column could hold something the reader would
never return — the reader de-duplicates on the way out and now so does the
write. The wide-event name for saving an answer used a different noun from its
two siblings, which makes a dashboard filter awkward for one feature. And the
contract said the restart body was required while the route accepts none.
The account payload resolved the setup state for the actor while its module
map, its cycle flag and every count the checklist reads beside it answer for
the active record. A guardian looking at a child's dashboard therefore got the
child's rows ordered by their own answers, and the child's reading count
tested against their own flow state. The two halves of that component have to
agree about whose record they describe, so the field is record-scoped now,
like the module map next to it.
Under a grant narrowed to one section the answers come back empty, for the
reason the module map is masked: they name the domains the record tracks, and
a grant that opens readings does not open that. The shape of the flow still
comes through.
The three write routes stay actor-only and say why: a delegable declaration is
a reviewed entry in the sharing surface guard, and a read that is wider than
its write is the honest shape while that entry is still to come.
The setup flow writes a record's module map, its glucose unit and its unit
preference. All three have a dedicated settings route that records an audit
row, and the flow recorded none, so "why did my modules change" had no answer
in the activity panel. Restarting the flow did write a row, which made the gap
look like an oversight rather than a decision, because it was one.
Same event names as the dedicated routes, since it is the same column and the
panel should group them. The module row names the keys that actually moved and
the domains kept for their content.
The derivation wrote an explicit off for every module the answers did not
name, which is right for a record being set up and wrong for one with years of
rows behind it. An established self-hoster has, for most modules, no stored
preference at all — they were default-on and nobody ever touched the toggle —
so the first confirm took away the navigation entry, the dashboard widget and
the settings entry for domains full of content. The rows survived; the
surfaces did not.
A derived off now asks first, and only for the modules it would actually
retract. Ownership is not re-answered: the measurement domains come from the
same map the MCP reads and the coach snapshot gates off, and each remaining
module names the one table its content lives in, exhaustively, so a module
added later fails to compile until somebody says where its content is. A
module whose domain holds rows keeps whatever the record already had, because
absence is what "on" looks like in the stored map.
The only gate on the derivation was Q1 being set, so one answer was enough to
turn the whole questionnaire into a module map. Two ways that landed wrong and
they compound. Somebody answers who the record is for, closes the tab, and
later finishes the legacy four-step wizard, which posts to this same endpoint
and knows nothing about these questions: the five unanswered ones each read as
their conservative default, eleven surfaces are written off, and the
once-only marker makes it unrepeatable. Or any client patches one answer and
posts the confirm straight back.
The gate now asks whether every question has an answer or a deliberate pass.
A half-answered flow derives nothing, is not stamped complete, and keeps its
confirm step pending, so finishing the questions later still does the work.
The legacy arm is untouched: a caller with no setup row still answers the
fixed acknowledgement it always did.
The spec's sixth question is skipped when the account already carries the
preferences, and the flow had no way to know: `User.unitPreference` is
nullable with null meaning "default metric", and the account payload coerces
that away, so nothing downstream could tell a chosen metric from an unset
column. The two columns are read raw on the server now, into the same
vocabulary the wire publishes, and a record whose account holds both reads its
`units` step as done — the question answers itself and is never put again.
Read-side only: the stored ledger keeps saying pending, because the person
really has not answered it, and an answer given later still lands.
The two writes the units answer makes — the display columns and the record of
the answer — go in one transaction, so a failure between them can no longer
leave the columns changed with no record of what changed them.
Four claims the unit suite cannot reach, because each is about what is in the
database afterwards: the answers persist and a replay writes nothing new, the
module derivation happens once and leaves a later Settings change alone, a
restart resets the steps and not the modules, and a delegate with a live
accepted grant and a switched session is refused on all three writes — with
neither record acquiring a row, which is the quiet failure a fall-back to the
caller's own record would produce.
The demo guard gains the three paths and, beside them, the case that matters
more: admitting one verb on a path must not open a sibling.
`POST /api/vaccinations/{id}/booster` is declared `("write", "profile")`, and
that is right for the act — filing the Impfpass and arming the booster it
suggests are one thing to the person doing them. The row it writes is not: a
`MeasurementReminder` belongs to the measurements section, every direct route
over the model declares it there, and creating one needs MANAGE. So a WRITE
grant scoped to the health background alone produced a reminder on the owner's
checkup list that the delegate could neither read nor change through the
section that owns it.
