Apply the prepare -> run -> finalize split to blood-pressure, pulse, bmi,
mood, general, and medication-compliance status, matching weight-status.
Each public generator keeps its signature and one-call behaviour by driving
its prepared card through the shared single-card path; the snapshot build,
cache read, read-only miss, and content-hash gate are now reusable by a
batched caller without changing what each card persists.
Introduce a prepare -> run -> finalize contract for the per-metric status
cards so the snapshot build and the provider call can be driven separately.
The public generator keeps its signature and one-call-per-metric behaviour;
weight-status now resolves through the shared single-card path. Add the
status-batch AI budget entry sized for one combined seven-metric call.
Two parallel changes both claimed 0178; the session-expiry indexes keep
0178 and the briefing-reroll column moves to 0179 so the sequence stays
unique and ordered.
Centralise per-surface AI budgets and right-size status maxTokens; add
seed support (pinned for reference surfaces, seedless for the briefing);
force JSON on Ollama and Anthropic; reroll only the daily-briefing prose
once a day for variety while findings stay stable; reuse the corrective
retry on the comprehensive path; extract shared prompt contracts as one
source of truth for grounding, tone, and the GLP-1 safety rule; reserve
the coach budget atomically before the call and reconcile burned tokens;
turn the local-AI private-host flag into a host allowlist.
`ALLOW_LOCAL_AI_PRIVATE_HOSTS` was a binary flag that, when `true`,
opened every private/internal host at once — including the cloud
metadata endpoint and any internal panel. Replace the binary check at
all three sites (write-time validation, local client, test resolver)
with a shared `isLocalAiHostAllowed` helper: a comma-separated host list
permits only those exact hostnames, while the legacy `true` still
permits any private host for backward compatibility. The safeFetch /
isPublicUrl floor is untouched; the allowlist only widens what the
operator deliberately permits. Docs and the SSRF-pin inventory test are
updated accordingly.
Raise primary touch controls across the non-Settings surfaces to a 44px
floor that condenses at sm+, add a shared ListRow primitive and migrate
the three bespoke list-row mini-cards onto it, and add a scroll-touch
utility for the horizontal scrollers.
Assert the send and stop controls render the `size-11 sm:size-9` tap-
target pattern so a future restyle cannot quietly drop them back to the
36px size the audit flagged. Renders the real component and matches the
emitted class string, the same model as the bottom-nav size test.
Sweep the primary touch controls that sat below the 44px tap-target
floor on phones onto the canonical `min-h-11 sm:min-h-9` (text) /
`size-11 sm:size-9` (icon) pattern, condensing back at `sm:` upward:
- Coach composer mic / stop / send icon buttons (were `size-9`).
- Dashboard hero CTA, link and action variants (were `size="sm"`).
- Vorsorge list-view "mark done" button (was `min-h-9`) and its kebab,
which now uses `size="icon-lg"` so the rendered control is 44px
rather than the icon variant's 40px.
- Illness list-view "log day" button (was `min-h-9`).
- Medication wizard Step 3 injection-sites disclosure (was `h-9`).
The Vorsorge dashboard-card capture action was raised alongside the
ListRow migration.
The bordered list-row "mini-card" appeared three times with three
shapes: the Vorsorge dashboard row (`rounded-md p-2.5`), the daily-
briefing rows (`rounded-md p-3`), and the measurement-list mobile row
(bespoke `rounded-lg p-3` div). Introduce `ListRow` (`rounded-lg border
p-3`, `asChild` per the shadcn convention) and migrate all three to it.
The primitive owns only the box shape; each caller keeps its layout,
colour, and interaction classes, so this is structural normalization
with no content or behaviour change.
Add a `.scroll-touch` utility that restores iOS WebKit momentum
scrolling (`-webkit-overflow-scrolling: touch`) and contains the
overscroll chain so a flick at the end of a horizontal row does not
rubber-band the whole page. Apply it to the insights tab strip and the
cycle tab list — the two in-app horizontal scrollers that lost inertial
flick under the default `overflow-x-auto`.
