The recovery page rendered a grid of compact device-score tiles plus a
redundant cross-link to the recovery-score anatomy at the top. It now
follows the same structure as every other metric page (blood pressure,
pulse, the HealthKit scaffold): each present device-native signal renders
a block carrying a short explainer, the max / median / mean stat strip,
the shared chart with the identical 7 / 30 / 90 / All toggle, and a
per-chart assessment. Multiple charts now carry multiple matching texts.
Register the seven device-native recovery / strain signals (ANS charge,
day / workout strain, cardio load, whole-day average / max heart rate,
energy expenditure) in the generic metric-status registry so each chart
gets a generated assessment, gate them on the recovery module, and give
each its own chart-overlay slot. Drop the duplicate top score-link block
(this surface is reached from the overview already).
Regenerate the OpenAPI contract for the widened metric-status enum.
The Coach can append a ---SUGGEST-REMINDER--- sentinel to propose a
measurement cadence (weight daily, the ESH/AHA 7-2-2 home-BP protocol,
structured glucose). The server resolves the cadence from a closed catalog
and emits an additive 'suggestion' SSE frame the chat renders as a one-tap
action card; accepting it creates a MeasurementReminder with origin COACH
through the same engine the Vorsorge surface uses (server-authoritative
schedule, finite course window self-expires via endsOn).
Non-naggy by construction: a frequency-cap cooldown, dismissal memory, a
master opt-out under Coach settings, and an explicit 'I measure enough'
stop path, all in the coach-prefs blob. Triple dedup at suggest, render,
and create so a cadence is never double-suggested or double-created, and
suggestions for a disabled module never surface. RHR/HRV stay passive --
the Coach surfaces the baseline, it never nudges to 'go measure' them.
The reminder DTO now echoes origin and endsOn for the iOS client.
Add `src/app/api/insights` to the inventory's walked trees so every
`route.ts` beneath it must classify: the 14 AI-narrative routes as
insights-module-gated, the Coach sub-tree as coach-delegated, and the
AI-settings / provider-chain / tile-layout / feedback / targets /
glp1-timeline routes as explicit infra-and-config EXEMPT entries with
documented reasons. A future ungated insights AI route now fails the
inventory BY NAME rather than leaking the surface over a Bearer token when
the account turned insights off.
Add representative insights-off → 403 module.disabled coverage to the
blood-pressure-status and correlations route tests (alongside the existing
insights-on → 200 path).
Require the `insights` module (in addition to the existing assistant-surface
gate and any per-metric data-module gate) on all 14 narrative insights
routes: blood-pressure / bmi / pulse / weight / medication-compliance /
mood / metric / derived / derived-batch status, cards, correlations,
narrative, pregenerate, and rhythm-events. A 403 module.disabled envelope
now precedes any provider or cache work when the account has insights off.
The per-metric sleep/glucose/recovery/mood gates and the
requireAssistantSurface("insightStatus"|"correlations") checks are kept;
this adds the module layer on top of them.
Route tests that now resolve the gate mock @/lib/modules/gate
default-enabled so their existing data-shape and assistant-flag assertions
ride through unchanged.
The batch derived endpoint (the one the dashboard wellness strip reads)
computed Recovery/Strain/Stress unconditionally while the single derived route
already gated them, so disabling the recovery module did not hide the rings.
Reuse the single route's DERIVED_MODULE map (now exported, with READINESS added
since it equals the computed recovery score) and drop disabled-module metrics
from the batch payload per request, without mutating the cached map.
Per-route tests assert the two-layer Coach module gate: a disabled
module (operator availability OR per-user disableCoach) yields 403
module.disabled before any Coach work runs, and an enabled module falls
through to the normal path. Covers the chat POST + list GET, about-me
GET/PUT, the questions GET/DELETE, and the adopt POST short-circuit
ahead of the locked read-modify-write.
