Run prettier --write on the four files the v1.4.37 W3 commits
touched (the schema, the route, the V2 component, and the two test
files). Drive-by also collapses two pre-existing whitespace drifts
in the InventoryItem type aliases and the glp1PostBodySchema refine
call so the file matches the formatter contract end-to-end. No
behavioural change.
Add two SSR fixtures to lock the v1.4.37 W3 fix in place: a mixed
taken+skipped seed asserts the distinct lucide icon + green styling
for taken rows and the muted outline styling for skipped rows, and a
German-locale skipped-only seed proves the table body never leaks the
"Eingenommen" label onto a skipped row. The seed helper threads the
new `status:"completed"` cache-key shard so the component reads its
own seed without firing a network call.
Marc reported that rows with `takenAt:null AND skipped:false` rendered
as "Eingenommen" with an empty leading column on the medication
detail page — the binary `event.skipped ? "Übersprungen" : "Eingenommen"`
mapping never accounted for the "missed / never confirmed" branch the
v1 component used to surface as "verpasst". The list now pins
`?status=completed` so those ambiguous rows stay off the detail-page
table (they remain visible on the calendar / today surfaces), and the
status cell branches explicitly: green secondary chip with a check
icon for taken rows, muted outline chip with a skip-forward icon for
skipped rows. Derived from `!skipped && takenAt` rather than `skipped`
alone so a malformed row cannot sneak past as "Eingenommen" again.
Extend the list endpoint with an optional `status` query knob —
`all` (default), `taken`, `skipped`, `completed` — so the detail-page
intake history can hide unconfirmed rows server-side without changing
the byte-stable contract the iOS Swift client and the existing
dashboard tiles depend on. `completed` (taken OR skipped) covers the
detail-page surface; `taken` / `skipped` are there for any future
single-arm consumer. Co-located route test pins the Prisma where
fragment per branch and the 422 on unknown values.
When DAY-bucket coverage exists for both BLOOD_PRESSURE_SYS and
BLOOD_PRESSURE_DIA, the /api/analytics default slice now derives the
five in-target windows (7d / 30d / all-time / prior month / prior
year) from rollup mean values instead of paging the trailing 365 days
of raw measurements. The new helper pairs SYS and DIA per-day means
by day key, runs the existing isBpReadingInTarget predicate, and
buckets days into the reporting windows. Partial-coverage and
brand-new accounts fall back to the legacy chunked aggregator so the
per-event pairing contract still holds where the rollup table cannot
guarantee parity. A path:"rollup"|"live" annotate on the meta dict
makes the branch selection visible in prod logs, matching the
v1.4.36 summaries-slice pattern.
The per-day-mean approximation is documented in the helper's header.
For typical multi-reading-per-day usage it lands within ~2 % of the
per-event count; outlier days mixing extreme readings can flip the
classification either way. The live fallback remains the per-event
source of truth.
The read-path warm-up's bare `catch {}` hid populator regressions —
a deadlock or pool exhaustion inside `recomputeUserRollups` would
return `{ recomputed: false }` silently and the read fall-through
to the live aggregator masked the failure until performance bled
through.
The catch now annotates `rollup_refresh_failed` plus the error
message on the wide-event pipeline (no-op when there's no request
context) and emits a `console.error` with the same shape so the
worker log path surfaces the failure regardless of caller surface.
Return shape unchanged — the read path still must not fail.
Unit test simulates a deadlock inside the inner recompute and asserts
both the annotate meta and the console fallback fire while the outer
call still returns `{ recomputed: false }`.
The generate route caught `FeaturesPayloadTooLargeError` from the
raw-mode `extractFeatures` call and retried in aggregated mode. The
retry ran the same `enforceSizeGuard` from `features.ts`, so a
very large medication history plus multi-year context could blow past
the 5 MB ceiling a second time → the uncaught throw bubbled to the
apiHandler's generic 500 path.
The retry now sits in its own try/catch. On a second oversize the
route applies the existing `applyInsightsExcludeFilter` with the
optional context blocks (anthropometrics, medications, sleep, steps,
hrv, resting_hr) excluded and re-extracts. A third failure annotates
`insights_payload_too_large` for ops visibility and returns 422
with a readable user-facing message rather than 500. Unit test mocks
all three extractor calls to throw and asserts the 422 + annotate.
The build-time row read its label from `t("builtAt", { time: "" }).trim()`
with a fallback to the gitSha key. The fallback was effectively dead
(the trim never produced an empty string) and the pattern silently
risked rendering two side-by-side "Build" columns when an
implementer dropped the trailing `{time}` placeholder. Replace with
a dedicated `settings.about.builtAtLabel` key across all six locales.
Extract the newer-version badge into an exported UpdateBadge
component. Renders as a 44 px-square anchor (or role=img span
fallback) with text-primary colour, focus-visible:ring-* on the
anchor branch, and aria-label/title carrying the latest tag so
keyboard + screen-reader users get the same signal sighted users
get from the icon. Drops the hardcoded text-dracula-purple in
favour of the semantic primary token, which already meets WCAG AA
on light + dark.
The v1.4.36 perf wave wrapped the briefing / correlation / trends
blocks in `<Suspense>` boundaries. The page is `"use client"`,
the blocks load via `next/dynamic({ ssr: false })` (no promise
throw on hydrate), and TanStack Query hooks never throw thenables, so
the boundaries never engaged — dead code, masking the real shape of
the delivery.