The daily token cap was read before the provider call and recorded only
after a successful reply, so concurrent requests could all read
`spent < cap` and all hit the provider, and an empty/sentinel reply
recorded zero spend despite burning upstream tokens.
Reserve the budget atomically before the call: a single upsert-increment
returns the new total, so concurrent requests serialise on the row and
cannot both pass the cap; over-cap refunds the reservation and surfaces
`coach.budget.exceeded` as an SSE error frame. Reconcile the reservation
to the actual token count immediately after the provider returns —
before any empty/sentinel short-circuit — so burned tokens are always
recorded. An integration test proves two concurrent requests cannot both
exceed the cap and that an empty reply still records its cost.
Extend the demo tenant with body composition, blood glucose with meal
contexts, cardio vitals, recovery and sleep scores, and realistic workouts
with heart-rate samples, all reconciled against the existing series and
dated relative to now. Replace the stale achievement ids the prior seed
referenced with verified registry keys.
The per-worker in-process "last working provider" hint stays volatile
and is deliberately NOT folded into the durable Postgres health ledger.
The ledger already shares the load-bearing skip signal across workers;
the only uncoordinated piece is a soft reorder whose worst case is one
redundant first-hop per worker per hour. Promoting it to a per-request
ledger write would add hot-path write amplification for a negligible
saving. Documented inline so the trade-off is explicit.
The grounding, tone/banned-opener, GLP-1 dose-safety, metric-identifier
ban and forbidden-filler rules were independently encoded across the
comprehensive, status, Coach and narrative prompts and had drifted. Move
the canonical wording into `prompts/shared-contracts.ts` (locale-keyed
fragments) and compose each surface from it.
Where the surfaces had drifted, adopt the stricter wording: most notably
the GLP-1 dose-prescription contract — previously present only in the
comprehensive and Coach prompts — now reaches the status cards and the
period narrative too. A cross-surface coverage test asserts every
enforced fragment appears verbatim in every surface, in de and en.
Extend the demo tenant so it reads as a long-term, fully engaged user
across the whole app, not a thin sample. All series span the existing
ninety-day window with the latest point on today and sit in healthy
bands:
- Body composition (fat, fat-free, lean, muscle mass, total body water,
bone mass, visceral fat, BMI) derived from the same-day weight and
body-fat series so the numbers reconcile with the headline tiles.
- Blood glucose in a healthy non-diabetic band — fasting, post-meal and
bedtime readings stored in mg/dL with the glucose context tagged.
- Cardio fitness and vitals: HRV (SDNN and RMSSD), oxygen saturation,
respiratory rate, periodic VO2 max, active energy, walking/running
distance and flights climbed.
- WHOOP-style recovery, day strain and sleep performance/efficiency/
consistency scores aligned with the seeded sleep nights.
- Three to four workouts a week (running, strength, cycling) with
duration, energy, average and max heart rate, distance, a sparse
per-workout heart-rate sample series, and per-workout strain in the
metadata blob.
Clear the workout tables in the same truncate pass so the seed stays
re-runnable.
The demo seed unlocked twelve achievement ids that no longer exist in
the registry (first-measurement, week-streak-measurements,
three-month-streak, …), so the achievements surface rendered them as
unknown placeholders. Map the unlocked set to live registry keys —
first-entry badges, logging and in-range streaks, compliance and
miss-free streaks, plus the self-context and seven-night sleep badges —
each on a plausible unlock date the seeded history supports.
The medication card and table render dose-derived stock from the cached
medications list, but the inventory write routes evicted no server cache
and the client invalidated only the per-item inventory key, so a new
container's stock kept serving the pre-write value for the stale window.
Hard-evict the user's medication caches on every supply write and
invalidate the list key alongside the inventory key on the client.
The /medications card and table render the dose-derived stock
(stockUnitsRemaining / stockDosesRemaining) straight from the
GET /api/medications list payload, which reads through cachedSwr on the
60s-fresh + 10-min-stale medications bucket. Registering, adjusting, or
deleting an inventory container changes that stock, but the inventory
routes never invalidated any cache — so the list served the pre-write
stock for the rest of the stale window. On the client the add / adjust /
delete supply mutations only invalidated the per-medication inventory
query, never the medications-list query that the card reads, so a mounted
list never refetched even once the server caught up.