The streaming POST and the conversation-list GET enforced only the
legacy operator assistant flag (requireAssistantSurface("coach")),
missing the per-user disableCoach opt-out and the operator
module-availability layer. Both handlers now run the two-layer
requireModuleEnabled(userId, "coach") gate right after auth, returning
403 module.disabled when the account has Coach turned off — even with a
valid Bearer token. The legacy assistant-flag check stays in place.
The derived-score, metric-status and mood-status insight routes serve a mix
of core-vital and toggleable-module signals. Map the module-owned metric ids
(sleep/recovery/glucose) to their module key and refuse with the shared 403
module.disabled envelope when the account has that module disabled; mood-status
gates wholesale on the mood module, mirroring the cycle gate. Metrics absent
from the map are core vitals and stay open.
A PUT that omits sections (an iOS tiles-only reorder) reset the user's
section customization to defaults, and symmetrically a sections-only PUT
reset the stored tiles. serializeInsightsLayout now takes the resolved
stored layout and falls back to it — not to defaults — for whichever
dimension the body omits; the PUT handler loads the stored layout first.
A first-ever single-dimension PUT (no stored layout) still fills
defaults, so the v2 blob stays complete. Pins the merge behaviour with
three unit cases.
A medication can declare itself as-needed: no schedules, never due,
never reminded, excluded from every compliance rate — but intakes log
ad-hoc, consume inventory exactly like scheduled doses, and the
history stays. The wizard offers it as a cadence choice and round-
trips it on edit; card, table and detail show a calm marker where the
next dose normally sits, the take-all dialog never collects one, and
the supply runway stays unknowable by construction. (#316)
Stock only decremented on the web card's take button — the status
toggle, the phone sync, the Telegram buttons and the token API all
recorded the dose and left the inventory untouched, which is why a
fresh pack kept reading full. Consumption now stamps the intake row
itself: every path that turns a row taken consumes through one module
— open container first, then earliest expiry with auto-open, spilling
into the next pack and flooring at zero — and every path that takes it
back (undo, delete, a take turned skip, a converged duplicate)
refunds exactly what the stamp recorded. Replays and pending sync
echoes can never double-charge, and the nightly convergence pass
refunds the rows it tombstones. A medication can declare units per
dose, so two 2 mg tablets book as one 4 mg dose and stock reads in
doses with the unit count alongside; containers carry a named type
including tablet packs, hold up to a thousand units for inhalers, and
the GLP-1 card reads the same stock as the inventory tab instead of a
separate ledger. The inventory endpoints join the documented API.
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.
The dashboard reserves chart-row skeleton cells from the saved layout
before the snapshot resolves, so a cold load paints close to its final
footprint instead of growing by a screen when data lands; the tile
silhouettes count only tile-capable widgets with blood pressure as two,
the greeting line keeps its box through name personalisation, and the
getting-started checklist moves below the charts where its arrival
shifts nothing. On the insights overview the targets build runs its
eight reads in parallel with a one-year floor on the glucose scan
instead of strictly in sequence over an unbounded history, the derived
batch gains a server cache where it previously recomputed sixteen
metrics on every read, both serve stale-while-revalidate, and the
health-score column reserves its exact footprint while the analytics
payload is pending.
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.
Stale-served insights responses now carry a revalidating flag; the
advisor and comprehensive queries poll on a bounded 25 s cadence until
the background warm lands, so an open session converges instead of
pinning yesterday's briefing for an hour. Interactive measurement
writes (single create, edit, delete, bulk-delete, restore) hard-evict
the analytics cache so a manually entered value is reflected on the
next read; background batch sync keeps the mark-stale path. The
visit-triggered force warm collapses to one job per user instead of
one per locale, and the umami script proxy failure now annotates a
named action with the failure reason instead of a bare flag.