The actual perceptual win remains intact: the page-level
`isLoading` gate is gone, the hero plus each section's own
loader skeleton paints inside the first paint budget while data fills
in. A comment in the file explains why no `<Suspense>` wrapper
belongs here so a future contributor doesn't re-add one.
NOTE for CHANGELOG drafting: the perf-wave bullet should read
"early-skeleton paint" rather than "Suspense streaming".
Genuinely streamed server children remain a v1.5.x track.
The v1.4.35/v1.4.36 read-swap gated on a global `COUNT(*) > 0`
against `measurement_rollups`. A user with BP fully rolled up who
logged their first WEIGHT measurement saw the per-write hook upsert
one WEIGHT DAY bucket, the global probe flipped to true, the read
path picked the bucket-derived branch for WEIGHT, and the all-time
WEIGHT count collapsed to whatever the trailing 90d narrow aggregate
plus the single fresh bucket could reconstruct.
The new shared probe in `rollup-coverage.ts` joins distinct
measurement types against the rollup table per user and surfaces
`{ type, has_buckets }` rows. Both `computeSummariesSlice` and
`buildComprehensiveAggregate` take the rollup-derived branch only
when every type the user has logged is covered; partial coverage
falls back to the live aggregate so the brand-new-type case stays
correct. Integration test seeds BP rolled up + WEIGHT raw-only and
asserts the WEIGHT summary count matches live SQL.
shadcn Button size="sm" renders at 32 px — below the project's
44 px touch floor adopted in v1.4.34.4. Use the default size with
an explicit min-h-11 sm:min-h-9 override so the pagination row
stays compact on desktop but clears WCAG 2.5.5 on touch.
Sort headers relied on hover:underline only and dropped below the
44 px touch target floor on mobile. Add focus-visible:ring-* and
min-h-11 sm:min-h-9 so keyboard + touch users can hit the column
header reliably.
Pulls the v1.4.36 W4c MeasurementType -> SourcePriorityMetricKey
chain out of the dashboard per-type loop into a typed switch helper.
Same call sites, same null fallback for the unmapped TIME_IN_DAYLIGHT
branch.
v1.4.36's perf wave added a `SELECT COUNT(*) FROM measurement_rollups`
probe at the head of `computeSummariesSlice`. The slim-slice route test
mocks `$queryRaw` with two `mockResolvedValueOnce` calls (aggregate +
latest); the new probe consumed the first slot, leaving the aggregate
read pointing at undefined and the route returning 500.
Seed `n: BigInt(0)` as the first mock value so the route deterministically
takes the legacy heavy-aggregate fallback path the rest of the test
expects.
The carrier chip on the admin login-overview audit table was empty
whenever the offline GeoLite2-ASN MMDB had no entry for the IP. In
practice this is every IP behind a non-DACH or cloud-NAT range — the
free-tier MMDB carries DACH carriers densely but thins out
elsewhere — and the chip went blank even though the city + country
were already populated from the parallel location lookup.
When `entry.carrier` is null but `entry.location` is non-null,
render the chip with the city/country string and a "carrier
unavailable" qualifier. Private/loopback IPs still hide the chip
entirely (both columns null).
The dedicated "Updates" card carried a "Check for updates" button,
a last-checked timestamp, and three result paragraphs covering
up-to-date / newer-available / unknown. All of it surfaced an answer
to one question — "is there a newer release?" — and the answer was
overwhelmingly "no, you're current". The card was three full rem of
chrome for that signal.
The 24 h auto-check still runs on mount. When it returns
`newer_available`, an ArrowUpCircle badge appears next to the
version line; hover/tap exposes the latest tag via the title
attribute. The link points at the release HTML URL when present.
No bouncing, no badge counter, no manual recheck button.
The settings/about route now reads as a tight identity card +
external links list, matching how often the surface actually
gets used.
The targets endpoint runs >1 s cold — pulls 30 days of measurements
for every type plus medications, intakes, mood and glucose to build
the consistency strips, streaks and target-met-this-week rollup. It's
read on every Insights mount and on every sub-page click, so the
cumulative cost dominates a power user's session.
Wrap the body in the existing analytics-style server cache with a 60 s
TTL and a per-user key. The invalidation helpers for measurement,
mood and medication writes evict the user bucket so the next read
paints fresh data without the cache lagging the underlying state.
The bfcache-friendly Cache-Control header lets a back-forward
navigation restore from the BFCache instead of repaying the cost.
The About surface (version + license + source links + update check)
used to live as its own Settings section and the user-card dropdown
exposed it from both desktop and mobile. The three cards collectively
get opened once or twice a year, mostly by the operator confirming
the running version against the latest release.
Move the surface to `/admin/about` and drop the dropdown link from
both the desktop sidebar and the mobile top-bar. The public-facing
landing at `/about` (credits + MaxMind attribution) stays untouched
since it's reachable without a session. The legacy `/settings/about`
route still resolves so external bookmarks don't break, but is no
longer surfaced in any nav.
The existing `<AboutSection>` is reused as-is — no markup changes,
the component owns its own heading so the admin renderer mounts it
without the SectionFrame wrapper.
