The comprehensive insight that feeds both the daily briefing and the
insights trend narrative is a single reasoning-heavy generation over the
full feature snapshot. On a large account it runs past the provider
client's fixed 60 s timeout and is aborted mid-stream, so the chain
reports an all-failed outcome, no cached block is written, and the
surface falls back to its static placeholder on every visit. The warm
paths bounded the same generation tighter still (45 s) and counted the
failure with no observable signal, so the blank state was permanent and
silent.
Make the per-request upstream timeout configurable
(CompletionParams.timeoutMs, default 60 s for every existing caller) and
raise it to 120 s for the comprehensive surface only, across all three
generation paths: the on-demand regenerate, the on-demand warm, and the
nightly pre-generation. Raise the warm-pass budget to 130 s so the
provider's own timeout governs a genuine completion rather than clipping
it early. Annotate the comprehensive failure in both the generator and
the nightly loop so a generation that keeps failing is queryable instead
of leaving the surface empty without a trace.
The dedicated conversation-history page gates on the operator master
flag and the per-user opt-out like the main Coach route, so add it to
both grep-based gate-site allowlists.
Drop the unused extractCoachClaims helper (+ its CoachClaim types) and the
unused CoachPlanStatus type — the deterministic grader uses the boolean claim
helpers directly, and the plan status enum is consumed via the const array.
Update the dashboard-snapshot suite for the locale-keyed cell: stub the i18n
context and add getQueriesData to the fake query client so the locale-aware
preload probe is exercised.
A Codex backend 400 on the richer Coach tool-call request (and the daily
briefing) propagated out of the provider chain runner unwrapped: a 4xx other
than 401/403/429 was not classified as a hard provider failure, so it bypassed
AllProvidersFailedError and rethrew from the chat route as an HTTP 500. Plain
status-card completions, which never send a tools array, were unaffected.
Harden the failure path so a provider error can never surface as a 500:
- Treat any 4xx a provider wire returns as a hard failure so it cascades and is
wrapped in AllProvidersFailedError. InsightSchemaError (the strict-JSON 422
surface) stays the only non-hard class, so that diagnostic is preserved.
- The chat route now also classifies a tagged provider error that reaches its
catch unwrapped and emits the graceful coach.provider.* SSE frame instead of
rethrowing. Only an untagged error keeps the 500 + error-reporter path.
Make the Codex 400 diagnosable: fold the redacted upstream response body into
the thrown error message and the chain wide-event, so the field the backend
rejects is visible in logs (it was previously dropped).
Fix the Codex request shape behind the tool-call 400:
- Send the required strict flag on each Responses-API function tool.
- Replay prior assistant turns as output_text, not input_text, per the
Responses API multi-turn contract — the violation only surfaced on round two
of the tool loop, which is the path that broke.
Codex-only; the OpenAI, Anthropic, local and mock providers are untouched.
A deterministic golden-set + red-team floor that gates every Coach change
offline, plus an opt-in live-judge layer that scores warmth and groundedness
with a pinned model only when COACH_EVAL_API_KEY is set.
- 42 golden cases across grounding, cross-metric, data-honesty, provider-parity
and own-baseline taxonomies, graded by behaviour-anchored binary criteria.
- Structured-claim grounding helpers extend the numeric verifier from numbers
to threshold / own-baseline / confident-verdict claims; the numeric API is
unchanged so the existing grounding tests stay green.
- Red-team battery asserts the existing inbound refusal, outbound dose screen,
acute red-flag escalation and learn-link guards fire.
- Nightly + manual workflow runs the deterministic floor always; the live judge
step runs only when the secret is present and is non-blocking.
Add a CoachPlan model: durable, user-scoped if-then implementation plans
the Coach proposes and the user confirms. The background memory-refresh
pass extracts plans the user agreed to and writes them as "proposed"; only
the user-facing PATCH on /api/coach/plans/[id] activates one. The free-text
cue / action / target are encrypted at rest with the shared Bytes codec;
metric, status and dates stay plain so the injection picker ranks without a
per-row decrypt. Active plans fold into the Coach snapshot's memory block
and survive the prompt-budget drop alongside facts. List, confirm and
soft-delete routes are Coach-gated and narrow the owner from the session.
Key the dashboard-snapshot query cell by locale so a locale switch reads
freshly localised tile copy on its own cell instead of serving the prior
locale's prose until the slot goes stale. The zero-arg key stays
byte-identical, so every prefix-based invalidation keeps reaching the
locale-keyed cells; the offline-banner and preloader match on the prefix.
