Clicking a type on the measurements list now drills into that metric's
insights sub-page when one exists, falling back to the existing
?type= filtered list when it does not. A reverse type-to-slug map backs
the navigation, preferring the dedicated single-metric page so an
ambiguous type resolves correctly (PULSE to pulse, not blood-pressure;
WEIGHT to weight, not bmi). No API change.
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.
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.
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 no-key and timeout fallback text carried zero reference to the user's
own readings and, on a provider timeout, rendered as hasProvider:true —
indistinguishable from a fresh assessment. Compose each fallback from the
per-metric signal the card already builds: name the current value, place it
against the user's own baseline, and close with one plain-language pointer,
in the same warm voice as the model surfaces. Degrade to the prior generic
tip only when a metric has no usable history. The timeout envelope now
reports hasProvider:false so the UI surfaces it as the computed summary it
is rather than mislabelling it as provider prose.
The floor reads Date.now() inside buildComparisonSnapshotForUser, so it
lands in [before - 400d, after - 400d] relative to the test's brackets.
The prior upper bound used before - 400d, which fails when a millisecond
ticks between the bracket read and the snapshot's clock read.
buildComparisonSnapshotForUser read a user's entire measurement history
per type with no date floor, then fed each series to summarize(). The
snapshot only consumes avg30 / avg30LastMonth / avg30LastYear, and the
widest of those reaches into the [365, 395)-day window (ageMs < 395 days
in meanOfWindow), so nothing older than 395 days can change a result.
Add a 400-day measuredAt.gte lower bound to every per-type read in both
copies of the builder — the page-blocking POST /api/insights/generate
handler and the shared comprehensive-generate path the nightly
pregenerate cron drives. The fan-out runs all seven snapshot types via
Promise.all, so on multi-year accounts the unbounded scan was tens of
thousands of rows per type. 400 days keeps a 5-day floor over the
strict-less-than 395-day reach.
Store additive per-bucket regression accumulators (sum_x, sum_xy, sum_xx,
sum_yy) so windowed slope, r2 and standard deviation compose from the
rollup tier instead of a live raw-table scan on every warm request
(migration 0190, backfilled by the boot queue). Median and the exact
anomaly count stay on a minimal live query. Parity with the live REGR_*
path within float tolerance, well below the read rounding.
Replace the CompletionParams { systemPrompt, userPrompt } string pair with a
structured { system, messages[] } request across all five AI clients. One
refactor unlocks three things: provider prompt-caching (Anthropic cache_control
on the stable system prefix, OpenAI/Codex automatic prefix cache, with
cachedInputTokens surfaced for observability), the tool-calling wire surface,
and clean multi-turn.
Each client maps the new shape to its own wire and degrades unsupported
features silently: the local OpenAI-compatible server never receives a tools
field (most reject unknown keys), and prompt-caching is best-effort everywhere.
CompletionResult gains toolCalls, finishReason and cachedInputTokens; the tool
types and per-provider mapping ship now, the tool-use loop does not.
Vision inputs fold into the user turn's content parts via singleUserTurn, with
each provider's pre-existing block ordering preserved. The corrective-retry
paths append to the last user message rather than a flat string. The Coach
keeps assembling one user turn so its snapshot-once and grounding behaviour is
unchanged. No wire schema visible to API consumers changed.
Behaviour is unchanged across every surface; the full suite stays green.
The summaries slice (computeFromRollups) and the comprehensive aggregator
(buildFromRollups) each ran a narrow live $queryRaw over the raw
measurements partition on EVERY warm-rollup request to fold slope7/30/90
and r2_7/30/90 via REGR_SLOPE / REGR_R2 - the un-eliminated live scan on
the happy path.