The Insights trends row fires three parallel
GET /api/measurements?type=…&aggregate=daily&limit=5000 requests on
mount (BP_SYS / BP_DIA / WEIGHT). The legacy date_trunc path scanned
the measurements table each time — ~3 s per call on Marc's account.
Add a `source=rollup` opt-in on the listMeasurements schema. When set
alongside `aggregate=daily` the route reads from measurement_rollups
DAY buckets instead, swapping a full-table scan for a small indexed
read against the ~5 k-row rollup table. Cumulative HK metrics get
mean * count to reconstruct the daily SUM the chart expects.
The chart-data client opts in for every windowDays > 7 fetch so the
trends row + every sub-page chart benefit. When the rollup returns
zero buckets for the requested window the route falls through to the
live aggregate so brand-new accounts still render their first chart
correctly.
Walk the features payload and the Coach snapshot through a shared
compactSections helper before JSON.stringify so the prompt never
carries labelled-empty keys ("sleep": [], "medications": {}, etc.).
Empty arrays, empty plain objects, and undefined values are dropped;
null is preserved (it signals "unknown", which the model handles
differently than "missing block").
The pre-v1.4.36 fallback rendered every metric pill while the
analytics fetch was in flight. Users with zero observations on a
metric briefly saw nav targets they couldn't act on, and on a
failed or slow fetch the pills stayed up forever — clicking
through landed on an empty-state page.
The strict gate flips the contract: no availability => only the
Overview pill renders, then metric pills light up progressively
as `summaries[METRIC].count` flips from 0 to ≥1. This matches
the auto-light-up behaviour the original metric-availability
helper was designed for.
The dashboard tile gate and per-sub-page empty-state gate already
applied the strict reading; the tab strip is now in lockstep.
Add insightsExcludeMetrics to the User schema and apply the list as
a feature-payload filter inside /api/insights/generate before
JSON.stringify. The Coach settings PUT mirrors its excludeMetrics
onto the new column so both surfaces share a single privacy
contract — no separate Insights settings panel needed.
Anthropometrics is special-cased to strip only height / age / gender
inside features.context, leaving the aggregate totals intact.
v1.4.35 left the regression intact: comprehensive-aggregator and the
slim summaries slice still ran the full COUNT/MIN/MAX/AVG `$queryRaw`
in parallel with the rollup read, only preferring the bucket-derived
values when the count happened to match. Net round-trip count saved
was zero on Marc's 311k-row account.
Promote the rollup table to the canonical source on the happy path:
gate on a cheap `SELECT COUNT(*) FROM measurement_rollups` and, when
non-zero, drop the heavy aggregate. The bucket composition feeds
count/min/max/mean, a narrow `$queryRaw` keeps the non-composable
windowed columns (stddev/anomalies/avg7/avg30/avg30_last_month + the
three slope tuples), and the legacy heavy aggregate stays as the
cold-mount fallback for accounts whose backfill is mid-flight.
Unit tests pin the contract: the heavy aggregate is NOT called on the
rollup-fresh branch. Integration test verifies byte-identical count /
min / max / mean against a parallel live SQL pass on a populated
fixture.
The Steps tile, Active-Energy tile, Walking + Running Distance tile,
Flights-Climbed tile and Time-in-Daylight tile all rendered the
*latest minute-level slice* from Apple Health as `summary.latest`
instead of the day's total. The user-facing result was a Steps tile
that read "1,247 steps today" when the actual day sum was closer to
8,000 — the tile was showing the freshest slice, not the running total.
SLEEP_DURATION already ran the per-day-source-pick + per-day-sum
collapse inline in /api/analytics. Extract that reduction into
`pickCumulativeDaySum` and broaden it to ACTIVITY_STEPS,
ACTIVE_ENERGY_BURNED, WALKING_RUNNING_DISTANCE, FLIGHTS_CLIMBED, and
TIME_IN_DAYLIGHT.
Each cumulative type maps to its own SourcePriorityMetricKey for
the canonical-source pick (TIME_IN_DAYLIGHT falls through because
there is no clinical-grade competitor today). After the collapse,
`summary.latest` is the most recent day's total, which is what the
dashboard tile expects.
Tests: 7 new unit cases on the pure helper; SLEEP_DURATION
integration regression still green.
Extend the coach exclude-metrics enum with two optional-context tokens
and surface them in the settings sheet under a separate "Include
context" group. The snapshot gates the GLP-1 weeklyContext block on
the medications toggle (skipping the Prisma lookup entirely when
excluded) and gates the new anthropometrics block on the
anthropometrics toggle plus at least one non-null profile field.
Empty blocks are never emitted.
The Settings + Admin shells centred the active chip via
`scrollIntoView({ inline: "center" })`. On narrow viewports the
strip over-scrolled past the leading chips so the first one or two
were unreachable until the user manually swiped right. `inline:
"start"` lands the active chip at the leading edge so the trail of
preceding chips remains visible.
The v1 IntakeHistoryList was retired in v1.4.28 on the assumption
it had no other consumer. The standalone /medications/[id]/history
page lost its dose timeline as a side-effect — non-GLP-1 medications
now render only the heading. This restores a slim, read-only v2.
