Add WHOOP-native DAY_STRAIN (0-21) to the Coach snapshot as a dedicated
dayStrain block, kept distinct from the computed STRAIN_SCORE (0-100) and
preferred native-over-computed, mirroring recovery resolution. Document the
glucose.clinical panel (TIR/GMI/eA1C/CV) plus the stillLearning and
isSpotEstimate gates in the system prompt, and the native dayStrain scale in
the derived paragraph. Extend the DE and FR/ES/IT/PL evidence-block unit lists
with mg/dL, mmol/L and min so a non-English coach can cite glucose and duration
values. Export GLUCOSE_PANEL_WINDOW_DAYS as the single source for the 30-day
clinical-panel window and import it across the analytics route, dashboard
snapshot, and Coach snapshot. Compute the derived metrics concurrently.
Hard-coding isSpotEstimate: true mislabels a dense Nightscout CGM stream
(~288 readings/day) as a spot estimate and undersells the data in the
clinical panel and coach. Compute readings-per-day over the covered span
and treat the series as continuous below the spot caveat once it clears a
~hourly bar (24 readings/day); sparse spot fingersticks keep the caveat.
The panel only renders the spot-reading line when isSpotEstimate, and the
OpenAPI field relaxes from a literal-true to a boolean.
Surface the server-authoritative sleep-rhythm read in the coach snapshot
so a reply can reference cumulative sleep debt and the chronotype band the
Sleep page and dashboard already show. The values come from the same
assembler those surfaces use over a fixed trailing-42-day window, never a
recompute, so the coach can only quote what the page renders.
Hold the learning gate: the chronotype band and social jetlag are emitted
only once the signal is ready; a calibrating chronotype is surfaced as such
and never asserted as a band. Document the new fields and minute units in
the system prompt.
WHOOP stamps a recovery on the wake morning while the computed proxy files the
same night under the day that ended, a systematic off-by-one that left both
rows alive as two competing recovery days once the native series was read.
Resolve recovery by a source-aware wake-day key in the user's timezone — the
computed proxy's day is shifted forward to meet WHOOP's wake morning — so one
night collapses to one canonical row without re-dating the stored stamp and
without a backfill. Thread the same resolution through the wellness reader, the
coach nudge trend, and the coach snapshot, which had each read recovery raw.
The native WHOOP recovery and the computed proxy stamp the same night on
opposite clocks: WHOOP files the wake-morning instant (day D), the proxy files
noon-UTC of the day that ended (day D-1). Bucketing raw measuredAt by calendar
day leaves both alive as two competing recovery days.
Bucket each source by the user's local wake-day instead — shifting the proxy's
day-that-ended stamp forward one day to meet WHOOP's wake-morning stamp — so a
single night collapses to one canonical row while distinct nights stay
separate. Resolve the coach-nudge recovery trend and the coach-snapshot dedup
through the same path so no surface blends the two series or drops a genuine
native value on a numeric tie.
Route the derived Sleep Score through the canonical night reconstruction
instead of its own UTC-day grouping, so a night logged by more than one
source is deduplicated to a single writer rather than summed. The score and
the readiness that folds it now read true per-night totals; multi-source
nights drop to their real figure. Emitted minutes and the night stage map
round to integers.
Resolve recovery to one canonical series — WHOOP-native when present, the
computed proxy as fallback — in the wellness reader, the doctor report, and
the coach snapshot, and stop presenting the proxy and the native score as two
numbers. A WHOOP-connected account's recovery now reads the native value on
the tile, the chart, and the printed report alike.
RECOVERY_SCORE is written by two sources for the same day: the WHOOP-native
percentage from the device sync and the COMPUTED proxy from the nightly
readiness job. The wellness reader hard-filtered to source COMPUTED, silently
dropping every ingested native row, and the doctor PDF blended both sources
into one min/avg/max. The COMPUTED proxy also equals the readiness blend
verbatim, so the Coach prompt carried the same number under both READINESS and
RECOVERY_SCORE.
Add resolveCanonicalRecovery: per day, keep the WHOOP row when present, else
the COMPUTED proxy. Drop the COMPUTED-only filter for RECOVERY in the wellness
reader and resolve to the canonical row; STRESS and STRAIN stay COMPUTED-only.