Drop the six REGR_* columns from both narrow queries and compose the
windowed slope / r2 (and the population sd via composeRegression) from
the per-bucket accumulators stored in measurement_rollups. The 90-day
DAY buckets are loaded once (already loaded in comprehensive; one bounded
findMany added in the slim slice), source-collapsed to the canonical
source, then summed over the day-aligned 7/30/90-day sub-windows. A
window whose buckets lack accumulators (pre-backfill) composes to null,
the same shape the empty REGR_* FILTER returned, so a transitional read
never returns a wrong partial regression.
median (PERCENTILE_CONT) and the exact |z|>2 anomaly count stay on the
minimal live query - both are non-composable across buckets. The cold
coverage-miss path keeps the live REGR_* regression. Annotate
slope_source:"rollup" + median_source/anomaly_source:"live" so a
regression to the heavy path stays observable (no-silent-cap rule).
The comprehensive briefing's bounded feature read (extractFeatures) was
paid more than once where it should not be: the on-demand route and the
shared generator each run a downgrade ladder that re-reads with identical
arguments, and a nightly warm has no shared object across consumers in a
single tick.
Add a request/run-scoped feature cache: withFeatureCacheScope opens an
AsyncLocalStorage store, getCachedFeatures memoises the first compute per
(userId, day, includeRaw, sinceDays), gated on a cheap salient-input
fingerprint so a mid-scope ingest recomputes and a probe failure fails
safe. Outside a scope it is a pure passthrough. Wired into the
comprehensive generator, the on-demand POST handler, and the nightly
per-candidate loop + forceWarmUser. Coach keeps its own snapshot.
The slow-moving status input fingerprint covered the card's own metric
rows plus mood and height, but not the FDR discovery channels that also
feed the prompt via getRelevantCorrelationsForMetric. A correlation
newly surfacing off a behaviour channel (e.g. daylight -> next-day
weight) left the card's own rows untouched, so the input gate re-stamped
the stale assessment and the discovery never reached the prose.
computeStatusInputFingerprint gains includeCorrelationChannels: it widens
the same grouped probe by the discovery behaviour + outcome channels
(deduped, mood excluded as it is mood-entry backed), so a channel change
flips the hash. Wired into weight-status, the one caller that folds a
relations block.
Weight and BMI move on a weekly cadence, but the per-metric generators
rebuilt the full snapshot (bounded findMany + per-series rollup reads +
correlation math) every day and relied on the post-build content-hash gate
to skip only the model call. Add an input gate that runs first: a single
grouped probe over the salient inputs (per-type count + newest reading, plus
mood for weight and the profile height for BMI) fingerprints what the
snapshot would read. On a non-forced run whose fingerprint matches the cached
row, the cached assessment is re-stamped under today's key and the whole
build is skipped. A miss falls through to the normal build and the existing
post-build gate.
Consent is checked before any re-date so a revoked-consent user never has
stale text presented as current. Forced runs never gate but still record a
fresh fingerprint for the next day.
The warm narrow query and the cold heavy query each referenced the
canonical-source subquery twice — once in the window_stats CTE and once in
the outer body — so the DISTINCT-ON pick over the live 90-day rows ran twice
per rebuild. Extract canonicalMeasurementsCte (the SELECT body without the
alias wrapper), bind it once as WITH cm AS (...), and select FROM cm m from
both window_stats and the outer aggregate. Output is identical; the dominant
self-join runs once.
canonicalMeasurementsFrom now wraps the same body, byte-stable for the
single-reference call sites (pinned by a new parity test).
Bound the comprehensive briefing's feature read to a 400-day window instead of
walking the full measurement history on every generation. All-time min/max/avg
are sourced from one grouped aggregation over the whole history, so the
allTime* fields stay honest while the unbounded row materialisation is gone;
the oversize downgrade ladder remains as the safety net.
Fold illness-episode and cycle state into the briefing prompt, reusing the
server-authoritative builders the Coach already assembles. Both are
module-gated and cost nothing for users without them, so a mid-illness or
mid-cycle user finally gets context-aware prose.
Ground daily-briefing numbers against the signal features, guard Coach replies
against fabricated doses and risk scores, gate the period narrative, feed the six
status cards a precomputed signal block, add SSE cancellation, and cap the Coach
snapshot read.
BP, pulse, weight, BMI, mood and medication-compliance still asked the
model to derive the recent-vs-baseline comparison from raw buckets — the
exact arithmetic small models do unreliably, and the pattern the
base-system prompt already advertises as precomputed.