- Paginated 25/page via the existing limit/offset query params
- Sortable by takenAt (default desc) and scheduledFor
- Status and source badges; no inline edit or delete
- Empty state with CTA back to the daily intake page
- i18n keys across all six locales (DE + EN curated, the AI-initial
locales mirror the EN copy per the v1.4.25 W9e maintainership policy)
The trends-row annotation slot painted "Mehr Daten nötig, um diesen
Trend zu kommentieren" whenever the advisor query was in flight or a
regenerate was firing — even when the model was about to deliver
fresh prose. Two cold-mount conditions collapsed into the same empty
hint: advisor pending, and advisor returned null.
`<TrendAnnotation>` now reads a tri-state `status` prop:
- `pending` — advisor query in flight / regenerate firing.
Renders a 3-line shimmer block.
- `needs_data` — advisor returned `annotation = null`. Renders the
"Mehr Daten nötig" hint.
- `generated` — advisor returned a string. Renders the prose +
optional confidence chip.
When `status` is omitted the legacy `annotation == null → empty`
mapping still applies (back-compat for existing callers).
`<TrendsRow>` derives the status from a new `loading` prop and the
per-metric annotation presence, with pending winning over needs_data
so a mid-regenerate cycle never flashes the empty hint between
shimmer and new prose. The Insights mother page now passes
`loading={advisor.isLoading || advisor.isRegenerating}`.
Added unit tests pin the three render-state branches plus the
back-compat fallback. New `trendAnnotation.pendingLabel` i18n key
across DE / EN / FR / ES / IT / PL drives the aria-label on the
pending block. Health Score paint lag is a parent-level analytics
query issue and stays for v1.4.37.
Replace the per-row rawMeasurements dump with DAY/WEEK/MONTH bucket
means read from measurement_rollups. The legacy shape blew past
Codex's 10 MB string limit on power-user accounts; the bucketed shape
stays O(types × buckets) regardless of row count.
Enforce a 5 MB ceiling on the serialised payload; the generate route
downgrades to the aggregated shape and annotates when the cap trips so
future regressions surface in logs.
The trends-row chart slot is now pinned to 140 px, but the default
`<ChartSkeleton>` paints a `rounded-xl` card with `p-4 md:p-6` and a
240 px chart band — overflowing the slot and reintroducing the
layout shift the height pin was meant to eliminate.
Add a `mini` prop to `<ChartSkeleton>` that renders the lighter
`rounded-md p-2` chrome plus a 140 px band. Expose
`HealthChartDynamicMini` from the chart-dynamic boundary so the
trends-row consumers pick up the matching loading shell without
having to wire the skeleton choice at every call site.
Trends-row swaps both `<HealthChartDynamic>` calls (BP + weight) for
the mini variant and the local `<MoodChart>` `next/dynamic` loader
now passes `mini` to its skeleton. The dashboard hero + sub-page
charts keep the default card chrome (`mini={false}`).
Each `trends-row-chart-slot` wrapper had `shrink-0` but no explicit
height. Recharts' ResponsiveContainer mounted with `width=-1
height=-1` and emitted a warning on every Insights cold mount; the
mood card visibly drifted taller than BP / weight as the chart
hydrated, contributing the bulk of the Lighthouse 0.46 CLS hit.
Pin the wrapper to 140 px (matches the chart-internal mini default).
Recharts now mounts at a known size, the three cards line up on the
same baseline from first paint, and the warning is gone. Test pins
the height on all three slots so a future refactor has to revisit
the row rhythm intentionally.
Removes the operator step that was implicit in the v1.4.35 foundation.
Self-hosters no longer need to run the backfill script after upgrade.
New `rollup-full-backfill` queue + handler folds one user per job at
serial concurrency. `enqueueBootTimeRollupBackfill` runs once at
`startReminderWorker` boot, discovers accounts with measurements but
zero rollup coverage, and enqueues one full-fold job per uncovered
user with a singleton key so fast restarts don't pile up duplicates.
Best-effort: discovery errors surface through the helper's return
value and are logged via workerLog; the worker boot never fails
because the backfill missed.
Tests: 4280 → 4285 unit + 228 → 230 integration. tsc + lint clean.
Switches two reader surfaces onto the persistent rollup table for the
linearly-composable stats. The comprehensive insights aggregator and
the slim analytics summaries slice now source count / min / max / mean
from the DAY rollup buckets via aggregateBuckets, plus the comprehensive
dailyByType correlation feed straight from the bucket means. Slope /
R² / standard deviation / anomaly counts stay on live SQL — they do
not compose across DAY buckets.
A defensive parity check against the live aggregate's COUNT(*) falls
back to live SQL when divergence is detected: covers the cold-mount
edge case where the rollup populator is mid-flight, and degrades
cleanly on the slim slice's all-time window for accounts that never
ran the explicit backfill.
Three new integration cases pin the parity contract end-to-end
against a real Postgres testcontainer: the aggregator's
count / min / max / mean matches live SQL byte-for-byte, dailyByType
matches a parallel date_trunc('day', ...) GROUP BY query, and a
freshly-written measurement is reflected on the next read.
tests: 4249 → 4280 unit + 222 → 228 integration. tsc + lint clean.
WIP toward v1.4.35. Lands the persistent rollup foundation from the
v1.4.34 cache-aggregation research §9. Reads still served by the
v1.4.34.1 live-SQL aggregator — the parity-safe read-swap
(count/min/max/mean + dailyByType from buckets, slope/r2/sd stay on
live aggregation) ships in a follow-up commit so v1.4.35 is the
single release that flips both foundation + read-swap.