Migration 0194 (additive CREATE TABLE). Encryption-rotation registry and
script cover the three new Bytes columns.
The A1–A3 source files carried v1.22.0 docstrings from before the work was
re-cut as the v1.21.2 point release. Comment-only sweep so blame and the
in-file version tags read consistently with the shipped version.
The builder is server-only by construction — only the metric-page route
imports it and it reads Prisma — and no other module in the tree depends on
the `server-only` package, so the import was both redundant and an undeclared
dependency. Removing it keeps the dead-code gate clean without weakening the
boundary.
The hero score card and the daily briefing accept tension, returnToBand, and
memory props, but nothing populated them server-side, so the lines never rendered.
Resolve all three server-authoritatively on the dashboard snapshot DTO (iOS rides
the same payload):
- tension — fires when the readiness composite's contributors disagree (a strongly
favourable one alongside a strongly unfavourable one), reusing the readiness
components the wellness ring grades and the shared disagreement detector.
Suppressed under a clinical red-flag (a coincident-deviation that no illness
explains) so the red-flag path is never reconciled into a calm verdict.
- returnToBand — runs the streak detector across the salient vitals and surfaces
at most one metric that has come back inside its own personal range after a
genuine prior out-of-band run.
- memory — maps the machine-derived coach memory block to the briefing's recall +
forward-look.
The blocks are locale-agnostic (contributor keys + a metric type); the insights
hero localises them through the existing readiness-contributor and metric-name
labels before handing the cards their resolved strings. Each value omits rather
than guesses. The snapshot cache key now carries the locale for the recall prose.
Add a deterministic narrative layer the Coach narrates rather than invents:
- Briefing callbacks + forward-looks: the daily briefing recalls the prior
period and points ahead, anchored to a figure, one callback per surface.
- Tension Verdict: when the readiness composite's contributors disagree
(good sleep but a rising resting pulse), surface a brief honest internal
read on the health-score card. A real clinical red-flag is never reconciled
away into a calm verdict.
- Streaks + return-to-baseline: a pure detector over a DAY-bucket series
closes a worry once a metric returns inside the personal MAD band after a
genuine out-of-band run; a gap day breaks a streak.
- Coach prompt: add develop-discrepancy, roll-with-resistance, and an
anti-persuasion clause; ratchet PROMPT_VERSION.
The tension flag and streak event are computed server-side; the card lines
render resolved fields and the model narrates them.
Replace the hidden composer prefill with a visible scope affordance: when
the Coach is launched narrowed to a metric, the hero shows a scope pill
naming that metric plus a tappable seed question, and the drawer sources
rail surfaces the same scope line. When the Coach opens unscoped, resolve
today's most notable derived signal server-side and offer it as a tappable
opener; fall back to the neutral greeting when no signal crosses the gate,
so the blank box is never cold and never fabricated.
Add an internal GET /api/insights/coach/seeded-question that reuses the
briefing's derived-signal detector to select the opener server-side; the
client renders the resolved DTO.
Render a compact two-line "Coach read" above each metric sub-page chart:
an own-baseline line ("Your usual range is X–Y; today's Z sits within /
above / below") and one top lagged-association line from discovered
correlations, framed descriptively. It makes a metric read seen-in-context
rather than isolated.
The strip is server-authoritative: a new GET /api/insights/coach-read
resolves the band via the median ± k·MAD baseline engine (with its 7-day
history floor — below it the strip says "still learning your range" rather
than inventing a range) and the strongest on-metric driver from the
existing FDR-controlled, effect-size-floored correlation discovery (no
lowered floor; the line is omitted when nothing clears it). The client
component only renders the resolved DTO, so web and iOS read identical
numbers. DTO shapes and the pure selection logic live in a client-safe
sibling shared by the component and the unit test.
Wired through the HealthKit metric scaffold (every generic metric page) and
the bespoke weight, blood-pressure, and pulse pages.
The post-hoc numeric verifier only graded the tool path, so local /
no-tools providers got no grounding net. Grade the no-tools prose
against the snapshot the model was shown — its structured sections,
which already carry the correlations-snapshot block — and flag any
cited number that traces to no snapshot leaf. Non-blocking: the count
is annotated under coach.prose.number_unverified and the unverified
token soft-stripped, never a stream-killer. Only runs when the full
figures were delivered this turn; a cheap follow-up has no fresh set
to grade and no-ops. Tool-path behaviour and tolerance unchanged.
