Revert the nine-group consolidation and the card relocations back to the
flat sidebar: sessions, trusted devices, and login activity return to
Privacy, and account deletion returns inline under Advanced. Drop the
/settings/layout redirect and the now-unused group and sign-in cross-link
i18n keys across all six locales.
Keep the collapse-by-default sessions and activity cards, the medical-history
cross-link from "About me", and the chronic-conditions consolidation.
Close the gaps the final verification pass flagged on the documents surface:
- Confirm no longer claims success when the route rejects a fact into its
`failed[]` array (HTTP 200). A unit-mismatch fact left silently PENDING with
a green toast now raises a warning naming how many couldn't be saved and why.
- Unify the German documents copy on "Angaben": the discard-confirm body, the
provider-unsupported note, and the module description still said "Fakten".
- After upload, sort by date-added rather than document date so a back-dated
upload always lands on the first page where its detail panel can open.
- Give the list its own load-error state with a retry instead of rendering the
"no documents yet" empty state on a failed fetch.
- Classify the local-OCR-disabled extract error to a clear message.
- Drop the orphaned settings.sections.notifications.inboxLink key (no call site).
The insights alert cards computed medication compliance with the naive
`schedules.length × days` denominator, so a fully-adherent weekly
injectable read ~14% and fired a false low-compliance alert (and a paused
medication mis-fired the same way). Route the alert path through the
canonical `calculateCompliance` engine like every other compliance
surface — pause eras, schedule revisions and dose windows honoured.
Document confirm: committing a numeric observation against an existing
biomarker silently replaced the stated unit with the catalog's, with no
conversion — a stated "6.1 mmol/L" against a mg/dL marker committed
"6.1 mg/dL". Reject on a unit mismatch (fail closed) so the user
reconciles on the review screen, the same posture as the unit-required
guard.
Wire grip strength, waist (with waist-to-height) and pain into the
briefing feature set and the grounding allow-set so the clinical-depth
signals reach the daily narrative; the mental-health screeners stay
excluded by construction.
Biomarker detail: fall back to a generic description for catalog-less
markers, and keep the heading + description with a retry on the error
state. Collapse the medication External-API and Lifecycle groups by
default. Re-enable the insights scroll-to-top regression and assert the
shipped main-content reset.
Align the EN haemoglobin/haematocrit display names to British spelling.
Drop the redundant `(userId, deletedAt)` index on inbound documents — a
strict prefix of the new three-column index. Trim the dead
MedicalDisclaimer variants to the one live `dataPosture`. Move the
crisis-resources card off the retired amber tint to a calm
destructive-subtle border. Add a registry invariant asserting every
biomarker signal key resolves to a real catalog slug.
Follow-ups on the v1.25.1 settings group consolidation.
- Sign-in security is no longer split across two top-level groups. Active
sessions, trusted devices, and the login-activity feed move into Account
> Security alongside the second-factor and passkey controls, so every
"who can sign in as me" control sits in one place. The Data & Privacy
pane keeps the data / retention / encryption story and cross-links to
the security home. Sessions and activity are no longer duplicated on the
Profile tab.
- First-child sub-tabs no longer repeat their group name: the section
titles become "Profile & preferences", "Connections", and
"Privacy & retention" across all six locales.
- Account deletion moves out of Advanced into the Account group as a
standalone card; research mode and the data reset stay in Advanced.
- Disambiguate the Spanish labels for the health record vs medical
history (they both read "Historial médico"); medical history becomes
"Antecedentes médicos".
- The section heading carries its id on a single breakpoint instance to
clear a duplicate-id-aria violation; the focus hook keeps targeting
both instances and picks the visible one.
- The in-page sub-tab strip scrolls its active tab into view on a narrow
viewport, matching the group strip.
- Redirect /settings/layout to the Display landing now that nothing links
to the old personalisation hub.
- Note in code that delivery channels deliberately stay under
Integrations (a channel is a delivery provider).
- Reword the upload toasts to the store-only reality ("stored", not "read"),
so the success message no longer claims AI ran and the panel it opens no
longer contradicts it.
- Classify upload failures (too large / wrong type / rate-limited / invalid
metadata) into specific messages; add a client-side 12 MB pre-check and show
the accepted formats and size ceiling beside the file field.
- Clear filters, search, and sort on a successful upload so the new document is
in the list the detail panel opens against, and add a "Clear filters" action
on the no-results state.
- List approved facts read-only after review so the link between a document and
the records it contributed stays visible instead of "Nothing left to review".
- Format date-group headers through the app's own date formatter, so they honour
the user's date-format preference instead of a textual medium style.
- Stack the toolbar until md and let it wrap; wrap the fact badge row on narrow
screens.
- Tie the disabled Approve to a plain-language reason, label the confidence
percentage, and render it through the locale percent formatter.
- Treat a fully-confirmed re-extract (422) as the calm already-reviewed note,
matching the partly-confirmed case.
- German: unify on "Angaben" for the extracted items and drop the anglicism.
Replace the terse one-line catalog descriptions with richer three-to-four
sentence explainers for all 30 biomarkers shown on the lab detail page:
what each marker measures, its physiological role, and what it relates to
clinically. The text stays static, factual, and non-diagnostic, with no
reference ranges baked into the prose. Updated across all six locales
(de/en/es/fr/it/pl) at exact key parity.
Default documentDate at store time. A store-only upload no longer lands
with a NULL filing date: when the client omits one, default it to the
upload day so the column is always set on a stored row. The library list
sorts, groups, and range-filters on the same effective date the UI shows,
so undated uploads no longer mis-sort under nulls-last, break date-group
headers, or fall through the range filter. The date stays user-editable
and may be cleared back to NULL via PATCH.
Block re-extraction once any fact is APPROVED. A partially-confirmed
document stays EXTRACTED, which the CONFIRMED guard does not catch;
re-extracting would delete the approved facts (severing the committed
record's provenance) and re-stage them as PENDING, letting a second
approval duplicate the committed lab / condition / medication. The
extract route now refuses with 409 (documents.inbound.alreadyPartlyConfirmed)
when approved facts exist; the user must finish reviewing or discard first.
RFC 5987-encode the original-download filename. The Content-Disposition
now carries both an ASCII filename= fallback and a filename*=UTF-8'' value
so a non-ASCII name survives intact; the stored set is inline-safe by
construction, so the dead attachment branch is removed.
Add a light per-user rate-limit backstop to the original-read, PATCH, and
DELETE routes (generous; self-DoS only — the routes are owner-scoped).
Lean the list counts: tally facts for the whole page in one grouped count
instead of materialising fact-id rows, and exclude REJECTED facts from
factCount. Extend the documentDate index with deleted_at to cover the
list scan. Document the deliberate plaintext-title PHI tradeoff at the
column and the store site.
Tighten the documents library review surface after a QA pass:
- Surface the provider-unsupported and already-confirmed notes as polite
live regions (role=status / aria-live) so keyboard and screen-reader
users hear them when the suppressed-toast paths fire.
- Render a document's detail panel directly after the row that opened it,
and tie the row's aria-controls to the panel id, fixing reading and
tab order.
- Gate Discard behind a confirm dialog and toast on failure.
- Disable Upload until a file is chosen, with an inline hint.
- Namespace per-fact editor field ids so simultaneously-open editors no
longer share DOM ids (label-for associations stay correct).
- Hide Extract once any fact is approved, mirroring the server guard, and
treat the already-confirmed response as a calm inline note.
- Add a document-date range filter, a kind fallback for the detail title,
and a screen-reader hint on the new-tab original link.
- Calmer, neutral needs-review badge and more legible row badges.
Adds the new strings across all six locales.
Login only wiped the in-memory query cache; the IndexedDB query snapshot
and the service-worker data/page caches from the previous account stayed
on disk. Logout and the 401/expiry redirect already drop all three. Run
the same offline-cache wipe on a successful login too, before the new
user's dashboard prefetch warms, so a fresh sign-in on a shared browser
inherits nothing from the prior session. The per-account restore guard
still stands; this removes the stale data at the source.
Settings shell: move focus to the section heading when the active section
changes within the mounted shell, so keyboard and screen-reader users are
placed in the new content instead of being left on the nav link they
activated. The first mount is skipped to avoid an unexpected focus move on
a deep link. Raise the inactive sub-tab text from muted-foreground to
foreground/70 so it clears the WCAG AA contrast floor over the tinted
sub-tab strip.
Evolve the inbound-documents screen into a real document library on top of
the store-first backend.
- Upload stays store-only and always works: the form now files a document
with an optional title, category, and document date in one step. Extraction
is a separate, explicit action.
- List becomes an organized library: documents grouped by date, with a title
(filename fallback), category and status badges, and pending/fact counts. A
toolbar adds title/filename search, a category filter, and a sort control
(document date / date added / title, asc/desc). Keyset pagination loads more
via the returned cursor through useInfiniteQuery.