The insights read path served a cold cache by generating inline — up to
ten seconds on first visit — and the nightly warm pass structurally
excluded evening visitors: their visit stamped the cache, the half-four
run skipped it as fresh, and it expired exactly at the next evening's
visit. Stale content now serves immediately with its timestamp while
one background job revalidates, the warm threshold drops to an hour,
forced warming fills both message locales, the analytics caches serve
stale-while-revalidate, the profile endpoint parallelises its reads,
the bug-report status drops a write per call, the umami proxy degrades
to a no-op script instead of an error, the insights route sheds its
heaviest static imports, and the medication page prefetches on intent.
The dashboard snapshot fires the moment login succeeds instead of
waiting for the page chunk (start ~670 to ~200 ms throttled). Guided
follow-ups are personalised from the user's snapshot and an answer gets
the coach's streamed reaction before the next question. Archived
schedule eras can be corrected through a superseding revision that
keeps the original as an audit trail, era writes serialise per
medication, and duplicate dose times collapse in validation. A rolling
medication whose next dose is tomorrow can no longer read as heavily
overdue — the card status derives strictly from the served next-due —
and the adherence bars keep a fixed geometry in every state. The
server-wide assistant key gains its admin card, the one duplicated
coach preference collapses onto a single owner, twenty-seven admin and
settings card headers move onto the shared pattern with sixty
containers normalised, and the remaining hydration warnings are gone.
The client bundle carries only the active locale — resolved on the
server and handed to the provider with no hydration flash, halving the
dashboard route to under a megabyte parsed (288 KiB gzip). The auth
shell renders the app chrome immediately instead of a spinner while the
session resolves in parallel, the widgets request folds into the
snapshot it duplicated, and insight pills prefetch their sub-pages —
a first visit to a sub-page lands in about ninety milliseconds instead
of nearly a second.
Editing a schedule's times no longer rewrites the past. Each material
change archives the outgoing cadence as a dated revision, and every
band consumer — history, compliance, write attribution, the auto-miss
cutoff — mints past periods from the schedule that was valid at the
time, so a dose era taken at its then-correct times stays on time
forever. The repair script can infer historical eras from consistent
old anchors behind an explicit flag. The welcome tour scrolls its
target into view, flips above it when space below runs out and falls
back to a centred sheet on any viewport, so no step can strand its
buttons below the fold. The avatar menu drops the about entry — that
lives in settings.
The composer starts as a single line with send beside it and grows only
when the text does; the shift-enter hint and the inline disclaimer are
gone. A typing indicator bubbles before the first token. The context
rail hides by default behind a toggle and stops scrolling sideways,
conversations open in the full view instead of a broken side drawer,
and the coach preferences move into Settings where the rest of the
configuration lives — context toggles and the exclusion list now derive
from one source so they can no longer contradict each other. A floating
bubble appears only while a coach-initiated nudge is unread. Clearly
stated allergies and conditions are remembered immediately through a
deterministic extractor instead of waiting for the twenty-turn
summary pass.
Every standalone intake anchors its slot on the take time since
v1.15.19, so guarding the nightly dedup on that shape excluded ordinary
drift duplicates from collapsing — the original point of the sweep.
Releasing a pinned dose now persists the user's choice on the row, the
dedup and the repair script key their exclusion on that flag, the read
model reports a released row as deliberately off-schedule instead of
re-snapping it within the anchor epsilon, and the attributed badge only
shows on rows actually bound to a slot. A few fixture names and stray
comments also drop personal references.
Invites move into their own admin section with a real ledger: status
badges, expiry, uses, a redemption history of who joined when (new
redemption table, revocation is now soft so history survives), and an
honest one-time link/QR moment. The assistant self-context grows
structured fields for conditions, allergies and focus — encrypted like
the free text — merges age and gender from the profile into one fenced
prompt block, and asks up to three targeted follow-up questions that
surface as tappable chips above the coach composer. A fourth nudge
trigger gently suggests a check-up when the self-context is stale and
the coach is actually in use.