Hard-evict the per-user medications + compliance buckets from every
supply-affecting write (inventory POST, the per-item PATCH / DELETE, and
the deprecated glp1 POST dose-change + ledger branches) so the next read
reflects the change rather than a stale SWR value. On the client,
centralise the supply mutations on a shared invalidateSupplyQueries helper
that drops both the medications-list key and the per-medication inventory
key. The card's neutral styling is unchanged.
Fold the three standalone module settings pages into the Settings shell so
they open in-context with the same left-nav, heading, and chrome as every
other section. Remove the standalone pages and the now-unused
ModuleSettingsFrame, and drop the orphaned settings i18n blocks.
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.
Promote the three per-module customise surfaces from standalone
ModuleSettingsFrame pages to first-class Settings sections. They now
render inside the Settings shell — same left nav, heading, and chrome as
account / ai / integrations / notifications — instead of their own
back-button-and-title frame outside it.
Add labs, illness, and vorsorge to the slug registry (after mood) and to
SETTINGS_SECTIONS with FlaskConical / Thermometer / CalendarCheck icons.
labs and illness gate on their toggleable modules; vorsorge has no
module, so its entry is always shown. Three thin section components wrap
the existing LabsSettings / IllnessSettings / VorsorgeSettings content,
mirroring MedicationsSection / CoachSection.
Delete the three standalone /settings/{labs,vorsorge,illness} routes so
the wrench deep links resolve through the dynamic [section] route, and
drop ModuleSettingsFrame now that nothing renders it. Add the section
title / subtitle / description keys across all six locales and remove the
dead per-module *.settings.* blocks the old pages consumed.
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).
Recent in-band BP, stable healthy weight, a resting-heart-rate series, a
full week of per-stage sleep nights, on-track medication doses, two
preventive-care reminders, a healthy lab panel, one resolved illness
episode, and a sample Coach conversation — all dated relative to now so
the public demo reads current for screenshots.
The measurement-reminder push already carries the reminder id in its
metadata, but the iOS allowlist stripped it before it reached the device.
Forward the key so a foreground or handled notification can post the
completion directly instead of routing through the capture deep link.
Tune the public demo tenant for App-Store screenshots:
- Extend the measurement loop to include today so every headline metric
has a current reading.
- Retarget blood pressure into the optimal band (~118/76), weight to a
stable healthy BMI, and add a clean resting-heart-rate series so the
resting-pulse tile scores it directly instead of the spot pulse.
- Write a full week of per-stage sleep nights (awake/REM/core/deep,
~7.5h, source APPLE_HEALTH) including last night, in minutes, and skip
the aggregate row for those nights to avoid double-counting.
- Schedule today's doses on-track: morning meds taken, the evening dose
pending and never auto-missed, so today never reads overdue or 0%.
- Add Vorsorge reminders (dental check-up, annual physical), one healthy
lab panel, one resolved illness episode, and an AI-configured Coach
with a short encrypted sample conversation.
All dates stay relative to now so the demo refreshes on every re-seed.
Add an opt-in `responseFormat: "json"` to CompletionParams that the
structured surfaces (comprehensive briefing + per-metric status cards)
set. On local (Ollama / OpenAI-compatible) it adds `format: "json"`; on
Anthropic it prefills the assistant turn with `{` and re-prepends the
brace to the reply, so the first emitted token is already inside a JSON
object. Both cut the first-pass JSON-failure rate that fence-stripping
otherwise has to absorb.
OpenAI already pins `response_format: json_object` and is unchanged. The
period narrative (plain text) and the Coach (streaming prose) leave the
flag unset, so their non-JSON paths are untouched.
Add an optional `seed` to CompletionParams and thread it onto the
OpenAI and local (Ollama / OpenAI-compatible) request bodies. The field
is dropped when unset, so existing callers are byte-identical.
Pin a stable seed for the status/reference cards (defaulted in the
status provider) and the period narrative, so a prompt change is
diff-able against a fixed baseline rather than masked by low-temperature
non-determinism. Anthropic's Messages API has no seed knob, so its
client ignores the field (documented inline). The seed is annotated on
the narrative wide-event.