- Detail panel: view or download the original, edit metadata (rename,
recategorise, set the document date), optionally run extraction, then the
existing review-then-confirm flow. When no scan provider is configured,
extraction surfaces a calm inline note instead of a hard error; the stored
document is untouched. Discard stays.
- Add a parameterised list key to the query-key factory so each filter/sort
combination keeps its own cache bucket while staying under the documents
prefix for invalidation.
- Rename the surface to "Documents" across all six locales (title, subtitle,
module label/description) and add the new toolbar, edit, extract, and
date-group strings. Route, model, and module key stay internal.
Add component and unit coverage for the grouped list, the detail/review
surface, the metadata edit form, the provider-unsupported note, and the
search-param/grouping/error-classification helpers.
Invert the inbound-documents surface from an OCR inbox into a document
library: a file can be stored without any AI provider, and extraction
becomes an opt-in action on top.
- Migration 0222: add `title` + `document_date` columns, a
`(user_id, document_date)` index, the richer kind taxonomy
(lab/imaging/prescription/insurance/vaccination), and a STORED status
made the column default. The status enum is recreated so the new value
can be set as the default within the migration transaction; kind values
use ADD VALUE. Nullable columns, no backfill.
- POST /api/documents/inbound is now store-only: bounded multipart read,
magic-byte MIME sniff, AES-256-GCM at rest, persisted as STORED with no
provider/consent/budget call and no egress.
- POST /api/documents/inbound/[id]/extract carries the relocated
extraction gauntlet (resolve provider, consent, rate-limit, budget, run,
reconcile, stage facts) against an already-stored row; absent a provider
it 422s the enhancement only and leaves the stored document intact.
- PATCH /api/documents/inbound/[id] edits title/kind/documentDate
field-by-field, owner-scoped by session userId.
- GET /api/documents/inbound gains search, kind filter, documentDate
range, sort, and keyset pagination via the parameter-bound query builder.
- Extend the validation schemas + DTO, regenerate the OpenAPI contract,
and add route tests for the store/extract/edit/list paths.
The settings left rail rendered one entry per section — 26 visible — which
read as a long flat list on desktop and a 26-chip scroll strip on mobile.
Group the sections into nine top-level entries and surface each group's
children as in-page sub-tabs, keeping the rail one level deep and the second
level inside the page.
Every section route stays live at its own URL. The rail is now driven by a
SETTINGS_GROUPS model; a reverse index maps each child slug onto its group so
deep links and the per-page settings links keep resolving and highlight the
right group, with the active child marked on the sub-tab strip. This
generalises the existing hidden-child highlight mechanism rather than moving
routes, so bookmarks and the contract surface are untouched.
Groups: Account (profile, sign-in), Tracking (areas, the per-domain pages,
targets), Display (dashboard, insights), Integrations (connections, channels,
source priority), Notifications, AI & Coach, API & Access, Data & Privacy
(export, health record, sharing, privacy, advanced), About. Per-domain
settings keep their view and data together inside one sub-tab — only the nav
groups, never a single concept.
Rename the Display group title away from "Dashboard" so it no longer collides
with its own Dashboard child. Add the nine group labels and the sub-section
nav label across all six locales. Drop the now-unused hub back-link.
The layered self-description standard already carries the canonical
intended-purpose line, the Coach standing note, the emergency note and
the data-posture note. The medication, insights and labs surfaces each
repeated a further context-specific disclaimer footer that added noise
without adding meaning.
Remove the three rendered footers and retire the now-unused
`insights` / `labs` / `medications` disclaimer variants from the shared
component (type, key map and doc). Drop the orphaned selfDescription
keys from all six locale bundles so call-site and reverse i18n coverage
stay in parity.
The root QueryClient is created once and survives every client-side
navigation, so logout then login on the same browser left the previous
account's non-user-scoped health-data queries in memory and the next
account read them before any refetch landed. Wipe the entire in-memory
cache on logout, on session expiry, and at the start of a fresh login,
alongside the existing IndexedDB and service-worker cache wipe, so no
session ever inherits another account's data.
Remove the standalone amber disclaimer card from the inbound-documents
review screen. The non-diagnostic essence now reads as the plain muted
subtitle line under the page title: upload a clinical document and review
the extracted details. Drop the now-unused disclaimer.title /
disclaimer.body keys across all six locales and the AlertTriangle /
CardDescription imports left orphaned by the removal.
Move the PHQ-9 / GAD-7 screener off `/insights/mental-wellbeing` to a
top-level `/mental-wellbeing` route so it no longer borrows the Insights
tab strip and layout shell. The page now uses the standard module-page
frame at the wider content width the other module pages use.
Replace the standing screening disclaimer block and the opt-in/private
line in the header with an info icon after the page heading whose
hover/focus tooltip carries the screening-not-a-diagnosis note. Drop the
now-orphaned `disclaimer` / `optInNote` strings across all six locales
and add the `infoTooltip` string in their place.
The questionnaire, scoring, severity result, PHQ-9 item-9 crisis-resource
handling, focus-on-submit, and 44px touch targets are unchanged. Repoint
the module nav entry and the registry comments at the new route.
The authenticated shell keeps a single <main id="main-content"> mounted
across every route change, so its scroll offset persisted between routes:
scrolling down in one section (e.g. Settings) and navigating to another
opened the next view already scrolled.
Reset the scroll container to the top on each pathname change, with a
window.scrollTo fallback for the body-scrolled shells. Back/forward
navigation skips the reset so the browser's own scroll restoration stands:
popstate arms a one-shot skip cleared on a macrotask so it can never stick.
Bring the lab biomarker detail in line with the Insights metric pages:
- Lead with the static catalog description directly beneath the heading,
the explainer-caption position the metric pages use.
- Drop the early big-value + reference-badge block that did not match the
assessment-card styling.
- Split numeric and qualitative markers: numeric keeps the stat strip and
chart, qualitative shows only the latest result text (no strip, no chart,
no range badge). The assessment card now renders for any marker with
readings and self-suppresses for qualitative-only markers.
- Move the full reading history off the detail page onto a dedicated
/labs/[biomarkerId]/values route, reached by a top-right list control,
matching the metric pages' show-all-values affordance.
The inbound-document store keeps the uploaded doctor letter encrypted at
rest (InboundDocument.contentEncrypted, AES-256-GCM), but there was no way
to read the original back — the decrypt helper had been dropped as unused.
- Restore decryptDocumentFromBytes in the document store, fail-closed on a
bad / missing key id.
- Add GET /api/documents/inbound/[id]/original: owner-scoped (userId-narrowed
query), module-gated, returns the decrypted bytes with the stored MIME
type. A PDF / image is served inline so the browser can render it; other
types download as an attachment. The filename is sanitised before it
reaches the Content-Disposition header. The bytes are PHI — never logged,
never served cross-user, and a decrypt error fails closed (500) rather than
returning ciphertext. Cache-Control is private, no-store.
- Surface a 'View original' action in the document review screen.
- Document the route in the OpenAPI registry and regenerate the spec.
- Cover the route with a test: owner-scoped success, attachment fallback,
404 for another user's document, and fail-closed on decrypt error.
i18n: viewOriginal across all six locales.
The chronic-conditions and Coach-focus self-context read as foundational
medical history, so the Anamnese section now surfaces them as a read-only
shared view next to allergies and family history, with a link to their single
editing home in personal context. Personal context links back to the medical
history. The card is coach-gated and reads through the existing
/api/coach/about-me query key, so nothing is moved or duplicated and there is
no schema change.
Both account-security lists now render behind a disclosure control and open
only on click. The trigger carries aria-expanded and aria-controls and is
keyboard-operable; the region is gated with hidden. State is UI-only and not
persisted.
The new-chat hero carried a standing self-description line. Remove that
instance; the greeting and composer stay. The MedicalDisclaimer component
and its insights/labs/medications placements are unchanged.
Grip strength and waist circumference resolved against a single sex-aware
bracket for every adult age. Replace each with age x sex norm bands:
- Grip strength: age x sex bands from Dodds et al. 2014 (PLOS ONE
9(12):e113637) British normative data — band low is the 10th centile
floored at the EWGSOP2 probable-sarcopenia cut-off (men 27 kg, women
16 kg), band high is the 50th-centile median. Strength peaks in the 30s
and declines from midlife, so the bands step down with age.
- Waist circumference: WHO increased-risk threshold (men 94 cm, women
80 cm) held flat through midlife and raised for older adults toward the
WHO substantially-increased-risk line, since the WHO cut-points derive
from cohorts aged 20-74. The standard threshold is never lowered.
The norms resolver gains an age-independent clinical cut-off row per sex,
used only as the no-age fallback so a profile with a known sex but no date
of birth keeps the sex cut-off rather than the coarse registry anchor. The
signal registry structure is unchanged — only the resolver tables move.