Resolve the doctor PDF wellness summary the same way so the report reads the
native value, not a proxy/native blend. Dedup the Coach snapshot: omit the
recovery line when it mirrors readiness (the no-native case), keep it when a
WHOOP-native value differs. Document the WHOOP-over-COMPUTED resolution on the
sync delta feed, which already mirrors each row with its source so the client
resolves one canonical series rather than charting two.
A WHOOP-connected user's recovery tile, chart, and PDF flip from the proxy to
the native value.
Add aiAvailable + managedBy to GET /api/user/ai-provider so a server-managed
or local provider is reported as usable even without a personal key. Resolution
origin comes from a presence-only helper that mirrors provider resolution
order; no admin keys or endpoints are exposed.
Let a user pick the Sensiplan double-check's second indicator — cervical
mucus (default) or cervix observation — mirroring the published NFP
standard where the temperature primary sign pairs with one chosen
secondary sign.
- `CycleProfile.secondarySymptom` (MUCUS default | CERVIX) drives which
second sign the engine confirms against. Casual users keep temperature
+ mucus untouched; the choice lives behind the advanced cycle settings.
- `CycleDayLog` gains the three standard cervix signs — position
(LOW/HIGH), firmness (FIRM/SOFT), opening/os (CLOSED/OPEN) — as nullable
columns (migration 0157), wired through Zod, the day-log upsert, the
DTO, the sync feed, backup/restore and OpenAPI. Additive and
back-compatible: existing profiles backfill MUCUS, existing day-logs
keep NULL cervix signs, iOS inherits the fields.
- The engine's `confirmSymptothermal` takes the secondary-symptom choice;
on CERVIX it confirms against `detectCervixPeak` (last fertile-cervix
day followed by three closed days — the cervix-closure rule) instead of
the mucus peak. The double-check still requires BOTH the temperature
shift AND the secondary peak: a single cervix sign never confirms, and
the still-learning gate and conservative posture are unchanged.
Engine derived clean-room from the Arbeitsgruppe-NFP / Raith-Paula &
Frank-Herrmann standard.
A fever, illness, alcohol or a late/short-sleep reading can fake or mask
a temperature rise. Add a per-day `temperatureExcluded` flag on
CycleDayLog (migration 0156): when set, the engine drops the reading from
the cover-line baseline AND the shift evaluation, so a spike neither
fabricates nor masks a rise (Sensiplan: the cover line is drawn over the
last six unbracketed values).
Wired end to end — schema + Prisma, the engine's `detectTempShift`
filter, the Zod input/patch schemas, the day-log upsert (partial merge +
change detection), the read DTO, the calendar/insights/coach selects, the
backup round-trip, and the OpenAPI contract (a new boolean on the day-log
and calendar-day DTOs for the iOS client). The log-day sheet gains a
"mark reading as disturbed" toggle that appears once a temperature is
entered, translated across all six locales.
Citation: de.wikipedia "Symptothermale Methode"; myNFP "Temperaturkurve
auswerten" — Störfaktor bracketing.
The dose-ledger adherence rate divides taken by taken + missed (deliberate
skips excluded), so the PDF compliance table's denominator column no longer
equals Taken + Skipped + Missed. Relabel it "Expected" so each row stays
internally coherent — Taken + Missed = Expected, with Skipped shown alongside
as informational and the rate reading Taken / Expected.
Also tighten the coach adherence-block wording: the headline is the
dose-weighted pool of taken / (taken + missed) across the user's scheduled
medications — exactly one medication's card rate for a single-med user, and the
honest overall adherence for several — never a per-day or per-week denominator
of its own.
The coach snapshot built its own medication-adherence figures: a per-day and
per-week timeline tallied off a "countable ledger rows in window" denominator
that differs from the cadence-aware `calculateCompliance(...).rate` every other
surface — the medication card, the detail page, the dashboard pillar, the
insights — reads. On an irregular, rolling, or partial-window schedule the two
denominators diverge, so the coach could reason from a rate the card never
shows the user.