Build the SIGNAL block (current / baseline / signed delta / normal-swing
verdict) with buildMetricSignal from each card's graded series and inject
it into the snapshot; BP carries two signals, compliance one per
medication. Each card's prompt section now leads from the signal block
and is told to state, not recompute. The server-computed compliance %
is unchanged.
The raw BP pulls in the comprehensive aggregator and the bp-in-target live
fallback read every source's row for BLOOD_PRESSURE_SYS/DIA. BP is a
multi-source metric — a Withings reading mirrored into Apple Health is two
rows for one event — so each shared reading was counted twice in the
in-target denominator and twice in the recency-weighted grade. The rollup
branch of the same fast-path reads source-collapsed measurement_rollups, so a
rollup-warm account and a rollup-cold account returned different BP
percentages and grades for identical data — a web/iOS and cold/warm parity
break.
Route both raw pulls through canonicalMeasurementsFrom so the cold/live path
picks the ladder-winning source per (type, day), matching the warm/rollup
path. Per-event rows of the winning source survive the collapse, so the
downstream 5-minute pairing still sees individual readings.
One-shot prettier pass so the new format:check gate starts from a clean
baseline. Formatting only — no behaviour change. Generator-owned and
vendored artifacts stay untouched via .prettierignore.
Prove one provider call covers every present metric, absent metrics are
omitted from the prompt and never fabricated, an all-served cycle makes no
call (the per-metric hash gate still short-circuits), and the batch degrades
gracefully: an omitted metric or a provider error falls each affected card
back to its single-card path, invalid JSON retries once, and a no-provider
cycle serves each card's precomputed no-key result.
The comprehensive generator did an ad-hoc JSON.parse with no corrective
retry and failed cold to `invalid-json` on a first-pass miss, while the
strict insight wrapper's `buildRetryCorrectionMessage` sat right next
door. Export that helper and reuse it: on a parse miss, append the
correction to the user prompt and re-run the chain once before declaring
failure. Unifies the JSON-robustness story across the two paths.
The content-hash gate re-stamps the prior text whenever the snapshot is
unchanged, so the "daily" briefing read byte-stable day-over-day. Decouple
phrasing freshness from data freshness for the dailyBriefing.paragraph
alone: on a hash-unchanged hit, re-roll just the paragraph once per
calendar day at a higher, seedless temperature, keeping every structured
finding verbatim from the cached payload.
A new `insightsBriefingRerollDate` column gates it to at most one extra
call per day, and a re-roll miss still stamps the day so a failing
provider cannot be re-driven on every page-open. The hash gate stays
fully intact for the status cards (reference assessments — monotony is
fine).
The overview charted each trend but never offered a way into the metric's
detail page — the steps page in particular exists and is used, yet Insights
never linked to it. Give every Trends-row chart a routed detail target and a
discreet "view details" link below the caption, so a metric the row surfaces
(steps included, when the briefing flags it) drills into /insights/<slug>.
An active illness/condition episode annotates, it never penalises: the
measured number is untouched, only the narrative changes.
- rest-mode.ts: one server-authoritative resolver (module-gated, fail-soft)
for "is the account in Rest Mode + which episodes back it". iOS mirrors the
resolved context, never recomputes the suppression.
- health-score: /api/analytics attaches a value-free Rest Mode annotation;
computeHealthScore stays pure (score never adjusted).
- recovery-score: computeRecoveryScore carries the Rest Mode context as of the
scored day; the persisted blend is unchanged.
- coach-nudge: pause cadence-nudges during an active episode (new
skippedDuringIllness counter, gate sits before the trigger search).
- achievements: bridgeFrozenStreakGaps freezes (does not break) every
day-streak across illness days; never invents a streak, a non-illness gap
still breaks.
- coincident-deviation: reframe a fired flag as illness-explained; vital
numbers untouched. Surfaced as a neutral line on the insights card.
- coach snapshot: illness history + active illness as load-bearing context so
the coach knows about Rest Mode (no decrypted note reaches the prompt).
- FHIR: illness episodes map to Condition (generic disease root, label on
code.text, patient-reported/unconfirmed) + a bounding Encounter.