- Reword two LOINC verification comments in the signals registry so they
no longer trip the forbidden-marker grep, keeping the meaning intact.
- Resolve the dead-code gate: gate the seeded-question opener on the
per-user ambient-suggestions opt-out (wiring the previously orphaned
resolveCoachAmbientSuggestionsEnabled, server-authoritative), drop the
unread decryptDocumentFromBytes helper and its now-unused import, and
remove the unused SignalKind type alias.
- Delegate undici's fetch to the global stub in the provider-chain
integration test so the egress-pinned client HTTP is mocked again.
- Defer the JSON example-download cleanup a tick so revoking the blob URL
no longer races the browser download start (stabilises the e2e).
- Sync the service-worker version fallback to v1.25.0.
The screener POST hardcoded source=WEB and ignored any client-supplied
externalId, so the native client could not tag provenance and an outbox
replay beyond the 24h idempotency-key window minted duplicate
administrations plus duplicate trend points.
Accept an optional closed source enum (WEB|IOS, default WEB) and an optional
bounded externalId, and persist both. Deduplicate on (userId, externalId)
when present: the route returns the existing administration on replay, and a
partial unique index over (user_id, external_id) WHERE external_id IS NOT
NULL (migration 0221) backstops a concurrent insert race (P2002 resolved to
the winner). The derived *_SCORE measurement keeps its stable per-assessment
externalId, so dedup at the assessment level prevents duplicate trend points.
Item answers stay encrypted and out of AI/MCP/export. Regenerate OpenAPI.
HealthKit emits HKQuantityTypeIdentifierWaistCircumference as a length
quantity in metres, but the mapping table had no entry for it, so a batch
upload of that identifier was silently dropped (mapAppleHealthEntry returned
null). Add the bidirectional mapping with metre-to-cm scaling so the round
trip lands on the canonical WAIST_CIRCUMFERENCE measurement (cm), matching
getUnitForType and the WHO display bands. Extend the mapping test.
Grip strength, pain (0–10 NRS), waist circumference, and waist-to-height
existed in the measurement enum, signal registry, and FHIR export but had
no web entry, no detail page, single non-sex-aware bands, and a split-brain
MCP exposure. This closes those gaps.
Web capture: the four signals join the manual measurement form, so a
web-only self-hoster can log them (the iOS client and the measurements API
already could).
Detail pages: flip detailPage on for the four signals and mount each on the
shared HealthKit metric template (chart, stat strip, Coach-read strip,
assessment card) via routed sub-pages and tab-strip pills, the same detail
spine every other metric carries.
Sex-aware bands: grip strength (EWGSOP2 men 27 / women 16 kg) and waist
(WHO men 94 / women 80 cm) now classify by the profile sex through the
age/sex norm resolver the assessment already threads; the resolver also
returns a sex-specific, age-independent band when no date of birth is set.
Fix the waist-to-height boundary so the NICE >= 0.5 rule is correct (the
last in-range value is 0.49).
MCP single source of truth: decouple the registry mcp facet from the Coach
snapshot — a signal can be MCP-readable without being on the Coach surface.
Mark the four physical signals mcp:true, derive the MCP rich-reads clinical
set from the registry flag instead of a duplicate allowlist, and relax the
invariant to coachSnapshot implies mcp (not the reverse). PHQ-9 / GAD-7 and
the environmental signals stay mcp:false and unreachable. compare_metric and
detect_changepoints now advertise clinical-signal support to match
get_metric_baseline.
The existing i18n guards prove key-shape parity but pass an English string
copied verbatim into another locale — the exact class of bug the v1.25
audit found (illness journal, PHQ-9/GAD-7 screener). Add a heuristic that
flags any es/fr/it/pl value byte-identical to en for a multi-word string,
with an explicit allowlist for the legitimately-identical keys (clinical
abbreviations, units, brands, OAuth labels) and auto-skips for URLs and
mask placeholders.
The detail page fetched a single limit=500 window and sorted client-side,
silently truncating any marker that crossed 500 readings. Switch to an
offset-paginated useInfiniteQuery: it loads a page at a time, a 'Load more'
control reveals the rest, and the chart and stat strip render the
accumulated set so loading more extends the trend.
The nightly discovery fan-out called isModuleEnabled per candidate user —
N serial DB reads on a large instance. Resolve the opt-in gate in-memory
from one findMany plus a single operator-availability read instead. The
per-day backfill upserts ran one awaited statement at a time (up to 730
for a full backfill); batch each location group's upserts into a single
transaction round-trip while keeping the idempotent semantics.
Wrap the home and travel city-search inputs in a form so Enter runs the
search instead of doing nothing, and show a 'no matching places' line when
a geocode returns nothing rather than leaving the user without feedback.
Validate the home timezone against the runtime IANA set (a bogus zone
would silently mis-key every stored environment day). Grow the result-row
and travel-remove controls to the 44px touch floor.
The inbound-document list and review panels used centered spinners; swap
them for the app's Skeleton pattern so the resolved content lands without
a layout jump. The low-confidence review badge moves from destructive red
to a calm amber, matching the surface's non-alarming posture. Approve /
reject / edit / save controls grow to the 44px touch floor, the upload
file input gains an accessible label, and each document row exposes
aria-expanded.
The sleep-debt computed-engine InfoHint only rendered in the still-learning
state — but the resolved full-debt card is the common case and where the
figure most needs the 'this is our own estimate, not a wearable's native
number' context. Show it there too. Separately, swap the period-narrative
loading shell's hand-rolled muted divs for the shared Skeleton primitive
so its shimmer matches every other loader.
The top-right Coach hand-off icon button and the info-hint trigger sat
below the WCAG 2.5.5 44px minimum on touch. Grow both hit boxes to 44px
on touch while keeping the visible footprint unchanged — the Coach button
is a fill-less ghost, and the info-hint offsets its padding with an equal
negative margin so inline layout is untouched.
The illness journal copy (load error, delete confirmation, edit/menu
labels, parent links, and the vital-sign labels) shipped in English in
every locale; translate it into de/es/fr/it/pl. Also add the new strings
the v1.25 surfaces introduce — the biomarker 'load more readings' action,
the environment city-search zero-result message, and an accessible label
for the document upload file input — across all six locales.
Pin the highest-consequence numbers in the release: every PHQ-9/GAD-7 severity
band edge, the >=10 action threshold, and the item-9 ANY-non-zero safety flag
(true on item9 1/2/3, false on 0 or for GAD-7). Add a POST happy-path test
that would have caught the dead-on-arrival bug — a valid body now parses,
persists server-authoritative scoring, writes the derived score measurement,
and surfaces the crisis set on a positive item 9.
The PHQ-9/GAD-7 totals flowed into the doctor-report / FHIR export
unconditionally while the AI/MCP surfaces were correctly fenced off. Map the
two `*_SCORE` measurement types to the mental-health module so the totals
export only once the account opts the module in (default OFF — privacy by
default), consistent with how mood/sleep/glucose gate. Item-level answers are
never a measurement, so they can never leak through this path.
On submit the form (and its submit button) unmounts, dropping focus to the
body — move focus to the result heading so keyboard/AT users land on the new
content. Announce the item-9 crisis-resource card via role="alert" so a
screen-reader user hears the resources. Bump the PHQ/GAD scale and
functional-difficulty option buttons to the 44px touch-target floor.
The handler passed the `safeJson` `{ data, error }` wrapper straight into
`safeParse`, so the schema never matched and every POST returned 422 — the
screener was dead on arrival. Destructure the envelope and read an unbounded
body no more: cap the request at 8 KB before validation, matching the sibling
clinical routes.
Add the missing write protections: gate both verbs on the opt-in module
server-side (a valid Bearer token included), wrap the create in idempotency so
a double-tap/retry cannot mint duplicate assessment + score rows, and add a
tight per-user write rate-limit.
The mental-wellbeing screener surface shipped built but unreachable: no nav
entry, no module key, ungated. Add a `mentalHealth` module key (opt-in,
default OFF — a depression/anxiety self-assessment is at least as sensitive
as mood) to the registry, the module-hub icon map, and the exhaustive module
maps. Give the check-in a nav home at /insights/mental-wellbeing, born-gated
on the module so it appears only once the account opts in, and redirect the
page home on a direct hit when the module is off.
Item answers stay encrypted and off the AI Coach / MCP regardless of the
toggle; this only controls whether the surface is visible and reachable.
The PHQ-9/GAD-7 item wordings rendered in English for es/fr/it/pl users
(de was already validated). Replace them with the officially validated,
public-domain Pfizer/PRIME-MD localized screener wordings (distributed via
the MAPI Research Trust) rather than ad-hoc translations — re-wording a
validated instrument invalidates the score. Localize the crisis-resource
copy too: translate the descriptive parentheticals while keeping the org
proper-nouns (TelefonSeelsorge, Nummer gegen Kummer, krisenchat,
Find a Helpline) and the US brand lines (988, Crisis Text Line) verbatim.