The three intake dialogs (log, edit, dose-history add) still used a raw
datetime-local input, so they ignored the user's date-format preference.
Route them through DateTimeField like every other date surface. Also
correct a stale diastolic-floor comment in the dashboard verdict.
The correlation discovery matrix gained two non-MeasurementType channels
(medication compliance, symptom severity), but the Coach get_correlations
reader still filtered its measurement query by hand and never built the new
channels. The channel keys leaked into the Postgres enum cast and threw for any
user with medication or illness data.
Extract the route's two channel-series fetchers into a shared module
(correlation-channel-series.ts) so both consumers build the channels identically,
then wire the reader to derive its type filter from discoveryMeasurementTypes and
fold the fetched compliance + symptom series into the discovery input. The reader
now humanises the channel keys to "medication adherence" / "symptom severity".
Render daily medication-compliance rate and illness symptom-severity as
first-class series in the FDR discovery engine, so the adherence-dip →
symptom-flare link (and compliance → vital drift) can finally surface.
Compliance pools every active medication's dose-history ledger into a per-day
taken/(taken+missed) rate, re-keyed to the user's display timezone. Symptom
severity reads the illness day-log functional impact, zero-filling healthy days
only across real episode spans so a user with no episodes yields an empty
series rather than a constant. Both flow through the existing n >= 20, BH-FDR,
effect-size, and shrinkage gates unchanged; a sparse series degrades to absent.
Each channel forms its own metric family (no self-lag tautology) and keeps the
association-not-causation framing. Add discoveryMeasurementTypes() so every
caller drops the non-measurement channels before a Prisma type filter.
The labs block already surfaces value + unit + range + date, and the Coach
prompt instructs it to quote them verbatim, so labs are cited end-to-end — no
gap. Pin that with a tool-level guard asserting value and unit survive
get_labs, so a future change to the labs pass-through can't quietly strip the
number the Coach needs to cite.
The Coach may point at a published /learn guide, but only a slug in the
catalog. That was a prompt instruction with no enforcement — a model could
hallucinate /learn/lower-your-cortisol and ship a dead link. Add a deterministic
post-filter on the assembled reply that keeps a published slug verbatim and
strips any reference whose slug is not in the catalog, tidying the prose around
the removed link. Makes the catalog's impossible-by-construction claim an
enforced guarantee.
The illness-recovery tool and the illness context block carried only labels,
lifecycle, and dates plus the recovery composite — never the recovery-gap, the
metric that drove it, the nadir, the pre-onset deviations, or the red flags the
illness card renders. So a recovery question got the composite, not the gap the
user sees, and a sustained-fever or low-SpO2 escalation was invisible
in-conversation.
Run the existing episode-correlation read-layer (the same engine the card and
the red-flag notifier use) for the most relevant episode — active first, else
the most-recently-resolved — and attach the computed scores to get_illness_recovery,
coverage-gated so a thin signal yields nothing rather than a fabricated number.
Document the illnessScores shape in the Coach prompt so it restates those
numbers verbatim and escalates a red flag rather than reassuring.
Each retrieval tool re-scoped the snapshot to a single source, so its cache key
never matched the inventory's full-source build — a four-tool turn paid four
extra full snapshot builds on top of the inventory's. Thread the inventory's
probe scope to every tool so a per-tool read lands the one cache entry the
inventory already primed, collapsing the turn to a single build (a window
override still gets its own correct build). Raise the loop's round cap from two
to three so a sequential cross-metric why-chain isn't starved; the per-round
token budget is unchanged and the absolute ceiling tracks one above so the
final round still forces prose.
The tool-mode path dropped the launch scope's sources: the inventory probes the
full source set, so a Coach opened from a metric page saw the whole inventory
and could roam, while only the no-tools path honoured the narrowing. Inject a
one-line FOCUS hint naming the launched domain(s) into the tool-mode user turn
so the model prioritises that metric and fetches its figures first, with every
other domain still reachable if the question leads there. Empty on a generic
open, so the prompt prefix is unchanged on that path.
The per-bucket OLS accumulators stored x on the raw epoch-day axis (x ~ 20540),
so sum_xx accumulated past ~1e10 and the square shed ~10 of a double's
significant digits before the value was ever stored. The sub-day x detail was
lost at accumulation time, which no read-side identity can recover; an
ill-conditioned window's composed slope/r2 then drifted from the live REGR_*
probe past the parity gauge.