- doctor-report: module-gated illness section in the aggregator + PDF table
(retrospective, neutral palette); FHIR Condition section in the bundle.
i18n x6 for the report + reframe copy. Tests: score-annotation-not-
penalisation, streak-freeze, coach-nudge pause, FHIR Condition/Encounter
mapping, doctor-report inclusion, Rest Mode resolver. typecheck/lint/test
green, openapi in sync, knip clean.
Editing any field of a mood entry re-sent its tag set, and archived
tag keys can no longer resolve — the diff deleted the entry's links to
them invisibly, against the archive contract the tag suite just
shipped. Submitted tag keys now govern active links only; links to
archived tags survive every edit, and the explicit purge stays the one
removal path. The insights surfaces — tag breakdown, influence, better
days, crosstab, the model snapshot and the narratives — render the
decrypted custom-tag label instead of the raw key, decrypted once per
tag server-side.
The external ingest endpoint was the one intake writer that never
attributed to the minted band: it stored a bare row beside the pending
slot, so an early take through the API showed as the last intake while
the slot kept demanding a dose, the ledger recorded an extra plus a
miss, and the day tally counted both. It now resolves the band and
converges onto the pending row like every other path, the Telegram
confirm and skip buttons do the same preferring the reminder's own
slot, and the nightly convergence pass heals the pairs this has
already produced — one band match, one live anchor row, both days'
rollups recomputed. The cache layer fences in-flight builders on
invalidation so a read can neither join a pre-write build nor have a
background refresh re-cache a pre-write body as fresh. Ad-hoc rows no
longer resolve a different slot through the six-hour radius, a bounded
early grace credits a take up to an hour before the window without
crossing the previous band, day-scale cadences taken early downgrade
the slot-day prompt instead of inviting a double dose, the overdue
pill ignores the projector's pending mints, created intakes carry the
same plausibility bounds as edits, slot moves keep the recorded
injection site and dose, both affected days recompute on a moved
toggle, deliberate slot-less skips converge instead of orphaning, the
coach reads adherence from the ledger, the cards and quick-add run in
the user's timezone with the served next-due as the gate, and
schedule-revision, wipe and restore writes invalidate the caches they
feed.
Per-status texts check AI consent before the unchanged-data refresh can
re-stamp them, the comprehensive fingerprint folds in the comparison
baseline and generation language so toggling either regenerates once,
and profile, threshold and source-priority writes evict the analytics,
targets and derived caches they feed. CSV export neutralises spreadsheet
formula prefixes on free-text cells while numeric columns stay typed,
the derived batch validates its type token against the measurement enum
and carries a per-user rate limit, and a dose synced from the phone app
counts as the interactive write it is. Background cache refresh
failures log a structured event instead of vanishing, the warm queue
runs visit-triggered jobs alongside the nightly walk, the formatter
memo evicts one entry instead of flushing wholesale with its option
objects frozen, the targets cold build bounds its fan-out, and time in
bed unions every writer's envelope while stages stay with the richest
recorder.
Staleness can re-enter a generated text: measurement age is hashed as a
bucketed tier, so the fingerprint stays stable day to day but flips
exactly when a reading crosses the boundary the prompt caveat cares
about. The ingest debounce becomes a one-hour minimum gap with the
hash gate carrying the budget, so a notable morning reading is
narrated the same morning instead of after the next nightly run. A
manual regenerate now refills the status cards through the same
hash-gated pipeline — changed cards regenerate, unchanged cards
refresh free, no eviction touches the hash baselines. The user's
self-context joins the comprehensive fingerprint so a newly remembered
fact reaches the briefing, the sleep writer collapse covers every
caller that reads nights, the batched compliance build fans out
bounded instead of sequentially, and a failed compliance read renders
a quiet retry state in the exact skeleton footprint instead of
swallowing the error.