Add the new mental-wellbeing module label/description and the nav label
across all six locales.
The inbound-document store encrypts the raw upload
(`inbound_documents.content_encrypted`) because it is sensitive clinical
data, but the per-fact staging columns persisted the same class of data in
the clear: `extracted_facts.data_json` held the transcribed clinical values
(diagnosis text, lab values, medication names, stated codes) and
`provenance_json` held the verbatim source span (up to 2000 chars copied
straight out of the report). The at-rest guarantee was defeated for staged
facts, which persist through PENDING and are retained after the decision.
Convert both columns to the AES-256-GCM `*_encrypted` BYTEA convention shared
with the note columns, going through the same Bytes codec. The upload and
fact-edit paths encrypt on write; the confirm path decrypts at commit time to
write into the normal structured stores, so the review-then-confirm flow is
unchanged. Reads are fail-closed (a bad key id throws).
Register both columns in the encrypted-columns registry and the key-rotation
script so a dropped legacy key cannot strand them (the guard test enforces
both). Migration 0220 swaps the plaintext JSONB columns for the encrypted
BYTEA columns; pre-confirm staged rows are ephemeral and cannot be
re-encrypted in SQL, so any left by a pre-release build are dropped rather
than retained as plaintext (their parent document keeps its encrypted
original).
Reflect the v1.25 surface without touching the changelog or version
numbers.
- README highlights now cover the new clinical signals (respiratory
rate, the opt-in PHQ-9 / GAD-7 mental-wellbeing self-checks with calm
crisis handling, grip strength, pain score, waist / body measurements,
longevity labs), the signal registry that wires each one across charts,
correlations, the coach, FHIR, and the API at once, structured records
(allergies, family history; immunizations noted out of scope), the
opt-in environmental-context module, inbound document extraction
(transcribe, never interpret), and the self-description posture (a
record-keeping and wellness tool, not a medical device).
- ai-providers.md gains a "what the Coach is — and isn't" section: the
non-diagnostic standard and the calm proactive morning check-in
behaviour.
- mcp-capabilities.md notes that the new measurement signals reach the
assistant through the existing metric and labs tools via the registry,
that the mental-wellbeing instruments / records / documents are
deliberately not exposed over MCP, and ties the grounding contract to
the project-wide self-description standard.
- getting-started.md lists the environmental module's Open-Meteo egress
as an opt-in outbound dependency.
Introduce one canonical "what this is / is not, not a medical device"
statement plus a layered, calm disclaimer set, and apply it on the
surfaces that earn a context note.
- New shared `MedicalDisclaimer` component renders the copy as plain
React text (no markdown renderer, per the XSS posture). Variants:
intendedPurpose, coach, insights, labs, medications, emergency,
dataPosture.
- New `selfDescription` i18n namespace across all six locales: the
intended-purpose line, the per-surface notes, the emergency note, and
an explicit data-posture statement (self-hosted, data stays on your
instance, encrypted at rest, BYOK / local model, no third-party
analytics).
- Apply once per surface, layered not plastered: a calm standing line on
the Coach new-chat hero, one framing line in the Insights layout shell
(covers the briefing and trends), a labs note (ranges vary, only a
clinician interprets), a neutral medications note, and the data-posture
line where the provider key is configured. Mental-wellbeing keeps its
existing crisis copy; records keep their existing note.
- Extend the Coach system prompt (EN + DE) with a self-description ground
rule so the Coach answers "what are you / where does my data go"
honestly and stays non-diagnostic.
App-authored copy was already observational; the few diagnose/treat
tokens are negations, idioms, or framed as the user's own self-reported
statement, so no softening was required.
Bring the new physical clinical signals onto the MCP read surface through
the existing rollup-backed reads. Grip strength, pain (0-10 NRS), waist
circumference, and waist-to-height ratio now resolve in get_metric_baseline,
compare_metric, and detect_changepoints, are discoverable through search
(present-only), and hydrate through fetch (metric:<KEY>).
These signals sit off the Coach snapshot by design, so the MCP layer owns
an explicit exposure allowlist and consumes the signal registry only for
grounding (backing measurement type, unit, population band). The PHQ-9 /
GAD-7 mental-health screeners and the environmental signals are absent from
the allowlist, so they never resolve through any read tool. The longevity
lab panel is already reachable through get_labs and the lab resource
template; no change needed there.
Settings → Environment now reads the override list as "location periods": a place
plus a date range that covers a trip OR corrects a stretch of earlier history,
overriding the home for the days it spans. Show the home's effective-from date
and prefill the backfill start with it, so the default span matches what the
server will actually store. Strings updated across all six locales.
Rework the per-day location resolution so it can no longer fabricate weather for
days before the home was set. Precedence, first hit wins:
1. an explicit dated location period (a trip, or a corrected stretch of
history) covering the day;
2. DEVICE — a reserved slot for a future client-supplied coarse per-day
location (ranks between the period and home; not produced server-side yet);
3. the home location, but only for days on/after its effective-from date;
4. otherwise the day is skipped.
The home route stamps `homeSince` whenever the home is set/updated and returns
it. The overview exposes it. Backfill bounds are now optional and default to the
conservative `[homeSince .. today]` span instead of a fixed reach onto the
current home; days before `homeSince` still resolve to a skip in the worker, so
even a wider explicit start cannot back-fill weather for the pre-home past. To
reach further back a user adds an explicit location period for that range.
Add unit coverage for the precedence (including the effective-from boundary and
the skip-not-guess rule) and the default backfill range.
Add `User.homeSince` (nullable effective-from instant) and a `DEVICE` value to
the location-source enum (migration 0219). `homeSince` records when the home
location became effective so day resolution can attribute a past day to the
home only on/after that instant, rather than to the current home for every past
day. Existing accounts with a home get `homeSince` stamped to the migration
instant; their deep past is filled by explicit location periods instead.
`DEVICE` is reserved for a future client-supplied coarse per-day location: the
enum value and the wire contract accept it now so the resolver can rank it
later without rework. Nothing writes it server-side yet.
Add the opt-in Environment module (default off), its Settings surface (home
location search, travel overrides, backfill, with Open-Meteo attribution), the
gated API routes (overview, home, geocode, travel CRUD, backfill), the nightly
pg-boss fetch job with a lookback window plus on-demand backfill, the query
keys, six-locale strings, and the regenerated OpenAPI. The module performs an
outbound weather fetch tied to a coarse location, so it ships dark and the user
turns it on deliberately; off means no job, no row, no egress.
Register the seven environmental-exposure signals through the signal registry
as a new "environment" kind (correlation-eligible; off Coach/MCP/FHIR in v1),
and fold their daily series into the correlation engine as lagged behaviour
channels. Weather/daylight now pair against mood, sleep, and vitals through
the existing FDR / effect-size / shrinkage gates, so a thin series degrades to
absent rather than a spurious link. All env channels share one ENVIRONMENT
family so two same-day fields never lag-correlate against each other.
Add the keyless Open-Meteo client (historical archive + geocoding) routed
through safeFetch, the per-day location-resolution + fetch/upsert service
(travel override wins over home, then degrades silently), and the shared
environmental-exposure field table the registry and correlation engine both
read. The base URL is operator-overridable for a self-hosted instance; the
fetch stays server-side so a private/LAN host is not blocked.
Add the opt-in environmental-context store: per-day EnvironmentContext
observations (temperature, daylight, sunshine, precipitation, pressure,
humidity, cloud, weather code) keyed by user + day and location, plus the
manual EnvironmentTravelLocation overrides and the coarse home-location
columns on the user. Location is city-level by design; the weather model
needs no finer than ~10-25 km and no GPS track is ever stored.
Document the /api/documents/inbound/* surface in the OpenAPI registry and
regenerate the spec. Add the inbound-documents copy across all six locales,
including the non-diagnostic disclaimer and the module label/description.
Add the opt-in inboundDocuments module (off by default, like the MCP
endpoint): ingesting a clinical document egresses it to the OCR/vision
provider, so the surface ships dark and the user turns it on from the
Modules hub. The born-gated /documents page carries the prominent
non-diagnostic disclaimer (we read, we don't interpret; nothing is saved
until you approve it; only your clinician can interpret these results),
uploads a document, and walks each extracted fact: source provenance and
confidence are shown, low-confidence facts are blocked from approval until
edited, and the confirm step commits the approved facts.
Ingest an uploaded doctor report / discharge letter through the existing
dedicated OCR/vision provider rails (the v1.22 document-scan provider/key,
reused — no new provider) and transcribe structured facts into staging.