The seam is fenced the way the visits service fences a checkup closure, with
one difference stated at the call site: this refuses rather than skips. There
the closure is one effect of a larger save and reporting it as not done keeps
the save honest; here the reminder is the whole request, and a 201 over a row
that was never written would tell the person their booster is planned while
nothing rings. The prompt that raises the call asks the same question client-
side, so a delegate outside the section is not offered it in the first place;
the dose itself still logs either way.
`vaccination-booster-scoped-delegate.test.ts` drives the owner, a delegate
holding both sections, a profile-only delegate and a whole-record one. The
scoped-refusal leg fails against the previous handler, which answered 201 and
left the row.
Nothing recomputes an answer. The module map they produced survives on the
account row, but the map says what is switched on, not that the person told the
flow they take medication daily and have a visit next month — so a restore
without the row opens a blank questionnaire for somebody who already answered
it, and leaves the checklist ordering itself from nothing.
Both ends live in one file, the arrangement the visits and the score already
use, because a reader who greps only the restore route gets a false negative
when the route delegates.
`modulesDerivedAt` is the field that makes this more than history and it is
asserted rather than defaulted: it is the once-only latch on the module
derivation, so a restore that let it default to null would hand the next confirm
permission to re-apply the questionnaire over every module decision taken since,
silently. The round trip seeds it non-null and reads it back.
The file's schema validates the three answer objects loosely and the restore
narrows them through the same fail-soft readers the account payload uses. Being
stricter there would refuse a whole account's restore over one retired chip.
The wipe takes the row for the reason the reset already takes the wizard's own
columns: the record the answers were about is gone, so the answers describe
nothing.
The three routes that act on the whole of an account — `DELETE
/api/settings/data`, `DELETE /api/settings/account`, `POST
/api/export/encrypted` — name no record. They resolve the caller and act on
whatever the caller owns, so the question a switched session puts to them is
what happens when the browser believes it is somewhere else.
They already refuse: all three resolve through `requireFreshMfaIfEnrolled` or
its erasure sibling, and both open with `requireAuth()`, which reads the acting
carrier and throws `sharing.not_permitted` for anything but `none`. What was
missing is that the contract never said so. `/api/settings/{data,account}`
published a 403 about narrow Bearer scope alone, and the encrypted export
published no 403 at all, so a client that switches records had no way to learn
which of its calls stop working short of trying one.
One sentence, spliced onto the three, byte-identical for the same reason the
delegable refusal is one string. `switched-session-destructive-refusal.test.ts`
drives each route with a live whole-record MANAGE switch and asserts both
halves: the status, the code and the wording, and that neither record moved —
a refusal raised after the transaction would satisfy the first alone. The
export leg additionally asserts no archive rode along with the 403.
`GET /api/cycle/insights` is declared `("read", "cycle")`, which is right for
the phase labels and wrong for what it compares them against. The crosstab rows
carry `lutealAvg`, `follicularAvg` and `delta` in display units for resting
heart rate, HRV, sleep duration, steps, weight, body/wrist/skin temperature and
blood glucose, plus the mood score read straight off `MoodEntry` — measurement
and mind data, published in full to a grant scoped to the cycle alone, with the
lagged discovery's narrated interpretation on top.
The visits service already had the right shape for this: a predicate over the
grant's sections, asked before the foreign rows are read. Both sections are now
fenced at the query rather than in the response, so a caller outside them never
loads the values in the first place and the rows read exactly as they do for a
record with nothing logged. The wrist-temperature read that sharpens the
prediction stays: it never reaches the response.
The predicate had two verbatim copies, one in the visits service and one in the
vaccination service, with a comment on the second saying a second spelling would
be a second thing to keep in step. It is one module now, `sharing/acting-domains`,
and the third consumer imports it rather than a feature service.
`cycle-insights-scoped-delegate.test.ts` drives four grants against a fixture
whose separation is large enough that the FDR gate cannot be the reason a row is
missing: the owner and an entire-record delegate get both figures, a cycle-only
delegate gets neither and no lagged row, and a cycle+mind delegate gets the mood
column and not the weight. Both scoped legs fail against the previous handler.
On the release that adds the ledger the table is empty, so on a host that has
been uploading nightly for months every account read Never, under a sentence
saying no copy had ever reached the bucket. That sentence is untrue on exactly
the morning an operator opens the new card for the first time, and it is a
claim the host cannot make: what is empty is the record, not the bucket.
The run now records every account it walks, not only the ones an object came
of. A walk that produced nothing moves the attempt instant and leaves the
success columns alone, so the two cases separate: no row at all is no record
yet, a row without a success is never. Never now means what the word says —
a run looked at this account and has never put anything there.