Fix the target-edit-sheet invalidation that used a bare ["insights"] array
not matching the factory prefix the reads nest under (a silent no-op), and
migrate 12 further bare queryKey call sites onto the centralised factory.
Replace the scattered `?? 0.3` / `?? 1000` temperature and maxTokens
magic numbers across the status provider, Coach route, comprehensive
generator and period-narrative generator with a single typed
`ai-budgets.ts` map. Each entry documents the output contract that
fixes its token ceiling.
Right-size the per-metric status output from 1000 to 250 tokens: the
card renders one `{ "summary": "..." }` of 2-4 sentences (30-60 words),
so 1000 was an order of magnitude over the real ceiling.
The target editor invalidated a bare ["insights"] array on save and reset.
That literal sat outside the centralised factory, so the broad insights
fan-out it intended never matched the reads, which key through
insightsTargets/insightsRoot. Route the editor's reads and invalidations
through userThresholds(), insightsRoot(), insightsTargets() and analytics()
so a saved target actually refreshes every insights surface.
Migrate the other call sites the lint rule does not reach onto the factory:
the Coach conversation rail + thread reads, the per-metric personal-records
badge, the per-day measurement drill-down, the dashboard intake
compliance-chart fan-out, the per-medication side-effects list, the
onboarding auth invalidations, the notification-preferences read, and the
PWA persister's auth-me lookup.
Add the matching factory keys: coachConversations, coachConversation,
personalRecordsByMetric, measurementDrilldown, complianceChartInline and
medicationSideEffects. Each new key reuses the prefix its reads already use
so no invalidation shape drifts.
The dev-only SHA-256 key padding was gated on the absence of a production
marker, so a staging or preview container with NODE_ENV unset silently
derived a weak key the operator never recorded — rotating to a real key
later orphaned those rows (decrypt is fail-closed). Gate the padding on an
explicit NODE_ENV of development or test; every other case, including unset
NODE_ENV, throws the same hard error as in production.
coach_messages is append-only and grows forever — a daily Coach user
accrues encrypted conversation rows with no lifecycle, while every
sibling ledger (push_attempts, audit_logs, host_metrics) already prunes.
Add a daily pg-boss cleanup that hard-deletes rows past a configurable
window (COACH_MESSAGE_RETENTION_DAYS, default 365, 30-day floor so a
misconfig cannot drop live in-context history). Wire it into the
maintenance registrar with the dead-queue contract guard and a focused
helper test.
The sliding-30-day session reaper and the passkey-challenge sweep both
filter on expires_at, but neither sessions nor auth_challenges indexed
that column, so the cleanup degraded to a sequential scan on a busy
instance. Every sibling expiry table already indexes its expiry column;
add the two missing indexes via migration 0178.
The floating Coach launcher inherited the button's dark foreground token,
which left the sparkle glyph as a dark outline on the purple-pink
gradient. The glyph is decorative and the button carries its own
accessible name, so the white mark reads cleaner without affecting the
control's non-text contrast, which the offset focus ring already
provides.
The expired-nonce callback test asserted the global fetch spy was never
called. A preceding successful-connection test fires its webhook
subscriptions fire-and-forget, and those promises resolve into the next
test's freshly installed spy, so the broad assertion failed whenever the
background calls landed in time. Filter the recorded calls to the token
endpoint and assert none reached it, matching the replay and cross-user
tests in the same file and keeping the check true to its intent.
KeyFindingRow and SignalRow rendered an identical tone bar + metric icon
+ headline/body/delta and the same link-when-routed behaviour, differing
only in their data-slot and body copy. Extract a single BriefingRow and
have both delegate to it.
The demo seed's random-walk BP generator could drift below the
physiological floor and mint a systolic-0 row (iOS #33). Clamp the walk
to the same band the input validator enforces (systolic >= 40,
diastolic >= 20), reusing the VALUE_RANGES floors, so a fresh seed never
writes a sub-floor reading. Add a read-side selection guard on the blood
pressure series endpoint that excludes rows below those floors from both
the series and the implicit latest point, so a corrupt row that slipped
in before the floor was enforced can never surface as the latest
reading. Export VALUE_RANGES so both sites share one source of truth.