The safety line is extract, never interpret: the prompt frames the document
as untrusted data to reproduce verbatim. Diagnoses map to Condition,
measured values to Observation, medications to MedicationStatement — with
STATED status only. A code survives only when the document writes it and its
system matches the resource (SNOMED/ICD-10 for a diagnosis, LOINC for a
value, RxNorm/ATC for a medication); nothing is inferred, no value is
range-flagged, no condition is linked to a medication. Every fact carries
per-field provenance and a confidence score; a fact below the confidence
floor fails closed — it cannot be approved until the user edits it.
Routes (apiHandler + Zod + session-narrowed userId, opt-in module-gated):
- POST/GET /api/documents/inbound — upload+extract (vision or local-OCR
text), and list. The raw document is stored encrypted at rest.
- GET/DELETE /api/documents/inbound/[id] — review detail; soft-delete.
- PATCH .../facts/[factId] — correct a staged fact before approval.
- POST .../confirm — the only write path: approved facts flow into the
labs / conditions / medications stores through their normal create paths;
rejected facts are discarded; low-confidence facts are refused.
Store an uploaded doctor report / discharge letter and stage extracted
structured facts for review. Two models:
- InboundDocument: the uploaded document, raw bytes encrypted at rest via
the shared AES-256-GCM Bytes codec (content_encrypted), with queryable
metadata (kind, mime, size, status, report date).
- ExtractedFact: the review-then-confirm staging row — fact type, status,
per-field provenance, a confidence score, and a needs-review gate, plus
the FHIR-staged payload (stated codes/status only) as JSONB.
Wire content_encrypted into the encryption-key rotation registry and the
rotation script so a key drop never strands a stored document.
Rework the proactive Coach nudge so it reads like one human checking in on
another instead of a terse English alert.
Copy: the title is a name-led greeting that rotates day to day ("Guten
Morgen, …" / "Hey …"; the 05:15 frame makes a morning opener safe) and the
body is one calm observation plus a gentle invite — never an imperative.
Drop the verbatim quoting of the user's decrypted self-context sentence:
the focus is referenced only by presence, abstractly, and is no longer
decrypted at all. Remove the 80-char mid-word clamp; each channel composes
to a length budget instead.
Cadence: suppress the nudge when the user already talked to the Coach in
the last ~24 h. With the existing rolling frequency cap (never two in a
window) and the no-trigger-no-nudge rule, the surface stays anti-nag.
AI enrichment (opt-in via the coach prefs blob, default off): when a
provider is healthy the body is composed through the model under hard
guards — a per-user budget gate, a tight per-call timeout, a per-tick
ceiling so the sequential tick can't stall across the user base, and the
outbound safety screen on the output. Any error, timeout, budget refusal
or unsafe reply falls silently back to the deterministic template, which is
always the default and the fail-closed fallback.
Localized greetings, per-trigger bodies and the focus note across all six
locales; the prefs opt-in is surfaced in the notification settings.
The language switcher wrote the choice to localStorage, the
`healthlog-locale` cookie and `<html lang>` only — never to the user row.
Server-side background work (the proactive Coach nudge, the Telegram test
message) has no request cookie to read and falls back to `User.locale`;
with the column left null those messages went out in English regardless of
the language the user reads the app in.
Add `PUT /api/auth/me/locale` (idempotent, validated against the supported
locales) and have the i18n provider mirror the active locale onto the row —
on every switch and once on mount, so the column is backfilled for users
who set their language before this landed. Fire-and-forget: a 401 on a
public page or an offline blip never blocks the UI flip; the cookie stays
the first-paint source of truth.
DELETE /api/measurements/by-external-ids now only tombstones rows the iOS
client minted through its HealthKit ingestion path (source APPLE_HEALTH).
An externalId the app never minted — an integration- or manually-sourced
row that happens to collide on externalUUID — falls outside the predicate
and is a no-op rather than a delete. Defence-in-depth so a foreign
HealthKit sample with a colliding UUID cannot be tombstoned by the
periodic reconciliation. The delete stays user-scoped as before.
Add a route test proving a foreign-origin externalId is not deleted, and
regenerate the OpenAPI spec for the narrowed contract.
The structured lab-create path (POST /api/labs) now accepts an optional
`source` marker ("MANUAL" | "OCR"). The on-device-OCR ingestion path posts
"OCR" so a reading captured from a phone-camera lab photo records its
provenance; the raw image never touches the server on this path. Omitting
the field reads as "MANUAL", so the existing hand-entry contract is
unchanged. The value is narrowed by a closed Zod enum and assigned
field-by-field, never spread from the body. Mirrors the "OCR" source the
server-side /api/labs/ocr/commit path already stamps.
Regenerate the OpenAPI spec for the widened request body.
The PHQ-9 / GAD-7 screener totals and the physical clinical measures
(grip strength, pain NRS, waist circumference, waist-to-height) are
continuous numeric series. Register their metric:<TYPE> tokens so the
chart-token drift guard passes and the AI insight prose can reference
them through the generic chart renderer.
Add the opt-in PHQ-9 / GAD-7 questionnaire surface at /insights/mental-wellbeing:
instrument picker, 0–3 item scale, server-scored result with the validated
severity band, the verbatim Pfizer/PHQ attribution, the screening-not-diagnosis
disclaimer, and a calm crisis-resource card on a positive item-9. The surface
deliberately does NOT mount the shared Coach-launch card — mental-health item
content stays out of the AI.
- i18n: full mentalHealth block plus the six new measurement-type labels and the
longevity-marker catalogue names/descriptions across all six locales
(de/en/es/fr/it/pl). Instrument item wording uses the validated German text and
the canonical English elsewhere (re-wording invalidates the score).
- OpenAPI: regenerate the contract for /api/mental-health/assessments (response
excludes raw item answers by construction).
- Exclude the new manual / instrument types from the HealthKit-counterpart guard.
Generalise sleep debt onto the existing per-metric source-priority picker
(a new sleepDebt ladder, native-above-computed, default WHOOP > Oura > Polar >
computed) instead of assuming one provider. The rhythm read resolves the active
source off the ladder and carries it on the DTO; when the active source is our
own COMPUTED engine — every user today, since no provider ships a native
sleep-debt balance — the Sleep page explains what the figure means and that it
differs from a wearable's native number. The computed model itself is unchanged
and correct (a short rolling balance of need minus sleep, surplus-credited);
only stale comments that still referenced the old 14-night window are corrected.
Add a per-user toggle that turns off the proactive ambient example suggestions
(the daily seeded opener on the Coach hero). The flag rides the existing coach
notification-prefs blob (coach.ambientSuggestions, default on) — no schema
change — and a shared hook makes it the single source every ambient suggestion
surface consults, so the seeded opener and its fetch both stand down when off.
The weekly/monthly review card rendered an internal methodology line plus the
raw MeasurementType tokens it drew on ('Based on: WEIGHT, BLOOD_PRESSURE_SYS…'),
which read as broken and added no value. Drop the provenance caption entirely so
no enum token is ever user-facing, and hide the card whenever there is no real
generated narrative (no 'preparing' husk), consistent with the hide-when-empty
pattern. The query keeps polling in the background, so the card appears the
moment a narrative lands.
The short-horizon trajectory tile read as a wall of text with no clear value
on the metric pages. Drop it from the weight page and the shared HealthKit
metric page (the opt-in forecast slot is removed entirely, not left dead), and
delete the now-orphaned tile + fan components. The trajectory engine stays for
the Coach narrative; only the metric-page surfacing is removed.
The composer mic advertised dictation but errored on tap in environments where
the Web Speech API cannot serve. Gate the control on the API being present AND
a secure context, map the locale to a region-qualified recognition language,
and retire the button for the session on a permission refusal. When dictation
is genuinely unavailable the mic hides rather than showing a control that
errors on every tap.
The ambient example opener was stretched to the full composer width, leaving
a large empty tail that read as broken. Size it to its text instead (capped at
the full width) and guarantee a single line: the question truncates with an
ellipsis on a narrow phone rather than wrapping or forcing a horizontal scroll.
Add a new "Medical history" (Anamnese) Settings section that hosts the two
structured-record managers — allergies and family history — each a
list + add/edit sheet + delete-with-confirm built on the shared query-key
factory and the ResponsiveSheet pattern. It sits between the illness journal
and the preventive-care reminders and is always shown (not module-gated, like
Vorsorge), since these are foundational health-profile records.
The section carries a plain non-diagnostic note ("your own records, not a
medical diagnosis"). Full i18n across all six locales (de/en/es/fr/it/pl).
Emit `AllergyIntolerance` and `FamilyMemberHistory` resources in the
health-record document bundle. Vendor-blind and non-diagnostic, matching the
illness `Condition` stance: the user's own label rides `code.text` (never a
machine-guessed clinical code); only the FHIR-fixed value sets — allergy
category, clinical/verification status, the v3 RoleCode family-member concept —
carry codings. Every resource is marked patient-reported (allergy
`verificationStatus: unconfirmed` plus an explicit "not a clinical diagnosis"
note) so a clinician's viewer never reads it as an adjudicated finding.