Users can now tell the assistant about themselves in an encrypted
free-text profile that feeds the coach and briefing prompts as clearly
fenced context, with the coach allowed one targeted follow-up question
when the picture is incomplete. Operators on closed registrations can
mint expiring invite links with a QR code from the admin area; a valid
invite opens registration exactly as many times as it allows and records
who redeemed it. A nightly pass watches three deterministic signals —
low adherence, blood pressure above its target band, a falling recovery
trend — and sends at most one coach nudge per week through the existing
notification channels, gated behind the same opt-outs as every other
assistant surface.
The auto-miss cron stamped every pending row missed a flat day after
its slot even though weekly schedules stay takeable for days, touched
tombstones because it never filtered deleted rows, and a verdict stuck
even after the dose was taken. The cron now derives its cutoff from the
schedule's window reach, skips deleted rows and bumps the sync counter,
and a take clears the auto-miss flag. Editing a schedule tombstones the
open pending rows of the old anchors and reconciles a stale window onto
the new times. The bulk write path attributes takes through the same
band engine as the single-intake routes and accepts forceSlotInstant
per entry. Configured dose windows now reach every compliance call site
through one shared schedule select, and the card window pill takes a
timezone, treats a degenerate window as a point and counts dose times
instead of schedule rows.
The medication list fired one compliance request per card every 30
seconds, and each request scanned the medication's full intake history
and expanded the recurrence calendar five to six times. The endpoint now
mints bands once per request, serves the sub-windows from that single
ledger, bounds the event read to a year of selected columns and sits
behind a write-invalidated 15-minute server cache; the cards poll at
five minutes. The dashboard drops two redundant checklist queries by
reading counts from the snapshot, the BMI chart derives from the weight
series instead of refetching it, the detail page no longer serialises
behind the auth probe and seeds its query from the list row. Expensive
analytics reads gain a shared per-user rate limit, and the service
worker stops serving cross-version shells, enables navigation preload
and trims caches in one pass.
Raising the middleware body cap to 512 MB for archive imports removed
the implicit ceiling that protected JSON routes, so any bulk endpoint
would buffer an arbitrarily large body before validation ran. Every
JSON body read under /api now carries an explicit byte cap sized to its
legitimate payload (64 KB for single-record routes up to 16 MB for the
measurement import), webhooks cap before signature work, and oversized
bodies return 413.
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.
- Layout PUT accepts version 1 as well as 2 (live iOS client still PUTs
version 1); the serializer normalises to the canonical v2 blob, so a v1
body never 422s. Mirrored in the OpenAPI schema + description; yaml regen.
- Add the Weight tile to the inline tile editor so it can be hidden /
reordered like every other Vitals grid tile.
- Replace the non-transitive Vitals tile-drag comparator with a pure,
total-order rebuild helper (reorder the Vitals subset, substitute it back
into the full list in original-order slots, densify).
- Gated-off section rows keep a disabled eye toggle in the same position
with the gate hint as a caption below the title, so the row never jumps.
- Cap the disclosed Vitals tile list height (max-h-[60vh] scroll,
overscroll-contain) so a tile drag does not fight the page scroll.
- Gate edit mode until the layout GET settles so a save can never flush
defaults over a real saved layout.
- Drop the duplicate top-right "Done" affordance: hide the customize toggle
while editing; the in-card Done / Reset own save and reset.
- Restore focus on enter (edit-card heading) and on close (customize toggle).
Tests: route regression for a version-1 PUT persisting a resolved v2 layout;
unit coverage for the new Vitals rebuild helper (order, non-Vitals stability,
per-tile visibility, no mutation).
Add an additive, optional `sections` array to the persisted insights
layout alongside the existing per-metric `tiles`, with the same
merge/dedupe/dense-order semantics. v1 blobs resolve forward to a valid
v2 layout with every section default-visible; the iOS `tiles` contract is
unchanged and a PUT may omit either array (defaults fill in).
- Phase education card no longer carries its own log-today button (the
top-level log entry covers it).