Route the coach's headline adherence % through `calculateCompliance(...).rate`
with the medication context (the ledger path), aggregating taken / (taken +
missed) across the user's scheduled medications over the coach window — exactly
the arithmetic the card's rate uses. The snapshot now carries a single
`compliance.rate` headline alongside the supplementary timeline, so the coach
quotes the same number the card renders.
A dose can now consume a fraction of a unit — a half, third or quarter
tablet for a split pill. The wizard offers a curated fraction set beside
the whole numbers (the most error-resistant input); units-per-dose and the
per-container counts widen from Int to Decimal, so a 30-tablet pack dosed
at a half reads 29.5 after one dose and the days-left projection follows.
The consumption ledger stamps and refunds the exact fraction, and existing
whole-number doses are untouched. OpenAPI regenerated — the three unit
fields move from integer to number — and the units-per-dose hint is
updated across the six locales.
Also anchors a first-use date relative to now in an inventory route test
that had aged past its 30-day in-use window.
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)
- The legacy GLP-1 ledger reads again: GET and the Coach snapshot fall
back to ledger-derived pen/dose figures when a medication has no
inventory items; items always win once they exist. The delta branch
on the POST is marked deprecated in favour of the inventory endpoints.
- Expired containers count as available nowhere: one shared supply
summary (ACTIVE/IN_USE with units) feeds the overview row, the
Bestand summary, the list payload, the GLP-1 card, and the runway
estimate; expired units surface as a muted suffix instead.
- The inventory endpoints speak one wire dialect: request fields renamed
to unitsTotal/unitsRemaining to match the responses, OpenAPI regenerated.
- The reorder dialog mirrors the page's grouping: active and inactive
sections, each sortable within itself, so a saved order can never
interleave them.
- Polish strings avoid the 2-4 plural trap, sub-dose quantities render
as below one dose instead of approximately zero, and the table's name
sort collates by the active locale instead of hardcoded German.
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.
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.
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 remember action under every persisted chat message adopts its content
into the matching self-context field — allergy statements land in
allergies, diagnosis wording in conditions, everything else in the
coach focus — reusing the adopt endpoint's dedupe and encryption, with
the result visible and editable where the self-context lives. The
deterministic fact patterns accept closed filler-word and possessive
variants that previously defeated them, and stored facts survive the
snapshot budget squeeze instead of being the first block shed on
data-rich accounts.
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 self-context adopt runs read-modify-write inside one transaction,
the admin shell renders nothing until the role is confirmed, and the
json-body helper merges Headers instances correctly. The compose image
reference accepts a pinned digest through an env var, and a promote
script resolves the published digest, verifies its signature and hands
the operator the exact pinning call — the documented path away from a
moving latest tag.
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.
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.
The pregenerate pass now warms the status cards even when the
comprehensive step is budget-blocked, failed or timed out — the gap
that left cards generating on first visit despite the nightly runs —
and the loop gains a per-user timeout so one hanging provider cannot
stall the tick. Schedule edits recompute their rollup days immediately,
the compliance cache keys on the timezone, open doses no longer count
against the nudge trigger, an invite sent against an open registration
still lands in the ledger, the self-report prompt fence uses explicit
data markers, the maintenance crons invalidate the caches they touch,
and a flaky audit assertion polls instead of racing. Six care-focused
achievements join the engine: miss-free streaks, measurement routine,
a complete self-context and a sleep-log run.
The freshly released metadata-action v6.1.0 exits without output on the
pull-request build path; both call sites pin back to the proven v6.0.0
commit. The dead-code gate also caught an unused read helper and an
unused analytics type left behind by the wave — both removed.
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 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.
Blood-pressure fixtures in the in-target integration suite were copied
from live readings; they are now synthetic with the same in/out split so
every assertion holds. Hard-coded operator hostnames move behind env vars
(DEPLOY_LOGS_URL, HEALTHLOG_PROD_URL), the codex client identifies via
the public repository URL, the OpenAPI servers entry points at a neutral
example host, and remaining workflow, docs and comment references to the
operator's instance or working notes are generalised. The compose
whitelist also gains DEPLOY_WEBHOOK_SECRET, which was missing and made
the deploy webhook unusable on compose installs.