The card names the last check on the never row, so the state is a fact with a
timestamp rather than an assertion. Runbook and all seven bundles carry the
same two sentences. The integration file proves both arms against real
Postgres, including that a failed run leaves the previous good copy on the
row.
The threshold was the schedule period with no budget on top of it. Two things
spend that budget on a host where nothing is wrong. The cron runs 02:30
Europe/Berlin, so on the DST fall-back Sunday the gap between two runs is 25
hours and every account on every host would cross one period on the same
morning. And the ledger instant is when that account's object landed, by
design, so on a cohort walked one account at a time an account reached an hour
later tonight than last night drifts past 24 with two clean runs behind it.
The schedule period and the verdict threshold are now separate numbers. The
period is still what the card quotes; the threshold is the period plus six
hours, applied to both steps, so due still means one run produced nothing and
stale still means two. The fall-back night is pinned in the unit file by its
real dates rather than by an arithmetic 25, and a genuinely missed night at 31
hours still reads due. The runbook says why the slack is there.
The column was INTEGER. A single object may occupy 8 MiB across 10 000
multipart parts, which is the 80 GB the uploader explicitly permits, and
Postgres INTEGER stops at 2.1 GB. An account between the two uploads fine and
then cannot be recorded, so the biggest account on the host is the one the
console would have to report on and could not.
BIGINT in the table, BigInt in the model, Number() where the route builds the
wire row so the payload type stays a plain number. The migration is edited in
place rather than followed by a second one: the table has not shipped, and a
column widened before its first release is one column, not a history.
The comparison with the in-database backup does not carry over, for the
record: that size is octet_length over a bytea Postgres caps at 1 GB, while
this object never touches the database at all.
The unit tests either side of this mock Prisma, so neither proves the
migration applies or that the writer and the reader agree on a shape. The
integration file runs the real nightly job against real Postgres with an
in-memory bucket, then reads the row back and asks the console's own
classifier what it says: fresh after a run, one row across two runs, stale
once it ages past two nightly periods, gone when the account is.
The component test pins the case an operator most needs stated: a host with
no bucket configured says so, rather than painting an empty list that reads
as "no problems". The accessibility scan's admin-backups response gains one
row in each state, because a card that paints only its header proves nothing
about the rows anybody reads.
Two small corrections ride along, both from writing the tests. The freshness
badge composes its classes through `cn` like every other surface, and the
error-reporting line waits for the settings to arrive instead of asserting
"nothing is sent" while the answer is still loading.
Two more creates the client replays out of its offline outbox ran the
handler again on the replay. A titration step carries no natural key at
all, so a lost success response left the medication with two dose
changes on the same day and the runway estimate reading off the wrong
one. A custom cycle symptom is unique on its label, so the replay did
not duplicate — it answered the 409 that uniqueness raises, reporting a
conflict the client cannot act on for a row it has no record of
creating.
Both wrap in `withIdempotency`, the composition their siblings already
ship, and both operations publish the header and the in-flight 409 so a
client generated against the contract can tell a retry is safe.
The integration file that drives the medication creates twice against
real Postgres gains the titration case: one dose-change row after the
replay, and a replayed 201 carrying `X-Idempotent-Replay: true`.
`POST /api/mcp/tokens` is deliberately not on this list. Its response
carries a freshly minted `hlk_` credential, and the wrapper refuses to
persist a secret-shaped body — publishing the header there would
promise a replay the mechanism will never perform.
A disaster-recovery snapshot carries the singleton `app_settings` row —
registration, the MFA requirement, the default locale and timezone, module
availability, the notification and AI configuration, the document size cap and
quota — and the restore wrote it back as a side effect of restoring one
account. An operator putting a single record back on Wednesday reconfigured the
whole host from Monday's file, silently, with nothing on screen saying so.
Rebuilding a host from a snapshot is a real case, so the write stays: it is now
`restoreInstanceSettings: true` on the request and a checkbox beside the typed
gate, off unless the operator ticks it. Restoring an account and reconfiguring
the installation are two decisions, so they are two answers.
The 422 confirmation sentence and the dialog copy drop the instance-wide claim;
the opt-in carries it instead, in all seven catalogues. Integration coverage
runs both arms: without the flag the settings row keeps its own values, with it
the file's values land.
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.