Schema:
- new MeasurementRollup model with composite PK
(userId, type, granularity, bucketStart) and a descending
bucketStart index. RollupGranularity enum (DAY/WEEK/MONTH/YEAR).
- Prisma migration 0067_v1434_measurement_rollups is additive:
CREATE TYPE / CREATE TABLE IF NOT EXISTS / CREATE INDEX IF NOT
EXISTS / ADD CONSTRAINT FK. No ALTER on existing tables.
- db-compat schema-bootstrap path mirrored.
Populator (src/lib/measurements/rollups.ts):
- recomputeBucketsForMeasurement: sync DAY; WEEK/MONTH/YEAR via
pg-boss (singleton-keyed).
- recomputeUserRollups: bounded full-range, used by Apple Health
import + backfill.
- ensureUserRollupsFresh: cheap stale-detection + bounded DAY-90d
recompute. Error-swallowing.
- SQL via $queryRawUnsafe + strict enum whitelist for date_trunc
unit (Postgres 42803 forbids binding the unit literal).
Write hooks on every measurement-mutating endpoint; each wrapped
in try/catch so a populator hiccup never fails the write.
Apple Health import worker recomputes the full range once at
completion (per-row hooks would multiply cost).
Backfill (scripts/backfill-rollups.ts): idempotent, single-user
serial, single-digit minutes for a 100k-row account.
Tests: +14 unit, +3 integration. setup.ts.truncateAllTables
extends to the new table.
Read-swap follow-up:
- comprehensive-aggregator + analytics/route + summaries-slice
each call ensureUserRollupsFresh as a read side-effect (warms
the table). The response stays on the live-SQL aggregator so
byte-shape parity with v1.4.34.5 is preserved.
- Next commit lands the partial read-swap: count/min/max/mean
and dailyByType from buckets (math-identical compose), while
slope/r2/sd stay on indexed live aggregation. Together they
ship as v1.4.35.
Pre-shipping state — do NOT tag v1.4.35 from this commit alone.
Five textareas in the app (Coach composer, bug-report description,
admin feedback note, medication-side-effects notes, medication-intake
JSON import) shipped `text-sm` (14 px) on every viewport. iOS Safari
triggers a viewport zoom-on-focus whenever an input renders text
below 16 px — so every tap on these surfaces yanked the page scale
and rebounded on blur. Switched each className to `text-base sm:text-sm`
so the mobile baseline clears 16 px while desktop keeps the compact
look the rest of the form chrome converged on.
No shared <Textarea> primitive exists yet — F-5's deferred work — so
the fix is per-call-site. Five files, one className change each, no
component re-shape.
The sticky header was h-16 with no safe-area handling. On notched
iPhones (~44 px status-bar inset) the iOS PWA painted the status bar
over the header — the HealthLog logo + auth controls clipped under
the system clock. Adding the inset as inline padding-top reserves the
overlay without touching the 64 px content baseline.
Inline `style` over a Tailwind class because Tailwind doesn't expose
`pt-[env(safe-area-inset-top)]` without arbitrary-value escaping and
the inline form is clearer in the diff.
The v1.4.34.3 code-quality batch added a `toJson<T>()` helper in
`src/lib/db.ts` to consolidate the seven Prisma JSON-cast sites.
Two existing route tests mock the entire `@/lib/db` module via
`vi.mock` — those mock factories now need to expose the new export
so the route's import resolves to a real function under test.
The mock implementation is the identity cast (`<T>(v: T) => v`),
which matches what production does and keeps the test's JSON-payload
assertions stable.
Three hotfix-ready findings from the v1.4.34.3 mobile-deep audit
plus the first two docs A8 audit identified as missing.
Mobile-deep:
- `viewport.themeColor` was pinned to `#282a36` (dark). The
Android URL bar + iOS PWA status bar stayed dark even when the
user was on the light theme. Replaced with the dual-media-query
form: `#ffffff` under `(prefers-color-scheme: light)`,
`#282a36` under `(prefers-color-scheme: dark)`. Browser chrome
now matches the active palette.
- Coach FAB was pinned at `bottom-20` (5 rem). On notched iPhones
the home-indicator safe area is ~34 px so the FAB collided with
the bottom-nav. Moved to
`bottom-[calc(env(safe-area-inset-bottom,0px)+5rem)]` so it
floats above the inset-aware nav.
- Sonner toaster shipped `theme="dark"` hardcoded. On a light-mode
page the toast contrasted badly against the surface. Switched to
`theme="system"` so Sonner reads the resolved theme from the
root element (set pre-paint by the inline theme script in
layout.tsx) instead of pinning dark.
Self-hosting docs:
- `docs/self-hosting/getting-started.md` — step-by-step from
`git clone` to first measurement, covering the 3-secret env
setup (post-SESSION_SECRET drop), `docker compose up -d`, the
first-user-becomes-admin contract, and the next-step pointers.
- `docs/self-hosting/reverse-proxy.md` — Caddy, Traefik, Nginx
Proxy Manager, Coolify, and bare-Nginx config examples. Closes
the README link at line 107 that previously punted to the docs
site.
Five hotfix-ready findings from the v1.4.34.3 code-quality audit.