A sha256 fingerprint of the compacted feature snapshot is stored beside
every generated text; nightly, forced and ingest-triggered regeneration
all skip the model call when the fingerprint matches and refresh the
timestamp instead. Warms cover only the user's resolved locale and the
blanket per-status eviction after a comprehensive generation is gone —
a lazily-read second locale regenerates on its own miss path. The
ingest debounce widens from thirty minutes to six hours and no longer
deletes cache rows, the visit-triggered warm re-checks freshness at job
start, backs off for an hour after a failure and is capped at two
forced runs per user per day, and its dedup window now outlives the
client poll horizon. An active day drops from hundreds of model calls
to a handful tied to actual data change.
A typed apiFetch wrapper owns the envelope unwrap, the ok-check and the
error read for every client call — two hundred call sites across the
app migrate onto it and a lint rule keeps raw api fetches out for good.
The legacy timezone helpers fold into the tz module and test fixtures
drop personal names.
A dose past its slot time but still inside its catch-up window now stays
on the medication card as overdue and takeable instead of yielding to
the next occurrence — the next-due field can therefore carry a past open
slot, flagged with nextDueOverdue. Y-axis padding no longer dips below
zero for series that cannot be negative, so a steps chart stops
suggesting minus nine hundred steps.
The two-o'clock status crons generated for every account, ignoring both
the per-user coach opt-out and the operator kill switch; they now share
one gated discovery that also skips users the pregenerate pass will
rewarm anyway, and the mood card finally gets the nightly run the other
six always had. Deleted measurements no longer feed any prompt. Locales
without an assistant language fall back to English consistently instead
of German texts with English fallbacks. A rate-limited regenerate stops
claiming success, the dashboard briefing state distinguishes a missing
provider from a pending generation and serves a stale briefing with its
timestamp instead of hiding it, narrative warming actually processes
oldest first, and worker failures reach the error tracker through one
shared reporter while the insight queues gain real retries.
The top insights navigation was driven purely by data availability: a
sub-page pill appeared whenever its metric had readings in the 90-day
window, with no connection to the per-tile visibility the customize
panel writes. Hiding a sub-page in Anpassen therefore never removed its
nav pill, and the two surfaces drifted apart.
Thread the resolved insights layout into the tab strip via the layout
shell. A pill now shows iff its metric has data AND its slug is visible
in the saved layout; data availability stays the floor, so a
layout-visible metric with no readings still gets no pill. Pills and
group children order by the saved layout order, falling back to the
natural slug order. While the layout query is in flight the strip keeps
the data-only gate, so first paint is unchanged and a slow or failed
layout fetch never suppresses a pill the user has data for.
Add a total slug to manager-group mapping in sub-page-metric so every
routed sub-page resolves to a category, a superset of the tab-strip nav
grouping that also covers the flat pills and the sleep / mood / events
categories the strip never collapses.
Every surface that applied a resting-pulse target judged the raw PULSE
series against a resting band. Apple Health maps all heart-rate samples
— including workout HR — onto PULSE, and the resting figure onto a
separate RESTING_HEART_RATE type. A single workout emits hundreds of
expected-high samples, so the resting-target evaluation read a large
share "outside target" and tanked the in-target % / consistency even
when the clean resting series averaged a healthy figure.
Judge the resting band against a resting series. A shared helper prefers
RESTING_HEART_RATE; when the user has none it derives a proxy from the
low percentile of each day's PULSE samples, so a dense workout burst
sitting in the upper tail cannot drag the resting estimate up. Applied
to the resting-pulse tile + its consistency strip, the pulse in-target %
in the assessment snapshot, and the dashboard / sub-page classification.
Raw heart rate, where it is merely charted, no longer carries the
resting-band overlay. The doctor report is unchanged — it reports raw
heart-rate stats factually, with no resting-target judgment to mislead.
- Medication cycle status: re-hydrate `nextDueAt` (it arrives as an ISO string
over the wire) before the Berlin day-bucketing, so "next dose in N days" no
longer drifts by a day near midnight or off Berlin. Covered by a new
string-input test.
- Wellness rings: drive the conic sheen off a fixed white so it always lightens
(a near-black light-mode foreground under overlay was darkening the tile into
a grey band); give the ring a transform transition so it scales smoothly with
the tile lift instead of snapping; cap the green-band glow at a few breaths
instead of an infinite drop-shadow repaint; deepen the light-mode stress and
recovery end-stops for non-text contrast; track the sm/md column count to the
active-score count so two or four scores never leave an empty cell.