The request-scoped capture kept the last verdict of any kind and fell back
to it when nothing had refused. That fallback fires in exactly one state —
no limiter refused this request — which is the definition of "this 429 came
from somewhere else". So a daily AI budget that is spent until midnight, an
hourly generation quota, and a provider-side 429 relayed onward all went out
carrying our own bucket's numbers: a remainder that says there is room left
and a Retry-After measured against an unrelated window. On the AI paths the
captured window can also expire while the provider call runs, so the dressed
delay could land at zero and tell the client to retry immediately into the
ceiling that had just refused it.
The cell now keeps refusals only. A request every bucket let through reports
nothing, and a 429 from another cause leaves undressed, which is correct:
there is no bucket to describe. The two-limiter rule is unchanged — the last
refusal wins, and a refusal still outranks a pass.
The unit case that pinned the fallback is turned round, and both directions
now run through the real wrapper: a limiter refusal gets the headers, a 429
raised without one gets none. The integration file gains the two shapes it
was missing — a route that consults two buckets must report the one that
refused, and a route whose 429 comes from the daily budget must carry no
rate-limit headers at all.
`GET /api/auth/me` publishes a boolean per module, and three different
situations collapse into the same `false`: the record has the module switched
off, the active grant's sections do not open it, or the operator switched it
off for the whole instance. A client holding the boolean has to guess between
them, so the empty state either says nothing or says something that may be
untrue -- offering a delegate a switch that is not theirs to reach, or telling
an account holder to turn on a module the operator removed.
`moduleAccess` carries the reason beside the booleans, which keep their exact
meaning: `enabled`, `disabled`, `not_granted`, `unavailable`, resolved
outside-in so the reason published is the one nothing further in can change.
`modules[key]` stays exactly `moduleAccess[key] === "enabled"` for every key,
and the two halves are built independently -- the boolean still runs through
`maskModulesToSections` -- so the test of that pairing is an assertion rather
than a restatement of a derivation. It is checked on the pure builder across a
matrix of grants, records and operator blobs, and again on the real route for a
scoped delegate, an unscoped one, and an operator-disabled module.
Both ends in one change, because a payload field with no reader is the failure
this repository keeps rediscovering. The reader is `ModuleDisabledNotice`, the
shared module-off empty state the empty-states audit asks for: it names the
reason in the reader's own language and offers the enable control only when the
record's own switch is what is off, since neither of the other two is fixable
from that page. The nutrients page adopts it and drops its bespoke block, which
had been handing every delegate a button that would refuse them.
Registered in the OpenAPI registry as `ModuleAccessMap` and carried on the
account payload; the native contract gains a field and loses nothing, since a
Bearer request carries no switch and never reads `not_granted`.
Response drift was structurally undetectable. `openapi:check` proves the
YAML matches the registry and the route-coverage guard proves the verb is
listed; both pass when the registered schema disagrees with what
`apiSuccess()` sends, and the coverage guard says so in its own doc
comment. That is how `/api/sync/changes` shipped five domains against a
three-domain contract and how the error envelope went years without
declaring the field the multi-issue 422 was built for.
Two bodies, the two that were broken: a full `/api/sync/changes` page
with live and soft-deleted rows in all five domains, its `cursorExpired`
short-circuit (a second shape under the same schema), and a 422 carrying
more than one issue.
`safeParse` alone would not have caught either finding, because Zod
strips undeclared keys instead of refusing them — a schema short of the
wire parses cleanly. So the parsed value is compared back to the raw
body; a strip means the contract is missing something the server sends.
Removing `cycleDays` / `cycles` from the sync schema, or `details` from
the envelope, fails all three cases.
`/api/measurements` and `/api/labs` sort on one non-unique column and
page with `skip`/`take`. When the sort key ties, Postgres does not
promise the same order for the query that fetches page N and the one
that fetches page N+1, so a client paging its own history can receive a
row twice and never receive another — silently, on both sides.
Ties are not hypothetical here. The partial unique index forbids two
live measurements of the same type at one instant, but the tie that
occurs in production is the cross-type one: a daily `stats:` rollup
posts steps, distance, flights and active energy at the identical
timestamp, and an unfiltered history list pages straight through the
block. A lab panel drawn in one sitting shares `takenAt` across every
analyte in it.
Appending `{ id: <same direction> }` to the `orderBy` makes the order
total. Same fix on the five siblings that page the same way — medication
intake history, illness day-logs, custom-metric entries, reminder
history, mental-health assessments — because the defect is the shape,
not the two routes it was found on. No wire change; `/api/sync/changes`
already ordered this way.
The fixtures seed ids in an order unrelated to insertion order, so the
assertion on order inside the tied block can only hold if the secondary
key is doing the work: with the tiebreaker removed the new file fails on
all three cases.
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.
Marc-André Bombeck