- Predicted-period calendar days render as a soft dashed fill instead of an
underline, so they read as period rather than an unfinished dash.
- The derived batch route now attaches the deterministic per-score assessment
for the five scores so the dashboard grid (and iOS) paints the explanation
without a second per-id round-trip; guarded so it never fails the grid.
The comprehensive endpoint paired sys+dia on the 5-minute window only, so
readings whose sys/dia timestamps drift past 5 minutes on Withings /
Apple-Health imports were dropped, skewing the in-target share the Home
Health Score renders. Route it through computeBpInTargetPct, which adds the
same-Berlin-day pairing fallback and shares one definition with the
dashboard tile, insight cards, AI snapshot and targets endpoint. Band
(ESH-2023 ceiling + 90/50 hypotension floor) and the 90-day window are
unchanged.
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>
When no AI provider is configured (or before the first AI summary is written),
the period-in-review card stayed empty. Compose a concise, factual, strictly
non-causal summary from the same structured digest the AI prompt is built from
— ranked metric changes vs the prior period, vitals that moved outside their
typical range, and an honest count of statistical associations — and persist it
with a `deterministic` provider marker so a later AI warm replaces it in place.
The read route now warms regardless of provider; the read-never-blocks posture
is unchanged. Fills the no-key demo's retrospective too.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The reference-grounding builder (buildSystemPromptWithReferences) and its
SOURCES list shipped to no one: both live paths called the plain system prompt,
so the model was asked to cite referenceId values it was never shown — and the
parser rejects invented ids, so the citation machinery the schema and UI
footnote already support stayed dark. Wire the builder into both consumers,
passing the metric sections the snapshot actually carries (a weight-only run
never lists BP guidance). Apply the fenced-JSON parse fallback here too.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Close the launch-blocking gap where the ConsentReceipt infrastructure was
built but never enforced as a precondition: a direct API caller with a valid
token could forward a health snapshot to a server-managed external LLM with no
receipt on file. Consent was enforced client-side only.
Add a consent guard (src/lib/ai/consent-guard.ts) and call it before the first
external-provider call on the server-managed path. The gate fires only when the
resolved chain could egress via the operator's global key (admin-openai); a
user's own BYOK key (openai/anthropic), their own ChatGPT OAuth account, and
the self-hosted local provider stay ungated — that egress is the user's own act
and the settings flow is its consent. The check fails closed: a BYOK-primary
chain with an admin-openai fallback still requires a receipt, since the runner
may cascade to the operator key. The surface maps to the consent kind (coach ->
ai_coach, insights -> ai_insights_only; master ai_full satisfies either).
- Interactive routes (insights/chat, insights/generate POST) throw
ConsentRequiredError; api-handler renders a 403 with
meta.errorCode = "consent.ai.required", mirroring assistant.disabled.* so
the client can render an inline grant-consent notice.
- Off-request pipelines fail closed quietly: comprehensive generation returns a
typed skipped:no-consent outcome, and the per-metric status family
(runStatusCompletion, gating every status card + the period-narrative warm +
the off-budget Coach memory workers) surfaces the no-key fallback instead of
egressing.
Also rate-limit POST /api/fitbit/sync (the one Fitbit route without a limiter):
a 5/60s baseline plus a tighter 1/hour bucket on the expensive fullSync walk,
keyed by user id, matching the sibling Fitbit routes.
Harden the Fitbit OAuth redirect_uri: assert it is an absolute https origin (or
http on localhost), lands on /api/fitbit/callback, and stays same-origin with
NEXT_PUBLIC_APP_URL — defence-in-depth behind Google's registered-URI check so
a Host-coerced or misconfigured app URL cannot redirect the authorization code
off-origin.