The records are loaded owner-scoped in the export route and folded into the
fhir + package bundle formats with their own Composition sections.
Add the `/api/allergies` and `/api/family-history` endpoints (list / create /
read / update / soft-delete), each wrapped in `apiHandler` with Zod
`safeParse` + `returnAllZodIssues`. `userId` is always narrowed from the
session/Bearer — never a body field — and the Prisma `data` is built
field-by-field (no mass assignment). Creates wrap in `withIdempotency`. The
free-text reaction/note are encrypted before write and decrypted fail-soft on
read via the server-authoritative DTOs.
Register the query-key factory entries and the OpenAPI route tables (new
`Records` tag); regenerate `docs/api/openapi.yaml`.
Introduce two structured health-record tables, kept as reference records
(not time-series signals):
- `allergies` — AllergyIntolerance-style: substance + category + type +
severity + status + optional onset. The substance label stays queryable
plaintext; the free-text reaction and note are AES-256-GCM encrypted
(`reaction_encrypted` / `notes_encrypted` BYTEA).
- `family_history_entries` — one condition per relative: relationship +
condition label + optional age-at-onset; the free-text note is encrypted.
Both are owner-scoped and soft-deleted. Migrations 0215/0216 are additive
and idempotent. The three new encrypted columns are registered in
ENCRYPTED_COLUMNS and wired into the key-rotation script, so a key drop can
never silently strand them undecryptable.
Add the validated mental-health self-assessment surface, beside mood tracking
(never replacing it) and off by default:
- MentalHealthAssessment stores per-item answers in a single AES-256-GCM Bytes
blob (registered for key rotation); only the server-authoritative total,
severity band, and item-9 flag are denormalised. The total projects to a
PHQ9_SCORE / GAD7_SCORE measurement row so the trend rides existing infra
without ever exposing item content.
- Item-9 safety: any non-zero PHQ-9 self-harm answer surfaces a calm,
locale-aware crisis-resource set (DE: TelefonSeelsorge 0800 111 0 111 / 112;
international fallback) and stamps crisisShownAt. No clinical interpretation,
no AI ingestion of item content, no third-party alerting.
- API + OpenAPI: POST/GET /api/mental-health/assessments; the response excludes
raw item answers by construction. Required Pfizer/PHQ attribution lives in the
instrument module, rendered verbatim.
The screener scores are kept off the Coach snapshot and MCP surfaces, and out of
exports by default — item content never reaches the AI pipeline.
Add the clinical-signals wave's new measurement types and wire them through
the signal registry so FHIR, correlation eligibility, ingestion, and the
measurement-list surfaces derive from one definition each:
- MeasurementType: PHQ9_SCORE, GAD7_SCORE, GRIP_STRENGTH, PAIN_NRS,
WAIST_CIRCUMFERENCE, WAIST_TO_HEIGHT (+ AssessmentInstrument enum and the
MentalHealthAssessment model that backs the screener scores).
- Registry: the six measurement/score signals plus the longevity lab panel as
kind:"biomarker" entries; the FHIR golden fixture is regenerated in lockstep.
Grip-strength and waist-to-height carry a null LOINC (NEEDS-VERIFY) and emit a
local text concept until a stable code is confirmed.
- FHIR: add the survey observation category for the screener / pain signals.
- Labs: extend the biomarker catalogue with ApoB, Lp(a), fasting insulin, and
the omega-3 index; add LOINC codes for those plus GGT and hs-CRP; add a
computed HOMA-IR helper.
- Exhaustive maps updated to keep the enum coverage, category, list-meta, and
personal-record drift guards green.
The requirePublicHost egress paths now dispatch through undici's own
fetch, so the existing global fetch stubs in the AI, GeoIP, moodlog,
send, and ntfy suites no longer intercept them. Delegate undici's fetch
back to the installed global stub so each suite keeps asserting its
request shape and canned responses without exercising the network.
CIMD client resolution swallowed every fetch failure as ssrf_blocked and
/authorize logged a bare invalid event, so an unreachable metadata host
was indistinguishable from a genuine SSRF block and the consent flow
returned invalid_client with no diagnosable cause.
Split the resolution failure into ssrf_blocked (a real private-host
refusal) and fetch_failed (timeout / connect error), carrying the
concrete SafeFetchError kind and cause as a short non-secret detail.
Thread the resolution reason and detail into the authorize.invalid
annotation behind a pinned meta shape, and cache a resolved CIMD
document for up to 24h so the consent flow does not re-fetch on every
hit.
The connect-time SSRF pin returned only the first vetted address and
attached an undici@8 Agent to Node's built-in fetch. Two failures:
- A host whose first DNS record is an unreachable IPv6 (no IPv6 route on
the container / LAN / Tailscale) failed the connect instantly, because
the single-address pin discarded the working IPv4 record behind it and
Happy Eyeballs never ran.
- Node's global fetch is backed by its own pinned undici copy; handing it
an Agent built from the project's undici throws UND_ERR_INVALID_ARG on
a dispatch-handler interface mismatch.
Hand undici the full vetted survivor set and enable autoSelectFamily so
it runs RFC 8305 Happy Eyeballs across the already-vetted IPs (rebinding
stays closed: every dialable IP passed isPublicIp in the same lookup).
Drive the pinned path through undici's own fetch so the dispatcher and
the fetch engine share one undici version.
Add an end-to-end regression that drives a real localhost-bound server
through the pinned path (not a mocked fetch, which hid the coupling),
covering the IPv6-to-IPv4 fallback, streaming, timeout, and abort.
The sleep page rendered an assessment under the duration/architecture charts
and a second one under the sleep-quality scores. The quality assessment mostly
never generated — most accounts carry no wearable sleep score — so it sat at the
foot of the page as a duplicate empty card or a redundant connect-a-provider
prompt.
Drop the quality-block assessment so the duration card is the page's single
assessment; the quality block keeps its score tiles. Remove the now-unused
grounded-note fallback and its deterministic grading helper, and the orphaned
quality-assessment message keys across every locale.
The per-metric assessment card carried the Coach hand-off as a bottom-right
text button, present on only some pages. Pin it to the card header's top-right
as a single icon button (no text label) with an accessible name and a tooltip,
and wire every assessment surface to it: the bespoke slug pages
(weight, BMI, pulse, blood pressure, mood), the generic metric scaffold, the
medication-compliance card, and the biomarker detail page already on the shared
card. Each opens the Coach with a localised, auto-sent opener narrowed to that
metric's snapshot source.
The button self-gates on the Coach triple (provider mounted, operator flag,
per-user opt-out) and the card only mounts it on the populated branch, so it
never advertises a control that cannot work. The auto-sent opener now
interpolates the localised measurement name rather than a raw token.
Replace the hand-maintained MEASUREMENT_LOINC literal and the BUCKETED_TYPES
correlation list with signal-registry projections. Both reads are byte-identical
to the previous tables (guarded by the registry-invariant test), so there is no
behaviour change — only one source of truth for each.
Introduce one declarative table describing each health signal across the
surfaces it touches — DB measurement type, detail page, correlation engine,
Coach snapshot, MCP read tool, and FHIR coding — discriminated on kind
(measurement / biomarker / score). Promote the proto-registry in
metric-status-registry.ts to the canonical definition and add thin adapters
that derive the per-surface tables from it.
The registry coexists with the enum-driven code: the DB enum stays the source
of truth for persistence and the registry sits above it. An invariant test
pins every derived table byte-for-byte to the values the hand-maintained
tables carry today, so cross-surface drift fails CI.
Drive the biomarker resource route against a real Postgres: PUT persists a
new reference window and rejects an inverted one, and DELETE drops the marker
together with its readings while leaving another user's marker and readings
intact.
Add coverage for the new biomarker-assessment seam and the marker detail
surface:
- the generator serves today's cache, re-stamps unchanged readings without an
LLM call, and regenerates only when the latest-reading fingerprint changes
- the read-only endpoint returns the envelope, 422s a missing id, and honours
the auth + insightStatus gates
- the detail page paints the stat strip, the chart placeholder, and the
assessment card
- the biomarker resource route drops the marker with its readings in one
userId-narrowed transaction and rejects an inverted target range
Hold the input gate independent of the worker's forced run so an idle marker
re-stamps cached text instead of regenerating, and list the new route on the
non-Coach gate allowlist.
Give each biomarker a detail page that matches the metric sub-pages: a
numbers-first Min/Max/Median/Mean strip over the numeric readings and a
data-grounded AI assessment card.