- Mood factor crosstab: bound the snapshot's cross-metric read to the 365-day
window and exclude soft-deleted rows, so the factor deltas the model cites
match the visible board over the same canonical row set (no recency-biased
truncation, no soft-delete skew); thread the user's source priority so the
per-day canonical pick agrees with the screen; render the delta neutrally
rather than green/red, since a higher vital is good for steps but bad for
resting heart rate and the board is observational.
- Log-intake: surface an error when a pinned slot can't resolve to an instant
instead of silently logging the dose through the unscheduled path.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Rated mood factors (work / sleep-quality / stress, scored per entry) now
participate in the cross-metric correlation surface. Until now their score
series was captured but read by zero analytics.
Two paths, both on the existing FDR-controlled engine:
- A factor x vital crosstab on the mood insights page and the mood
snapshot: for each factor the user scores, split its days low vs high by
the factor's own median, then run the same Welch test + day floors +
Benjamini-Hochberg correction the tag crosstab uses. Surfaces lines like
"sleep runs ~38 min below baseline on the days you rated work low"
(n-gated, q <= 0.10, never fabricated on thin data).
- Factor channels (FACTOR:<key>) fold into the period-narrative discovery
matrix as both lag source and lag target, so the drivers the coach and
narrative reference can cite a factor relationship.
An inverse factor (higher rating = worse day) gets the documented sign-flip
once, at the series boundary, so "up" always reads as a better day and the
low/high split and phrasing stay honest. Strings land in all six locales;
the surface carries the standing "association, not cause" caption.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add an additive, non-breaking assessment field to GET /api/insights/derived:
{ text, source, updatedAt } | null, keyed to the requested metric id for the
five per-score ids (READINESS, SLEEP_SCORE, RECOVERY_SCORE, STRAIN_SCORE,
STRESS_SCORE). The field is null when the metric is not one of those or when
its status is not "ok". A deterministic text always fills it — composed from
the score's own contributors, so provider-less accounts and the demo are never
empty — and warmer provider prose overrides it once cached, warmed out of band
so the GET never blocks on a model round-trip. de/en, following the v1.12.12
deterministic-narrative model. The batch grid read returns assessment: null;
the detail sheets use the single-metric route.
Introduce a shared buildMetricSignal() helper: a token-lean, signal-rich
descriptor (current, own baseline, signed delta, spread, normal-swing verdict,
normal-range placement, contributors) computed server-side so the model states
the comparison instead of re-deriving it from raw buckets. Thread it into the
generic per-metric assessment snapshot and lead the base prompt from it. Add a
PROMPT_VERSION constant to the base system prompt so prompt changes are
traceable.
Regenerate docs/api/openapi.yaml.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Anthropic and the local provider do not send a JSON mode, so a compliant model
still wraps its { "summary": … } reply in a code fence or prefixes a sentence,
which broke JSON.parse and surfaced the raw fenced string as the assessment.
Add a shared stripJsonFences helper (drop the fence, narrow to the first { …
last }) and apply it as a parse fallback in parseSummaryFromContent. No-op on
already-clean JSON.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The generic metric-status generator read SLEEP_DURATION row-per-stage and
fed the snapshot a single sleep STAGE (~16 min) as the current value, so the
sleep prose reported a fraction of an hour for a full night. Route the
SLEEP_DURATION status/snapshot through reconstructSleepNights — the same
per-night, source-deduped, midnight-aware time-asleep total the dashboard,
series, and targets surfaces use — so the snapshot's series, summary, and
latest value are all the night total in minutes (the registry unit and normal
range are already minutes). Gated to SLEEP_DURATION; no other metric's status
path changes.
The readiness wellness index could surface a contributor value of 0 for an
account that recorded resting heart rate / HRV / respiratory rate but had no
scorable baseline: a non-finite baseline center or today value fell through
clamp100 to a misleading 0 that reads as a real worst-case sub-score. Return
null instead so the client drops the contributor, and document the null-vs-0
contract: null means no usable signal (no baseline, no recent reading,
non-finite input); a finite value — including a genuine 0 — means the baseline
is established and the latest reading really deviates that far.