Rebase x to a fixed origin (epoch-days of 2020-01-01) at write time so the
squared terms stay O(1e7) and exact. Slope, r2, and population sd are invariant
under the affine x-shift, so the live raw-epoch probe still parity-matches the
cross-bucket compose; composeRegression needs no formula change. The per-bucket
slope/r2 columns stay on the raw axis (shift-invariant).
Migration 0193 recomputes sum_x/sum_xy/sum_xx from measurements on the rebased
basis, one set-based update per granularity. Recompute-from-source is idempotent
and rerun-safe; it touches only rows that already carry the accumulators and
leaves pre-migration NULLs for the boot re-fold.
Un-quarantine the DST-boundary parity test: it now matches live REGR_* within
the existing 1e-9 relative bound, with the WEIGHT case still bit-identical.
The acute red-flag clause hardcoded its blood-pressure numbers as prose
literals and named no glucose or fever floor at all, so the Coach could
silently drift from the dashboard hero and the notification engine. Compose
the systolic/diastolic, glucose, and sustained-fever thresholds straight from
clinical-floors.ts and echo the previously-missing glucose and fever lines, so
every acute number the Coach states is bound to the one source of truth.
Add an outlook contract (gentle forecast, what-to-expect, anticipatory if-then
— all conditional, ranged, association-framed) composed beside the tone
contract on the Coach and the comprehensive briefing, so a reply can sharpen
expectations without a false promise. Tighten the tone contract against
over-validation (affirm the effort, not an unsafe choice; stay neutral when
there is nothing to praise) and route why/conflict questions through
get_correlations and its coincident-deviation flag before answering.
Coverage asserts the clause numbers equal the constants and the new contract
reaches both surfaces.
Promote the static /learn guide catalog into a typed concept-to-slug
lookup the deterministic UI surfaces can consume. learnUrl is the only
sanctioned /learn URL builder and learnLinkForMetric fails closed for
an unmapped id. A test asserts every mapped slug resolves in the
catalog, so the mapping can never point at a guide that does not exist.
The SpO2 sustained-low scan passed episodeDays straight to runFlag, relying
on the reader to deliver them in day order, while the fever path already
sorts its unioned series. The run scan is calendar-consecutive, so feed it
chronological input by construction: sort the filtered SpO2 points before
the scan, matching the fever path and removing the latent ordering coupling.
Both lastStableReturn and computeSymptomReturn counted the in-band settle
run by array-index adjacency instead of calendar days, so three sparse logs
spread over weeks could register a stable return and move the headline
recovery gap. Extend the run only while the prior point is in-band and the
calendar day immediately before, matching the runFlag rule, bounding the
run's span to its logged-day count.
The slot-dedup pass ran a findFirst (latest takenAt) plus a findMany
(all live rows) per medication — 2N round-trips, 40 for a 20-med user.
Pull every live intake row for the user's medication set in one ordered
findMany, group in memory by medicationId, and derive each medication's
lastIntakeAt (max non-null takenAt) from the same grouped slice the
findFirst used to scan. The read cost is now one query regardless of the
medication count.
Behaviour is unchanged: the grouped rows arrive in the same
scheduledFor-asc order the heal and snap passes depend on, and the
derived lastIntakeAt matches the old desc-ordered findFirst.
The activity sync ran a per-(entry, field) findFirst followed by an
update or create, roughly entries x 3 x 2 round-trips — about 2,190 on
a 365-day backfill. Flatten the entry/field grid into the concrete rows
the sync wants to land, resolve every overwrite-vs-insert decision in
one findMany keyed on the same four dedup columns, then run a single
createMany for fresh slots and one transaction for the changed-value
overwrites. A backfill is now a bounded handful of queries regardless of
the day count, and an unchanged re-sync is a no-op on the write path.
The overwrite contract is unchanged: an existing WITHINGS row for a
(type, measuredAt) slot keeps its row and only rewrites value; a fresh
slot inserts with the stable externalId. createMany uses skipDuplicates
so a concurrent sibling sync landing the same slot is tolerated by the
four-column unique.