Warm the insight generator (briefing + key findings + insight prose) and
the Coach to the same premium-motivating advisor register already shipped
for the per-metric assessments. The intro and role framing drop the cold
clinical-summariser voice for a warm advisor who reads the user's data
against their own history, names earned wins, and frames an unfavourable
finding as one doable step.
Every safety and quality guard is preserved: no diagnosis, no
prescription, no platitudes, no bare number-echoing, honest about
unfavourable values, the banned positivity openers, the doctor-consult
call-to-action, and the JSON output schema + keyFindings tone enum stay
byte-identical. PROMPT_VERSION 4.27.0 -> 4.28.0 (also covers the Coach,
which imports it). Native FR/ES/IT/PL personas warmed to match.
The per-metric and wellness-score assessment prompts gain a premium-advisory
layer: name the earned win, build a little momentum, frame an unfavourable
finding as an opportunity, and close forward-looking. Every existing guard
stays in force — no platitudes, no bare number-echoing, autonomy-supporting,
never alarming, moralising or diagnostic, and the banned-opener and
forbidden-phrase lists are unchanged. The composite-score archetype warms to
a WHOOP/Oura-style recovery coach while staying descriptive, not clinical.
PROMPT_VERSION 5.0.0 -> 5.1.0 so the provenance can slice prose before/after
the tone edit. Cached accounts converge to the new voice within ~24h.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Push the cycle-reminder cron's gating into the query instead of loading
every CyclePrediction row each tick: filter on the user's
cycleTrackingEnabled / predictionEnabled relation and restrict to rows
whose nextPeriodStart falls inside a tz-generous [today-grace, today+lead]
string window. The per-tick cost now scales with the in-window cohort, not
the whole table; the exact per-user-timezone window still runs in JS.
Make the period-boundary shortcut transactional and idempotent: wrap the
close-prior + open-new (or stamp-current) cycle mutations in a single
$transaction so two concurrent taps cannot double-close a prior cycle or
compute a length against a stale read, and add withIdempotency so an exact
key replay is a no-op. The boundary day-log now ranks the incoming flow
against any existing same-day entry and never downgrades a richer flow.
Decode sync cursors additively: a live beta v1 token keeps its
measurement / mood / intake watermarks and simply fresh-scans the two
domains added in v2 (cycleDays, cycles) rather than forcing a full
re-init. Correct the stale "no live consumer" comment.
Thread the user's timezone through the Coach cycle snapshot so "today",
day-of-cycle and days-until-period derive from the user's zone — matching
the calendar and insights routes — instead of wall-clock Berlin.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add the menstrual-cycle phase as a channel in the existing FDR-guarded
correlation machinery, the same way rated mood factors were folded in —
no new statistics, the same Welch + Benjamini-Hochberg pipeline, the same
day floors and q = 0.10 gate.
Two complementary mechanisms, both confined to a gender-gated surface:
- a categorical luteal-vs-follicular contrast per outcome metric (resting
heart rate, HRV, sleep, steps, weight, basal/wrist/skin temperature),
surfacing one headline finding (resting-heart-rate-by-phase, falling
back to HRV); and
- the continuous phase ordinal folded into the lagged-Pearson matrix.
Phase never reaches the general correlations route — the temperature
channels held out of the general matrix are un-blocked only here, behind
the cycle gate, so a reproductive inference can never leak onto a
non-gated surface. Every row shows n, effect size, p and q; nothing is
fabricated on thin data and nothing is framed as causal.
Surface it through a new gated GET /api/cycle/insights and a drop-in
card that mirrors the mood-factor crosstab.
Add a cycle dimension to the Coach snapshot: current phase + day-of-cycle,
the next predicted event (period range + confidence + method, fertile
window goal-gated), and the headline phase finding — present only for a
cycle-enabled account, so a non-cycle snapshot is unchanged. A new ground
rule keeps the Coach descriptive: never contraception-grade, never a
"safe day" claim.
OpenAPI regenerated; phase-correlation + snapshot-inclusion tests added;
copy localised across all six locales.