Add aria-expanded to the coach drawer's history-collapse control so its
expanded state is exposed, matching the expand toggle. Give the Vorsorge
trend strip an sr-only summary of the latest value and seven-day
direction, and key the bars on measuredAt rather than the array index.
Remove the integration keys left dead after the integrations refactor:
the WHOOP and Fitbit overlap notes, the Fitbit experimental note, the
per-provider credentials-help strings (WHOOP, Fitbit, Polar, Oura) and
the Nightscout help. None carry a t() call site any longer. Add the
Vorsorge trend-strip screen-reader summary strings (latest value plus
seven-day direction) across all locales.
Add a coarseOnly option to buildTieredSeries that skips the raw 0–14d
read and the DAY (14–30d) / WEEK (30–90d) band reads, returning only the
MONTH / YEAR bands plus the anomaly envelope. The Coach snapshot's
coarse timeline tail consumes only month/year + anomalies, so the
near-term bands were read and discarded every turn; it now passes
coarseOnly. The briefing path is unchanged. Look the learn-guide prompt
example up by slug rather than by array index so reordering the catalog
cannot silently change the example link.
Drop steps from the signals-of-day set: Apple Health steps ingest as
many intraday stats-prefixed samples, so the newest raw row is a
partial-day fragment, not a daily total — same exclusion already applied
to per-night sleep. Add a glucose signal off the canonical stored value
(the model reads the snapshot, not display units) so the block matches
the priority comment that already claimed it; drop the dead sleep entry
from SIGNAL_PRIORITY. Track the raw extreme magnitude when keeping the
most extreme recent anomaly so a larger anomaly is no longer dropped by
comparing against an already-rounded value. Thread the injected clock
through trendSlope so the emerging-trend cutoff is deterministic under a
test clock. Rename the anomaly day-count to anomalyDaysAgo so the
snapshot hash staleness-tiers it instead of letting the raw count drive
the fingerprint. Cap the returned set at three to match the schema.
Wire the Coach snapshot's core-metric history onto the shared tiered
series so blood pressure, weight, and pulse carry a coarse tail the
in-window weekly fold could not produce — 90d-1y monthly and >1y yearly
buckets plus the preserved peak/trough envelope, so a spike from months
ago survives a coarse bucket. The tiered read skips the raw 0-14d band
the snapshot already holds, so no duplicate round-trip; the coarse tail
is bounded per band and sheds first under the prompt-budget degrader, so
token size holds.
Surface the active illness episodes and Rest Mode state to the Coach
prompt: the snapshot already carries the illness block, and the system
prompt now instructs the Coach to read an off-band vital as plausibly
illness-explained and avoid pushing a heavier cadence while the user is
in Rest Mode.
Add a static catalog of the public /learn guides and inject a compact
list into the system prompt. The Coach may reference at most one guide
by topic and URL when genuinely helpful, and only a slug the catalog
lists — a fabricated /learn URL is impossible by construction.
Every Settings → Integrations block now paints one consistent shape:
heading (icon + name + status pill) → a white primary description → a
grey secondary description that ends with the setup-guide link rendered
inline, instead of a standalone bottom-of-card element.
OAuth-only (WHOOP/Fitbit), BYO-credential (Withings/Oura/Polar) and
manual-key (Nightscout) cards share the same skeleton via the new
IntegrationCardDescription helper; the credential form stays as an extra
block where it applies. The "where to find these credentials" hint moves
above the Save button instead of sitting below it.
Rewrite the per-provider description strings across all six locales into a
parallel two-part shape with even length, and drop the now-orphaned
overlap/experimental sub-notes and the standalone setup-guide link.
Wave A removed the inline ask-the-Coach / suggested-questions block from
the overview hero (Coach is reached via the FAB + drawer). Update the
stale assertion that still expected onAskCoach/onPickPrompt + the launch
context on the page.
Marc-André Bombeck