(cherry picked from commit 0422ac012765f30069a41f662c2b025197ed9a88)
The per-metric assessment cards are tightly grounded but their form had
converged: every card opened the same way, forced a "one doable step"
even when nothing was actionable, and a metric that stayed steady for
weeks produced near-identical paragraphs. Squeeze more value from data
already computed server-side, without touching the grounding floor or any
safety contract.
- Surface FDR-surviving cross-metric correlations on the per-metric card.
The discovery engine already screens behaviour×outcome pairs with a real
Pearson + exact p-value + Benjamini-Hochberg FDR control; that
intelligence only reached the period narrative. A new read-only consumer
runs the same full-matrix discovery and filters to the pairs that
involve the current metric, then feeds the engine's own conservative,
descriptive, never-causal interpretation strings into the prompt as
grounded relations. Best-effort: a correlation hiccup can never block the
generation it only decorates.
- Streak-aware repetition. Derive a steady-run length from the graded
series (consecutive recent weeks within a band of the user's own
baseline) and tell the model when it has already reported "no material
change" N times running, so it acknowledges the continuity in one clause
and pivots to a different facet instead of restating the same level.
- Stop forcing a step; add controlled variety. The closing step is now
conditional — when nothing is genuinely actionable the model affirms and
names one thing to watch rather than manufacturing the filler that lets
platitudes back in. A deterministic per-render variety token (seeded from
user + metric + day, never Math.random / Date.now) rotates the opening
angle so consecutive cards and days don't read identically, and the
phrasing temperature lifts 0.3 → 0.45 while the facts stay pinned by the
snapshot and the forbidden-phrase guards.
- Surface the computed data strength (n + recency) into the prose prompt so
it hedges on thin data instead of guessing what "few" means, matching the
UI confidence badge.
- Add locale-matched few-shot examples (a grounded assessment + a labelled
banned-filler counter-example) to the assessment system prompt, which
most helps weaker local-provider models follow the contract.
The own-baseline grounding, computed-not-hallucinated stats, schema-
enforced output, filler-phrase ban, and correlation-as-association
discipline are all preserved and re-pinned by new guard tests.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
(cherry picked from commit c9d04cab1523e0e57a3f50c87c9f88684b01b26f)
The tag × metric crosstab summed cumulative metrics (steps, active
energy, sleep duration) per day across every source with no canonical
pick, so once two sources reported the same day — Fitbit + Apple steps,
Fitbit + WHOOP sleep — the per-day total double-counted and inflated the
with/without averages and the Welch delta the surface ranks on. Run the
metric rows through the same per-day source-priority + device-type picker
the analytics steps/sleep path uses before bucketing, keyed by the
metric's priority ladder and the user's Berlin calendar day. Same-source
per-stage rows still sum into the night total; only cross-source twins
collapse. Thread the user's source-priority blob through the read so the
pick honours the configured ladder.
The cross-metric read was a flat `take: 5000` over the 365-day window
ordered `measuredAt desc`, unioned across per-stage sleep + per-sample
pulse/HRV; a multi-source, high-frequency user blew past it and silently
dropped the oldest months, biasing the crosstab and the long-window
correlations toward recent data. Lift the cap to cover the realistic
worst case for the window and annotate a wide event when the cap is hit
so a truncated read is observable rather than silently wrong. The read
stays raw rather than moving to the DAY rollup tier because that tier
buckets on UTC midnight while these aggregates key on the user's Berlin
day, which would skew the day pairing at the boundary.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
(cherry picked from commit aa72ffba6e3bcac9f97be4adc26d689d4f352e98)
Extend the mood-relations surface from mood to a health metric. For each
structured mood tag present on enough days, compare a metric's value on
tag-present vs tag-absent days: active energy on workout days (same day),
sleep length on nights tagged sleep (same day), and next-day recovery
after a tag like alcohol or food (D to D+1 lag). The comparison reuses the
existing statistics engine wholesale — the Welch two-sample t-test, the
per-group day floors, the confidence band, and the Benjamini-Hochberg FDR
control already used by the influence and discovery boards — so no new
statistics code lands. Only pairs that clear p < 0.05 and the FDR control
surface, ranked by adjusted q then absolute delta and capped at eight rows.