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>
The BYOK Anthropic client fell back to the retired claude-3-5-sonnet-latest;
match the chain-runner default (claude-sonnet-4-6). The admin ai-settings GET
displayed a gpt-4o-mini fallback that disagreed with the actual provider
default (gpt-4o); align the display string so the surfaced model matches what
the provider would use.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The v1.12.1 diversity block (variety lead, data-strength line, steady-run
repetition signal, FDR-screened cross-metric relations) reached only the
generic archetype cards. The seven hand-tuned headline cards — the ones users
see most — ran at temperature 0.3 with only the previous-context block, so they
stayed prone to repetitive openers and forced filler steps. Thread the same
assessmentContextBlock through the specialized get*UserPrompt builders and
raise their temperature to 0.45 to match the archetypes.
Soften the unconditional "close with one step" across the specialized sections
and the activity / body-composition / sleep archetypes to the base prompt's
conditional form — close with a step only when the finding implies one,
otherwise affirm and name one thing to watch — so a steady metric no longer
manufactures a filler step. EN/DE parity preserved; guard computeSteadyRun
against an absent graded slice.
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>
The insights ground-rules carried changelog tokens ("v1.4.16 phase B8", "W4d")
inside the prompt strings the model reads — token waste and inconsistent with
the project voice. Move the rule content to clean numbered rules; the markers
belong in code comments, not shipped prompt prose. Update the test pins to the
cleaned rule headers.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The OpenAI BYOK + admin-shared fallback defaulted to a small model and
Anthropic to a 2024 Sonnet, so a user on the operator's shared key got the
weakest model for the Insights assessments. Bump the OpenAI default to gpt-4o
and the Anthropic default to claude-sonnet-4-6; both stay overridable by user
config. Sync the provider docs.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The per-metric assessment prompt (base-system) and the scope-hardened
insight system prompt (insight-generator) each carried EN and DE as two
parallel template literals kept in sync by hand — a standing drift risk
flagged in the AI-quality audit.
Model each prompt as one section skeleton: a single ordered list where
every section supplies its EN and DE fragment side by side, and a builder
joins the fragments with a blank line. The section order and prose are
unchanged; the DE calibration is preserved verbatim. A section edited in
one locale but not the other now shows up as an asymmetric edit in one
list instead of two literals silently diverging.
Pure refactor — every emitted prompt string is byte-identical for both
locales across every metric, surface and reference combination. A new
locale-skeleton-parity test pins the structure (each section carries both
fragments, composed prompt equals the joined fragments); the existing
grounding, filler-ban, forbidden-phrase, refusal and exact-phrasing guards
pass unchanged.
(cherry picked from commit 12cd718d6d602a48cafc14ad06d82282fdaacb27)
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)
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)
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.
The snapshot builder ran its independent reads sequentially on a cache
miss: feature extraction, the shared measurement query, the mood,
compliance, sleep and workout table reads, and the GLP-1, derived,
trajectory and memory helper blocks each awaited in turn. None consumes
another's result, so they now fire concurrently and resolve in batched
hops — feature extraction and the measurement read first (both gate the
per-metric blocks), then the remaining reads in one Promise.all. The
source guards are unchanged, so a disabled source still issues no query,
and the synchronous block assembly keeps its original order, so the
provenance envelope is byte-for-byte the same. Cuts several round-trips
off the cold build; the warm path is cached and unaffected.
- Drop the unused `now` field from the coach `ExtractOpts` type; fact
extraction never read it (conversation-summary keeps its own `now`).
- Correct the no-ladder tiebreak docstring in measurement-read: the
collapse picks the alphabetically smallest source name, matching the
live-SQL `ORDER BY … source` tiebreak, not max count.
- Replace the hand-inlined canonical-source subquery in the summaries
rollup-path narrows query with `canonicalMeasurementsFrom(rank, "90 days")`
for live/rollup parity with its siblings.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Add requirePublicHost to the three outbound safeFetch sites whose host is
user- or operator-controlled and lacked the connect-time pin: the OpenAI
and Anthropic clients (BYO base-URL override) and the IP-geo lookup
(IP_GEO_LOOKUP_URL env). A base URL resolving to a private/metadata
address is now rejected. The LOCAL AI client intentionally stays
conditional (LAN escape hatch). Redirect-follow is already closed by the
safeFetch defaults.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