SESSION_SECRET drop (F-3):
The env var was required by `docker-entrypoint.sh`, listed in
`.env.example`, `docker-compose.yml`, `CONTRIBUTING.md`, and the
scaling guide — but `grep -rn SESSION_SECRET src/` returned
nothing. Sessions are server-stored in `prisma.session` keyed by
the cookie value; the secret signed nothing. Every self-hoster
carried a ceremonial 64-char hex string that the app ignored.
Removed everywhere except the README (separate batch). Each fresh
install now generates 3 secrets, not 4.
toJson helper (F-9):
Seven sites cast Prisma JSON writes via `value as unknown as
Prisma.InputJsonValue`. Each call re-invented the same escape
hatch. Replaced with one helper in `src/lib/db.ts`:
`export const toJson = <T>(v: T) => v as unknown as Prisma.InputJsonValue;`
Migrated all 7 call sites: `auth/me/doctor-report-prefs`,
`dashboard/chart-overlay-prefs`, `dashboard/widgets`,
`integrations/healthkit`, and 4 sites in
`apple-health-import-worker`. The helper has a one-line WHY
comment + a unit test pinning the cast shape stays stable.
db-compat docstring (F-10):
`src/lib/db-compat.ts` looks like accumulated backwards-compat
cruft on first read. It's actually the schema-bootstrap path that
runs ALTER-TABLE-IF-NOT-EXISTS so a fresh container syncs without
`prisma migrate deploy`. Added a header docstring naming the role
and pointing at `prisma/migrations/` as the canonical history.
Passkey types (F-11):
`src/lib/auth/passkey.ts:103,169` used `response: any` with
per-line `eslint-disable`. Imported `RegistrationResponseJSON` and
`AuthenticationResponseJSON` from `@simplewebauthn/types` and
dropped the disables.
Zero-TODO CI gate (F-15):
The audit confirmed product code has zero TODO / FIXME / XXX /
HACK markers — genuinely rare and worth preserving. Added a tiny
GitHub Actions workflow at `.github/workflows/no-todo-markers.yml`
that fails the build on any future re-introduction. Test files
are excluded; product code only.
Eight hotfix-ready UX/a11y findings from the v1.4.34.3 UX audit
land as one batch.
i18n leaks (3 primitives):
- Dialog + Sheet close-X carried hardcoded `<span sr-only>Close</span>`
so a German user heard 'Close, button' on every modal. Both lift
`useTranslations()` and read `common.close`.
- ChartSkeleton's sr-only loading announcement was English-only.
The component is already a client island; reads `charts.loadingLabel`
from each locale now. New key shipped in DE / EN / FR / ES / IT / PL.
- Notifications-section breadcrumb `<nav aria-label="Breadcrumb">`
was hardcoded. Routes through `t("nav.breadcrumb")` with a new
key in each locale.
Focus rings (mobile + tablet nav):
- Bottom-nav primary links + the More overflow trigger relied on a
`text-muted-foreground` → `text-primary` colour transition only;
keyboard / switch-control users saw no focus indicator at all.
Added `focus-visible:ring-ring/50 focus-visible:ring-2
focus-visible:outline-none focus-visible:ring-offset-2` matching
the v1.4.33 sidebar treatment.
- Mobile top-bar trigger had `focus:outline-none` without a
`focus-visible` replacement. Same treatment applied.
Tap-target floor (WCAG 2.5.5):
- Input / NativeSelect / Select / DateTimeInput were locked to
h-10 (40 px) on every viewport. Switched to `h-11 sm:h-10` so
mobile clears the 44 px floor while desktop keeps the 40 px norm
that the Button primitive already converged on in v1.4.33.
Admin chip-strip auto-scroll-into-view:
- The admin shell shipped 13 chips but lacked the
`scrollIntoView({inline: 'center'})` effect the settings shell
got in v1.4.33. Admins tapping a chip past the right edge landed
on the new page with the strip scrolled to the left and the
active chip off-screen. Copied the settings-shell pattern verbatim.
List-page subtitles on mobile:
- Measurements / Mood / Medications pages hid their descriptive
subtitle on `<sm` with `hidden sm:block`. Comparable apps keep
the subtitle on mobile because it doubles as contextual help.
Dropped the hide rule; the H1 still anchors the visual.
Insights empty-state CTA vocabulary:
- `ALLOWED_ADD_TYPES` on `/measurements` accepted tokens (`GLUCOSE`,
`TEMPERATURE`, `HEART_RATE`, `BMI`) that didn't match the
`MEASUREMENT_TYPES` enum the form uses. Several `/insights/*`
empty-state CTAs emitted those mismatched tokens and the dialog
silently failed to open. Replaced with a single source of truth
derived from `MEASUREMENT_TYPES` plus a normalisation map for
legacy tokens. Added BODY_TEMPERATURE to the form's enum row.
New unit test walks every `?add=` token emitted under
`src/app/insights/**` and asserts it resolves to a real form type.
HSTS gains `; preload` so the domain becomes eligible for the
Chromium preload list. Withings CSP host moves from global to a
path-gated entry (mirrors the existing AI-host gating) — closes a
DOM-XSS exfil shape on every non-Withings page. `getClientIp` warns
once per process when `TRUST_PROXY_HOPS` and the XFF chain length
disagree, and exposes a tagged `getClientIpOrTrustWarning` helper so
future callers can route to a tighter universal rate-limit bucket
when the chain looks tampered with.