(cherry picked from commit 46b5fc2cea1b6e80bee0445bfdacf5790442c67d)
Token-by-token streaming was buffering on the server. The chat route
enqueued every tokenised SSE frame synchronously inside the stream's
start() callback, so the runtime coalesced them into a single network
read and the client painted the full answer at once despite the
per-token render path already wired in use-coach. Make the emit callback
async and yield to the event loop between token frames so each one
flushes as its own chunk; the visible cadence now reads ChatGPT/Claude-
style. Refusal and error paths stay single-frame.
The conversation history was a permanent left column on lg+, eating chat
width on laptops and reading as a takeover behind the narrowed drawer
cap. It now lives behind an always-visible "Conversations" toggle on
every viewport and opens as the left-edge tray on demand; the thread
keeps the full width by default and never loses it on mobile.
The provenance block (source chips plus raw key-values) rendered fully
expanded under every answer and was often taller than the reply. Fold
the chips into the existing collapsed disclosure alongside the key-
values so the whole grounding is one click-to-expand surface, collapsed
by default, with the disclosure's aria-expanded/aria-controls intact.
The clinical-decisions disclaimer rendered both at the bottom of the
thread and in the sources-rail footer. Consolidate to a single line
directly above the composer, always visible on every viewport.
The auto-growing composer (single line growing to a capped multi-line
with internal scroll, Enter-to-send, Shift+Enter-newline) and the
aria-live streaming region are preserved. Plain-text React children
throughout; no new dependencies.
Apple Health (and some WHOOP shapes) write BOTH an unspecified ASLEEP
aggregate row AND the granular CORE/DEEP/REM breakdown for the same
sleep period. The per-night reconstruction summed both, so a real ~8 h
night surfaced as ~16 h across every sleep surface — the chart showed
~16-18 h points and the target-range 30-day average read ~16.4 h/day.
Prefer the granular partition: when any CORE/DEEP/REM stage is present
for a night/source, count only those; fall back to the bare ASLEEP (or
a stage-less legacy/manual) row only when no granular stage exists.
IN_BED and AWAKE stay excluded; dual-source nights still collapse to one
canonical source via the user's sleep priority ladder. The same rule
applies inside the per-source asleep tiebreak so a source writing both
shapes is not over-weighted.
Align the surfaces that bypassed the shared helper:
- insights/targets: SLEEP_DURATION was averaged as raw per-stage rows
(minutes) and fed into a target band expressed in hours, producing the
bogus 16.4 h/day. Reconstruct per-night time-asleep, convert to hours,
and classify + roll up consistency on the nightly hour totals.
- coach snapshot: route the sleep block through reconstructSleepNights
so the prompt narrates the same nightly numbers (was summing every
stage row, including IN_BED + AWAKE).
- sleep card chart: read /api/measurements/series?kind=sleep (night
reconstruction, hours) instead of the generic measurement
daily-average path, and label the axis in hours instead of minutes.
Add regression fixtures: a night with both a bare ASLEEP row and the
granular breakdown asserts the granular total (no double-count); a
dual-source night with both shapes per source; and a multi-night 30-day
average that lands at ~8 h, not ~16 h.
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.
The hero strip, daily-briefing card and trends row all read the advisor
payload through `useInsightsAdvisorQuery`, which POSTed `/api/insights/generate`
without `force`. That POST generates inline on a cache miss, pinning the
page-load path on the full provider chain (up to the client's 8 s abort) before
the briefing paints — even though a read-only lift of `User.insightsCachedText`
costs nothing.
Add a read-only GET to the route that serves the cached payload immediately and,
when the cache is stale or missing and a provider is configured, enqueues an
out-of-band warm on the existing `insight-pregenerate` queue. The read mount now
uses the GET; user-initiated regeneration keeps the POST force path. Mirrors the
stale-while-revalidate posture of the per-metric status and period-narrative
reads — no provider call ever sits on a navigation request.