The assessment is served by a new read-only endpoint
GET /api/insights/biomarker-assessment?biomarkerId=&locale=, returning the
same envelope as /api/insights/metric-status so InsightStatusCard consumes
it unchanged. A new generator reads the marker's LabResult history and
reuses the shared provider runner, the stale-while-revalidate read-only
path, and the persist helper. Regeneration is gated on a fingerprint of the
latest reading, so the assessment is rebuilt only when a new reading lands;
an idle marker re-stamps the cached text without an LLM call. The
biomarker: scope rides the existing on-demand status queue.
The stat strip and the assessment card both self-gate on numeric data, so a
qualitative-only or brand-new marker paints neither.
Add a control beside the marker actions that opens a focused sheet for
the reference low and high only, so a user can tune the band without the
full marker editor. Persists through the existing biomarker PUT and
invalidates the detail, catalog, and result reads.
The per-marker catalog description used to render on every labs overview
row, both in the list and the card layout. It reads better as a caption
beneath the detail-page heading, mirroring the metric pages' explainer
line. Resolve the catalog slug from the marker name and fall back to the
user's own context note.
Hold the submit handler in a ref updated from an effect and dispatch
the auto-send through a microtask, so the first-turn send no longer
calls setState synchronously inside the effect.
Deleting a biomarker now removes its recorded values in the same
userId-narrowed transaction instead of unlinking them via the SetNull
FK. Removing a marker is a deliberate "drop this and its history"
action, so leaving orphaned, unlinked readings behind was surprising.
Update the delete-confirmation copy in all six locales to state that
the values are removed too.
Thread an autoSend flag through the coach-launch flow so a card's
"ask the coach about this" hand-off dispatches its seeded prompt as
the conversation's first turn instead of only seeding the composer.
The flag rides the launch context, the layout mount, the drawer, and
the shared conversation surface, where a ref-guarded mount effect fires
the send exactly once for a fresh thread.
Localize the previously hardcoded English assessment opener in the
metric status card behind insights.coach.assessmentPrompt with a
{metric} placeholder, and localize the resting-pulse coach prefill.
Both keys land in all six locales.
Convert the per-biomarker reading history from a static reverse-chronological
list to a shadcn Table with clickable sort headers. Reuse the shared
useTableSort hook and SortableHead chrome already backing the measurement and
mood lists; the sample-date column opens descending, value/range/note open
ascending. The sort runs client-side over the in-memory reading feed.
Lift the byte-identical sortBy/sortDir/toggleSort triple the measurement
and mood lists each carried inline into a shared useTableSort hook. Both
lists now compose their page-reset and selection-clear side effects around
the hook's toggle rather than re-implementing the direction math, and the
date columns keep their descending-first default via descColumns.
Surface the generation-failure signal in the daily briefing card. When a
held briefing is shown but the last refresh failed, the footer states it
plainly and offers a retry that runs the existing explicit regenerate —
no warm-on-mount. When there is no briefing to fall back on, the empty
state reads "couldn't generate" with the same retry instead of the
generic "no briefing yet". The hint is suppressed when no AI provider is
configured, where that hint already owns the surface and a retry is
futile. New copy lands in all six locales.
The daily briefing and the per-metric status assessments ignored the
per-user response-timeout setting: they hardcoded their per-surface
AI_BUDGETS default, so raising the timeout for a slow self-hosted or
local backend had no effect and the surfaces still aborted early. Only
the coach threaded the setting through.
Resolve the effective timeout from User.aiResponseTimeoutSeconds in one
shared helper and thread it onto every non-coach generation call:
- runStatusCompletion is the single chokepoint every status / reference
surface funnels through (per-metric cards, the batched assessment, the
derived assessments, the period narratives), so resolving it there
applies the setting to all of them at once — on both the upstream
call's own timeout and the outer wall-clock cap that previously clipped
a raised value at 60 s.
- The comprehensive briefing generator and the user-initiated regenerate
route thread it onto the main generation, the JSON retry, the grounding
retry, and the paragraph re-roll.
On a failed generation the briefing keeps its last good text — no cache
row is written, so the previous payload stays intact — and records a
dated, append-only failure marker (no migration; reuses the audit-log
JSON pattern). The read path compares the marker against the last
successful generation so it can tell a held briefing from a current one.
A render failure on `/c/<token>` previously fell through to the full-app
error screen, which carries app chrome the public share view never wants.
Add a route-scoped, chrome-free error boundary that shows a localised,
stack-free message and a single retry. Add the matching copy across all
six locales.
The passphrase gate is a client component, but the page passed it a `t`
translator function. A function cannot cross the server→client boundary,
so a fresh (passphrase-protected) link threw a generic server-components
render error instead of painting the unlock form. Pre-resolve the gate
copy on the server into a plain, serializable `strings` object and pass
that instead. Locale resolution stays server-authoritative.
The page test mocked the gate away and only checked the token prop, so
the regression slipped through. Tighten it to assert the gate receives a
serializable contract: no prop value is a function, and the copy arrives
as a pre-resolved plain string object.
Move DOCTOR_REPORT_TYPE_LABEL_KEYS into a jsPDF-free module so leaf
consumers can import the map without dragging jspdf + jspdf-autotable
into their graph. The PDF core re-exports the map, keeping a single
source of truth and its existing import surface intact. Point the
clinician share view at the new module so the public share route no
longer pulls the PDF renderer into its server graph.
Remove the in-app water-intake logging feature: the WATER_INTAKE
measurement type, the per-user daily-goal field, the goal-ring card and
its endpoint, and the related query keys, OpenAPI route, and locale
strings. The feature is dropped from this release; the baseline-drift,
breathing-screening, and labs-changes insight cards are unaffected, and
the pre-existing TOTAL_BODY_WATER body-composition type stays intact.
A cluster of related correctness gaps in the compliance denominator and the
insight surfaces that read it:
- Paused medications. A doctor-ordered pause flipped `active` off and stamped
`pausedAt`, but resume nulled `pausedAt`, so the interval was irrecoverable
and every paused day became a missed dose after resume — collapsing
adherence across the Health Score, the BP gate, the Coach snapshot and the
report. Persist pause intervals as durable eras (new `MedicationPauseEra`,
migration 0211), close the open era on resume, thread the eras through the
compliance context and drop slots inside a paused interval from the
denominator. Every compliance call site now fetches and passes the eras.
- Health Score eras. The Health-Score fast path was the only compliance
caller that skipped the shared schedule select and schedule revisions, so
after a schedule edit it scored past days against the current schedule
while every other surface used the historical era. Switch it to the shared
select and thread revisions (and pause eras) like the canonical path.
- Insight bucketing timezone. The per-metric status series bucketed every
user in Europe/Berlin, so a near-midnight reading landed in the wrong day
and the relative-day phrasing was off for non-Berlin self-hosters. Thread
the resolved user timezone and key off `userDayKey`.
- Correlation significance. User-facing correlations surfaced from five
paired points with strength read straight off the coefficient. Route them
through the rigorous engine (at least twenty pairs, Student-t p below 0.05)
so small-N coincidences no longer render as confident findings.
- Resting-pulse input. The mood-versus-"resting pulse" correlation fed the
raw PULSE day-mean, which folds in workout HR, under a "Resting pulse"
label. Feed `resolveRestingPulseSeries` (clean RESTING_HEART_RATE, else a
daily low-percentile proxy) so the label matches the input.
A sustained fever (three calendar-consecutive days at or above the fever
red-flag floor) or sustained low SpO2 was already classified "seek care",
but the only escalation path ran inside a GET on the episode correlation
route — no push fired unless the user opened that tab, and no cron drove
it. The proactive on-write safety check covered only blood pressure and
glucose; the two signals with the strongest safety case had the weakest
delivery.
Add sustained-run temperature and SpO2 checks to the on-write
`runSafetyFloorCheck`, wired from both measurement-ingest seams. The run
length and clinical floors are shared with the retrospective detector (no
duplicated thresholds), day bucketing honours the user's timezone, and
escalation reuses the existing red-flag notification with its push-attempt
dedup so a febrile user logging repeatedly is alerted at most once per
reason per day.
`estimateDailyDoseCount` only special-cased FREQ=MONTHLY and FREQ=YEARLY.
A modern FREQ=WEEKLY rrule (the encoding the create path stores, with an
empty daysOfWeek) fell through to the legacy branch at seven days per week,
so a once-weekly injection counted as one dose per day. The detail card's
"lasts about N days" and the daily low-stock job — which share this math —
fired reorder alerts roughly seven times too early (fourteen for bi-weekly).
Parse FREQ=WEEKLY honouring INTERVAL and BYDAY count:
perDay += (times * bydayCount) / (7 * interval).
The two cumulative drains soft-deleted per-sample rows and wrote a daily
total but never recomputed the affected DAY rollup bucket. The rollup tier
reads only `deleted_at IS NULL`, so the stored `sumValue` kept the
pre-drain double-counted total indefinitely — dashboard and every rolled-up
week/month/year step or energy figure stayed wrong and never self-healed,
because the backfill only re-flags days with no DAY row and the
daily-series read-swap only re-folds on a day-count change.