The result rides on GET /api/mood/insights as a new additive
`tagMetricCrosstab` field and renders as a confidence-gated, observational
card on the mood insights page, framed as an association in your own data
rather than a cause. The endpoint is server-rendered and not in the OpenAPI
contract, so no contract change is required.
Deprecate the standalone moodLog integration. Mood is now tracked fully
inside HealthLog through entries, structured tags, and rated factors, so
the external webhook and reverse-sync bridge no longer add anything. The
webhook route, the push and pull helpers, and the validation schemas carry
an @deprecated note pointing at the native path; the surface stays
functional this release and is slated for removal in a future major. The
settings card shows a deprecation hint, localised across all six locales.
- Mood discovery footer no longer pairs a mood-only numerator with the
full behaviour-by-outcome denominator. Reword it to surface the mood
relationships out of all day-to-day pairs the engine tested across
every metric, so the count it reports is honest about its scope.
Update all six locales.
- Rank the better-days board on a comparable scale: standardize the tag
mood-point delta into a Cohen's-d effect (delta over the pooled
with/without mood SD, clamped to [0,1]) so it sits alongside a metric
|r| instead of the old |delta|/2 heuristic. Fall back to |delta|/2 only
when the pooled SD is undefined. The pooled SD is ranking-only and is
stripped before the row enters the snapshot payload; the raw delta and
r the cards show are unchanged.
- Normalize the Withings and Whoop resume buttons to the min-h-11 scale
token, dropping the arbitrary min-h-[44px] spelling.
- Give the API token copy button a 44px tap target on small viewports,
scaling back to the tighter icon size from sm up.
- Drop the dead cached prop from InsightStatusCard and every caller; the
card has carried no cached affordance since the badge was removed.
Two changes that turn the descriptive relations surfaces into a
statistically-gated discovery view on the mood page.
Mood as an outcome: the FDR-controlled all-pairs correlation discovery
engine treated mood only as a behaviour (a lag source). Promote it to an
outcome channel too, so the scan also surfaces "behaviour today → next-day
mood" relations (e.g. more daylight today → better mood tomorrow) alongside
the existing "mood today → next-day outcome" direction. The self-pair is
skipped and Benjamini-Hochberg already controls the larger pair family this
opens up. The discovery route now feeds the shared mood daily-mean series
into both roles.
Surface the discovery slice: the engine was computed but never rendered.
The mood Insights page now shows the mood-involving pairs that survive the
false-discovery correction, each phrased observationally with its paired-day
count, r, and BH q-value, plus the honest "X of N tested" footer. The
surface self-fetches the correlation route and degrades to nothing when the
operator has disabled it. Observational only, with the standing disclaimer.
Two new observational surfaces on the mood Insights page, both computed in
the single mood-aggregates engine so the charts and the coach snapshot stay
in sync.
Influence on mood (per tag): for each frequent flat and structured tag,
compare the average daily mood on days the tag is present against the
counterfactual baseline of days it is absent, with the delta and a
confidence band. The band comes from a Welch two-sample t-test reusing the
exact Student-t p-value already behind the correlation cards, not a naive
approximation. Both groups must clear a five-day floor; rows rank by
absolute delta and cap per axis. Sparse tags, single-day tags,
all-same-value days, and zero-delta tags are dropped cleanly — never NaN.
Better days board: one ranked list folding the tag deltas and the
mood x health-metric correlations (sleep, steps, pulse, weight, blood
pressure) into a single effect-size-ranked, confidence-gated surface — the
headline answer to "what goes with my good days?". Near-zero and
under-sampled correlations are excluded; the list is capped and observational
only, with the standing "association, not cause" caption.
Both surfaces ride the mood-status snapshot too so the assessment prose
never contradicts the on-screen relations. New copy in all six locales.
Marc-André Bombeck