Service worker now resolves push-notification payload URLs against
`self.location.origin` and rejects off-origin redirects (falls back
to `/`). The German-only offline fallback HTML body was replaced
with a language-neutral icon + 'Offline' wordmark so EN / FR / ES /
IT / PL users don't see German on a cold offline boot. Manifest
gained `"scope": "/"`, `"id": "/?source=pwa"`, and
`"display_override": ["standalone"]` so future sub-path deploys
don't accidentally inherit root scope and install-prompt heuristics
anchor on a stable id.
+16 new test cases across two new files covering HSTS preload, CSP
gating on/off, the once-per-process warning, the tagged helper's 4
shape branches, SW same-origin guard, offline-locale-neutrality, and
manifest hygiene. Existing proxy/SW suites stay green.
README:
- Add a "How it compares" matrix against Withings web, Apple Health,
Oura, Garmin, and generic CSV. The audit found this section absent;
self-hosters arrive comparing alternatives and bounce when there's
no head-to-head.
- Promote Apple Health import + AI Coach as standalone Key-Features
bullets. Apple Health was the v1.4.34 banner feature with zero
grep matches in the README until now; the audit ranked it the
single highest-intent search keyword for ex-iOS users. AI Coach
+ Insights moved out from under "Multi-Provider AI Insights" into
a dedicated bullet that names the Coach, Daily Briefing, Weekly
Report, and Health Score surfaces.
- Status block under "What it is" so visitors see release cadence
+ current focus (native iOS client) without trawling the changelog.
- Badge row: drop "PRs welcome" + duplicate Self-Hosted, add a live
"Latest release" badge + a GHCR multi-arch badge so the landing
page surfaces version + container distribution at a glance.
- Tagline tightened to lead with "self-hosted health tracker" + the
two integration anchors (Withings + Apple Health) — both phrases
are search-keyword fronts the GitHub search index picks up.
- Drop the dead SESSION_SECRET env entry from Quick Start (3
secrets, not 4) and the Environment Variables table — the env
var was documented as required but never read by code (per the
code-quality audit). Companion env-removal in docker-compose /
.env.example / entrypoint ships in a parallel batch.
- Add a one-line OpenAPI pointer above the API-Reference table so
integrators discover the codegen source.
Layout / OG metadata:
- openGraph + twitter cards now carry images, a higher-intent
description, and a summary_large_image card. Uses the existing
logo-readme.png as the OG asset; replace with a 1200x630 dashboard
capture when one ships.
GitHub repo metadata (via gh CLI, no git diff):
- description tightened from 290 chars to 156 chars, keyword-front.
- homepage flipped from a personal tenant to demo.healthlog.dev so
sidebar clicks land on a working demo instead of a login wall.
- Topics: dropped 6 generic/PII-adjacent entries (glp-1, mounjaro,
tracking, health, dashboard, bloodpressure), added 6 high-intent
ones (apple-health-import, withings-alternative, glucose-tracker,
mood-tracker, ai-insights, personal-dashboard).
Planning artifacts:
- 7 audit reports drop into .planning/round-audit-*.md covering
mobile-security, UX, code-quality, README/discoverability,
perf, mobile-deep (pending), test-coverage, docs-depth. These
are the working notes for the v1.4.34.4 audit-driven release.
Two findings from the v1.4.34.3 security audit, both with the same
shape — every authenticated surface a user owns should rotate when a
credential-rotation path fires, and unfamiliar discovery sub-paths
must not inherit "no auth" status by default.
destroyAllSessions:
Previously only ran `prisma.session.deleteMany`, leaving every
long-lived `ApiToken` row (Bearer credentials issued via
/settings/api-tokens) and every native-client `RefreshToken`
intact. A stolen iOS Bearer kept working after the user rotated
their password or after an admin reset; only the browser cookie
flipped. Wraps the cleanup in a $transaction that flips
`apiToken.revoked = true` and `refreshToken.revokedAt = now()` in
the same atomic block so the audit trail survives. Both lookups at
`src/lib/auth/api-token.ts` and `src/lib/auth/refresh-token.ts`
already filter on these columns; no second code path needed.
Same destroyAllSessions call already runs from:
- POST /api/auth/password (user-initiated change)
- POST /api/admin/users/[id]/reset-password (admin reset)
- DELETE /api/settings/account (account deletion)
All three paths now wipe every transport in one shot.
Well-known proxy bypass:
`src/proxy.ts` PUBLIC_PATHS carried "/.well-known/" with a prefix
match — a future route at `/.well-known/foo` would inherit "no
auth" status silently. Tighten to an exact-match Set
(WELL_KNOWN_PUBLIC_PATHS) seeded with the one path we actually
serve (`apple-app-site-association`). New routes must be added
explicitly.
Also drops the stale `/api/auth/codex/callback` PUBLIC_PATHS
entry — that route doesn't exist in the codebase (the Codex
OAuth flow uses `/api/auth/codex/{device-poll,device-start,
disconnect}` instead).
Tests:
- New destroyAllSessions assertion in src/lib/auth/__tests__/
session.test.ts pins the three updateMany/deleteMany call
shapes inside a single $transaction.