The rollup read path windows the 7/30/90-day regression on
startOfUtcDay(now - N days), while the live-SQL fallback filtered the
REGR_* aggregates on the wall-clock NOW() - INTERVAL bound. An account
crossing a coverage boundary (warm to cold) therefore saw the boundary
day's readings flip in or out of the slope window depending only on
whether the rollup tier was populated, returning a cache-dependent
slope for the same request.
Anchor both cold-fallback paths on date_trunc('day', NOW() AT TIME ZONE
'UTC') - INTERVAL 'N days', the SQL twin of startOfUtcDay, so warm and
cold return identical boundary-day membership. The avg7/avg30/median
windows stay wall-clock since the warm narrow query computes them the
same way.
The FDR cross-metric driver pairs were reachable only through the
get_correlations tool, so a local/no-tools provider got the coincident
flag but zero discovered drivers and the cross-metric layer silently
degraded. The system prompt's fallback to a snapshot driver field
pointed at a field that did not exist on that path.
Attach a bounded top-3 slice of the same gated, ranked driver output the
tool serves to the no-tools snapshot floor, gated on the derived layer
and shed first under prompt-budget pressure. Local users now reach
parity on the flagship cross-metric capability.
Significance alone surfaced trivial drivers: a large-n pair with r=0.16
narrated as a confident "tends to go with" link while explaining ~2.5%
of variance. Add three gates before a pair is ranked as a driver:
- exclude same-metric-family lagged pairs (mood->mood, BP component
self-lag) as serial-autocorrelation tautologies, widening the existing
exact self-pair skip to the whole family.
- shrink each Pearson estimate toward null by n/(n+10) so a thin-data
correlation cannot out-rank a deep one on an inflated point estimate.
- floor the shrunk effect at 0.2 (drop below) and bind 0.2-0.3 to a
hedged "faint hint" phrasing tier, 0.3+ to confident phrasing scaled
by sample depth. The reported r and p stay the honest raw statistics.
Rank by shrunk effect magnitude, q as tie-break.
The coincident-deviation read grouped the latest day with a UTC date
slice while the sibling readiness read already used the tz-aware day key.
For a user east or west of UTC, a late-evening or early-morning reading
landed on the wrong calendar day, so a fired flag could compare a vital
from the wrong day against its band and narrate '2 or more vitals out of
band today' on the wrong day.
Thread the user's timezone into the latest-day read and mint the day key
with userDayKey, matching readiness. The Coach derived-snapshot and the
correlations reader pass the real account timezone through. Adds a
regression covering a UTC+2 user whose late-night reading rolls to the
next local day.
The hypertensive-crisis floors were defined in three modules with a
divergent diastolic value: the notification engine and the Coach acute
clause used 180/120, but the dashboard hero used 180/110. A reading like
170/112 lit the red critical-BP banner on the hero yet never tripped the
notification alarm or matched the Coach's stated acute number — one
surface said crisis, the other two stayed calm on the same row.
Promote the absolute floors into a dependency-free clinical-floors leaf
and have the hero, the safety-floor engine, the status registry's fever
band, and the illness escalation all import from it. The crisis diastolic
floor is the guideline-correct 120 (ACC/AHA hypertensive urgency); the
wider 110 hero net is dropped so the surfaces tell one story. A coverage
test pins that every consumer resolves the same constants.
The daily AI-token gate was a flat 25,000/day for every provider. That cap
exists to bound the OPERATOR's API bill, but a ChatGPT-OAuth (Codex) or BYOK
chain egresses on the USER's own plan/key and costs the operator nothing, so
gating it on the operator-cost ceiling is wrong. A single gpt-5.x reasoning
turn legitimately reports 20k-40k gross total_tokens (re-sent system prompt +
inventory + tool defs per round + hidden reasoning, summed across rounds), so
one or two turns exhausted the flat cap and locked the user out of a plan they
pay for.
Classify the chain's cost owner by its primary provider via resolveDailyCap:
an admin-openai primary (operator pays) keeps the operator-cost cap; every
user-egress primary (codex/openai/anthropic/local) gets a generous
abuse-only ceiling. Threaded into the Coach chat gate and both OCR-extract
modes.
Raise the operator-cost cap 25,000 -> 200,000 so the operator-key path also
survives a normal day of reasoning turns, keeping gross-token accounting
rather than reworking the reconcile math.
Subtract cached input tokens at reconcile: the Responses-API gross total still
includes prompt-cached input the user did not re-pay for. The codex client
already parsed cached_tokens; sum it through the tool loop and bill
total_tokens minus cached.
Marc-André Bombeck