Add the GDPR/trust surface for the durable Coach facts: GET lists the
caller's active facts decrypted, DELETE /[id] soft-deletes one, and a
bulk DELETE clears all — every verb apiHandler + requireAuth +
requireAssistantSurface("coach") + ownership-scoped, with an unknown or
cross-user id collapsing to an idempotent no-op that never leaks
existence. Registered in the OpenAPI contract.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
fetchConversationWithMessages now returns the decrypted rolling summary
(fail-closed → null), and buildHistoryWindow substitutes it for the dead
placeholder so a long conversation keeps memory of its elided turns.
Once a conversation grows past the turn cap the chat turn fires a
fire-and-forget coach-memory-refresh so the next long turn is fresh,
without adding a provider round-trip to the interactive reply.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add a calm "your week / month in review" summary to the Insights
overview, generated from the period-narrative context (deltas, personal
bands, FDR-controlled associations, coincident days).
- New insight_narratives table (migration 0113): one durable row per
(user, period, locale); generated prose AES-256-GCM at rest following
the coach-message precedent; labels-only provenance stays plaintext.
- generatePeriodNarrative feeds the context into the provider chain with
a tight, descriptive-never-causal prompt; an insufficient context
yields no narrative, a missing provider no-ops, neither fabricates.
- Nightly warm cron on week (Mon) / month (1st) boundaries, budget-gated
per user; the read-only GET enqueues a single-user warm on a cold read.
- Read route is stale-while-revalidate and never blocks on the provider;
the card renders plain text (no markdown), a single provenance
disclosure, and a fixed footprint (no CLS). i18n in all six locales.
Promote the volatile per-worker last-working cache to a Postgres
provider-health ledger so the fallback chain shares one health signal
across every instance, modelled on the atomic-upsert rate limiter.
The ledger holds a durable negative cache: an auth-class failure
(401/403) benches that provider for a cooldown instead of re-burning the
dead round-trip on every generation; a sustained hard failure backs off
briefly. The local model is never benched — it stays the guaranteed
floor of the chain. Every ledger interaction fails open, so a DB error
falls back to today's chain walk exactly.
When the user's primary provider fails auth-class, the Coach emits a
distinct credential_expired frame that deep-links to reconnect rather
than a generic "try again later".
Adds the provider_health table (migration 0112) and the
errorCredentialExpired copy across all locales.
The semantic-token swap means the trend arrow/delta now emit
text-success/text-warning (pixel-identical in dark, AA-safe in light), and
the batched derived route joins the insightStatus allowlist alongside the
single route. Update the trend-card sentiment-colour assertions and the
coach-route gate inventory accordingly.
The Insights cold mount fired 14+ independent GET /api/insights/derived
requests — Readiness, Sleep and the vitals baselines are compute-on-visit,
and all of them share one Prisma pool. That is the pool-starvation class
that surfaces as the app hanging then recovering.
Add GET /api/insights/derived/batch?metrics=a,b,c[:type] that resolves the
requested metrics server-side under a bounded p-limit(4) with the profile
loaded once via the shared loadBaselineProfile, returning a map of the same
flat Derived<T> envelope keyed by the request token. Add the useDerivedBatch
hook + the insightsDerivedBatch query-key factory entry; the single-metric
route + hook stay for the per-score detail pages. The single route now also
adopts loadBaselineProfile instead of re-inlining the profile read. Honors
the existing 8s ceiling, retry:0, replace-with-fallback, and warms nothing
on visit. Register the new route in the OpenAPI contract.
When a derived wellness score (readiness or recovery) crosses a notability gate
— outside the green band AND confidence at least medium — the insights
generator now appends a SYSTEM CONTEXT block so the briefing can narrate the
shift observationally. Mirrors the GLP-1 plateau detector's append mechanism:
reads the one derived contract, returns null (the common case) at zero token
cost, and instructs the model to frame the signal as an observed pattern, never
a diagnosis or a dose recommendation. Add the readiness/recovery sourceMetric
enum values plus their briefing href + icon + trend-chart mappings.
Marc-André Bombeck