- Existing proxy well-known + privacy + onboarding redirect
suites stay green.
Per maintainer directive — the v1.4.34 IW-B "Frag den Coach" sparkles
button sat next to the "Hinzufügen" dropdown on the dashboard hero and
read as visual noise rather than a useful surface. Remove it; the
Coach drawer remains reachable from the /insights tree (its original
home), and the auth-shell `<CoachLaunchProvider>` + drawer mount stay
intact so any future page that wants its own opt-in CTA can call
`askCoach()` without churn.
The hook usages this leaves orphaned in `src/app/page.tsx`:
- `useCoachLaunch` (only consumer was the CTA)
- `useFeatureFlags` (only gate was `featureFlags.coach`)
- `Sparkles` icon import
All drop out cleanly.
The v1.4.34 server-cache primitive (IW-G) provisioned eight per-route
LRU instances and wired three. Land the remaining five reads plus the
chart-overlay-prefs invalidation hook the new dashboard-widgets wrap
exposes.
Reads wrapped (60 s TTL unless noted):
- GET /api/mood/analytics → caches.moodAnalytics, key userId
- GET /api/workouts → caches.workouts, key includes
all params (limit/offset/since/
until/sportType)
- GET /api/bugreport/status → caches.bugreportStatus, key
"singleton" (10 min TTL);
isAdmin layered per request
- GET /api/medications → caches.medications, key userId
- GET /api/dashboard/widgets → caches.dashboardWidgets, key
userId (5 min TTL)
Write paths already call the matching invalidateUser* helper from the
v1.4.34 invalidation matrix; the new dashboard-widgets cache also gets
busted from /api/dashboard/chart-overlay-prefs PUT, which writes the
same User.dashboardWidgetsJson blob through a partial-update path that
previously bypassed the matrix.
The canonical-dedup test suite for /api/workouts now imports
__resetAllCachesForTests in beforeEach so a fixed userId fixture
doesn't carry cached state across cases.
The Cloudflare "no available server" 503 cascade on
/api/measurements?aggregate=daily resolves as a downstream consequence —
once /api/insights/comprehensive stops monopolising the pool, the origin
accepts new connections fast enough that Cloudflare's circuit-breaker
no longer trips.
The /api/insights/comprehensive handler used to walk every measurement
of every type from the trailing 90-day window into JS, then ran
summarize(), BMI, BP classification, target adherence, and four Pearson
correlations against the same pile of rows. Post Apple Health import the
unbounded findMany routinely returned 100 000+ rows, dominated request
time (~29 s in a production HAR capture), and starved the connection
pool — sibling endpoints cascaded into Cloudflare 503s.
Replace with a SQL-side aggregator in
src/lib/insights/comprehensive-aggregator.ts:
- Per-type DataSummary built via $queryRaw with AVG / MIN / MAX /
STDDEV_POP / REGR_SLOPE / REGR_R2 on the
measurements (user_id, type, measured_at) index. Anomaly count uses
STDDEV_POP to match the JS divisor (n, not n-1) and the same
round-then-threshold ordering as summarize().
- DISTINCT ON (type) for the latest raw value per type.
- date_trunc('day') daily means for correlation pairings.
- Two bounded findMany for BP sys/dia raw rows so the 5-minute
pairByTimestamp contract for bpPctInTarget survives byte-identical.
- One conditional MIN(measured_at) for dataSpanDays.
Route handler split into a buildComprehensiveResponse() helper wrapped
through caches.analytics (60 s TTL) keyed on `${userId}|comprehensive`.
The 60 s TTL converts every consumer within the window — page mount,
Coach drawer, recommendations grid — to a Map lookup; cold path drops
to a few hundred ms against fixture data matching the production volume
that triggered the bug.
One sanctioned semantic divergence: weight × BP correlation switches
from pairByTimestamp(24 h tolerance) to a daily-key join. The directive
authorised the swap for parity with the daily-mean shape mood × X
correlations already used; consumers comparing exact r/n against legacy
output may observe small drift.
Unit + integration tests cover the aggregator shape, cache hit / miss
counters, wide-event annotations, and end-to-end envelope parity.
The scatter-correlation card wrapper relied solely on Tailwind's
aspect-ratio utilities to derive its height from width, and explicitly
unset any inline height. When the dynamic-loaded chart mounted in
place of the h-[180px] skeleton, ResponsiveContainer briefly measured
a zero-sized parent before the browser resolved aspect-ratio against
the parent's width, emitting a width(-1)/height(-1) console warning.
Add a min-h-[180px] floor (matching the loading skeleton) so the
wrapper always has a non-zero height at first paint. The aspect-ratio
rules continue to drive the final card shape once layout settles.
Three integration tests against the real Postgres container pin the
v1.4.34 IW-G contract:
1. Ten concurrent /api/analytics reads against a cold cache fan into
a single builder call (misses === 1; remaining 9 split between
stampede joiners and warm-cache hits, with no second miss).
2. Priming analytics then writing a measurement evicts the cache —
the next read misses again and the response reflects the new row.
3. Compliance buckets re-read after an intake POST flush the cache
and surface the new taken-count.
Drops the unused _request parameter from buildAnalyticsResponse() so
the analytics route lints clean. CHANGELOG carries one Added line per
the marathon's user-facing voice rules.
Marc-André Bombeck