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.
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.
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.
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.
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.
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.
Marc-André Bombeck