Call `recomputeBucketsForMeasurement` after each per-day write, mirroring
the daily-mean drain that already does this. Added per-drain (not hoisted)
so the mean drain, which recomputes inside its own writer, is not
double-recomputed.
Date-only daily scalars from Oura and Polar were stored at UTC midnight
(`${day}T00:00:00.000Z`). On read these rows re-bucket through
`userDayKey(measuredAt, userTz)`, so for any west-of-UTC user the day
shifted back one: a June 1 SpO2 or readiness row filed under May 31 and
fell out of alignment with the HR/sleep rows that carry true instants.
Extend the existing `canonicalDailyTimestamp` helper to accept an optional
timezone and anchor a date-only day at local noon (noon UTC when no zone
is supplied) — a full twelve hours inside the calendar day, so it
round-trips through `userDayKey` to the same day for every zone within
plus/minus twelve hours. Route every Oura day-keyed mapper (via the shared
`dayAnchor` fallback) and all five Polar daily mappers through it; real
wake/measurement instants stay untouched. Fitbit already anchored at noon
and is now consistent; Whoop and the Withings cursor anchors were verified
not to be measurement anchors.
Add three read-only awareness cards to the Insights overview, each drawn
off an existing engine and each honest about empty state — the section
un-mounts rather than surfacing a husk.
- Health status: surfaces what is drifting from the user's own normal by
folding the personal-band deviations (coincident-deviation engine) with
the dated, sustained level shifts (changepoint detector) at
GET /api/insights/health-status. Awareness only, never a diagnosis.
- Breathing screening: summarises the last 30 nights of the per-night
sleep-breathing-disturbance index plus device-flagged events at
GET /api/insights/breathing-screening. Mirrors the device's own
classification and states plainly that it is a screening signal, not a
diagnosis — discuss with a clinician.
- Lab changes: pairs the two most-recent numeric lab panels and reports
each shared analyte's signed delta and reference-band standing at
GET /api/insights/labs-changes. Neutral framing only.
Each reshaping helper is pure and unit-tested for its present/absent
states. Register the three sections in the insights-layout universe so
they merge in as default-visible, wire them into the overview and the
edit-mode title map, add the query-factory keys, document the routes in
the OpenAPI contract, and translate every new string across all six
locales.
Add a WATER_INTAKE measurement type and a per-user daily goal so the
Insights overview can track water intake against a goal ring. Logging
rides the existing measurements create path (type WATER_INTAKE, unit ml,
summed cumulatively); GET/PATCH /api/hydration read the day total and own
the goal. The goal-ring math (empty / partial / met / exceeded) lives in
a pure helper, free of Prisma and React, and is unit-tested in isolation.
Wire the new type through the measurement enum, value ranges, cumulative
HealthKit handling, categories, display transform, list metadata, record
direction, and the chartable-token allowlist. The HealthKit DietaryWater
import stays on the nutrition hold, so the type carries no import mapping
by design.
Document GET/PATCH /api/hydration in the OpenAPI contract.
Cap and page the unbounded reads that let a multi-year tenant pull its
whole measurement history into memory on hot paths.
- extractFeatures: read newest-first with a 6000-row cap (mirrors the
Coach snapshot) and reverse to ascending, so a 365-400 day window no
longer materialises hundreds of thousands of PULSE / glucose rows per
Coach turn and nightly briefing. Full-history extremes still come from
readAllTimeExtremes, so the cap never relabels an extreme.
- Add a (userId, measuredAt) index on measurements (migration 0210) for
the type-less range scans that extractFeatures and the consolidation
passes drive; the existing (userId, type, measuredAt) cannot serve them
as a prefix.
- runConsolidation: keyset-paginate the per-(user, type) source scan on
(measuredAt, id) instead of one unbounded findMany, removing the giant
single result-set allocation that drove the boot storm. Output is
identical; covered by a paging unit test.
- Stagger the boot-time backfill enqueues with increasing startAfter
offsets so the heavy full-history loads no longer all drain in parallel
onto one tenant at startup. Non-boot callers keep their immediate send.
- Backups: replace load-all-then-stringify with paged narrow-select reads
for the weekly and off-host writers, bounding transient heap. Off-host
still includes soft-deleted rows. Covered by a serialize/restore
round-trip test.
- Drop the per-request auth.bearer.success audit insert; the wide event
already records auth_method and user_id, so the polling client no longer
drives a write + pool checkout on every request. Failure-path audit
stays.
- Coach conversation detail: cap the message include to the newest 200 so
a long thread no longer decrypts every message on each open; response
shape unchanged.
- Admin audit-log: floor the scan at the retention horizon so the listing
and its count stop sweeping open-ended over the substring filters.
Encrypt MedicationSideEffect.notes, MedicationDoseChange.note and
MedicationInventoryItem.notes with AES-256-GCM, closing the last plaintext
PHI columns left after the v1.23 note-encryption rollout. Add the three
*Encrypted Bytes columns (migration 0208), register them in the encrypted-
column rotation set, route every read/write through the note cipher, and add
a transactional fail-closed boot-time backfill for legacy rows. Plaintext
columns drop in a later release once the backfill reports zero remaining rows.
Require a fresh second factor on password change for MFA-enrolled accounts,
matching the existing step-up on MFA-disable and recovery-code regen. Redact
free-text note values from the Zod validation excerpt that can reach a wide
event.
Add isError branches with a retry affordance to the mood, illness, and
vorsorge lists so a failed read no longer renders as "no entries yet"
and invites duplicate re-entry, matching the labs and cycle pattern.
Make the admin backups table reachable on a phone with a card fallback,
name the admin language and timezone selects and the AI raw-data switch
for screen readers, and replace the touch-unreachable title= help on the
glucose clinical panels with a tap-capable info popover.
Route the password-strength label through the AA-tuned semantic tokens,
underline the register login link so it is not distinguished by colour
alone, give the labs load error alert semantics, and explain why the
target editor Save button is disabled with aria-invalid and a live
alert.
Subtract both the top and bottom safe-area insets from the full-page
Coach height so the composer no longer slides behind the bottom-nav on a
notched phone, and pad the Coach drawer composer and the ResponsiveSheet
bottom-sheet footer for the home-indicator band.
Raise the cycle and illness day-log severity controls to a 44px touch
floor, add hit-slop to the custom-symptom remove button, give the shared
day-log chips a min height, and let the mood-tag action cluster wrap so a
360px row no longer clips the tag name.
The 60s poller used to unregister service workers, wipe CacheStorage,
and call location.reload() the moment it saw a version bump, with no
prompt — firing mid-form, mid-message, or mid-wizard and discarding
whatever the user had typed. Keep detecting the new build, but now show
a non-blocking, dismissible toast with a Reload action; the eviction and
reload run only when the user chooses to reload. Show the toast once per
detected version.
Wrap the shared HealthChart plus the mood, medication-compliance, and
scatter-correlation charts in role="img" with a concise aria-label that
states the metric, value range, latest reading, and trend direction, so
VoiceOver and TalkBack announce something for every chart cell and the
~50 insight pages behind the shared component.
Add the same role="img" summary to the mood and compliance heatmaps,
route their tooltip dates through the locale- and timezone-aware
formatter instead of a hardcoded German format, and clamp the pinned
tooltip to the viewport so it no longer overflows the right edge on a
narrow phone.
Chart and heatmap text-alternative summaries, the version-update
toast, the touch-help trigger label, list load-error messages, and
the target-editor validation hints — across all six locales.
The confirm:false preview echoed the would-be record without running the
type / range / instant validation, so a preview could show a value the
commit then refused. Run the same shared validation in the preview and
surface wouldFail with the error + reason the commit would return, so a
preview and its commit never disagree.
Also note at the REST read-method allowlist that the MCP-audience guard
assumes GET/HEAD/OPTIONS are side-effect-free — a future side-effecting
GET must revisit that audience binding.
Add three grounded read tools, each a thin façade over an existing
server-authoritative engine (no new analytics, userId from the session):
- get_medication_schedule: next-due / overdue per active medication via
the canonical recurrence engine (computeDisplayDue), with as-needed
flagged and no scheduled due time.
- get_integration_status: per-provider sync health (connected, last sync,
reauth-required/failing) from the integration-status ledger, so a user
can see why their data is stale. No secrets or tokens.
- get_preventive_care: the user's already-configured Vorsorge reminders
with their persisted next-due dates and an overdue flag; it surfaces
configured reminders only and invents no screening logic.
Each declares an output schema, read-only annotations, and {present:false}
on absence. Page the search tool across the full assembled result set
(the prior 200-row scan made later items unreachable even through the
cursor) and give each result a per-item deep link where an id exists.
The doctor-visit summariser was duplicated and had drifted: the
healthlog://report/doctor-visit resource omitted mood, the glucose
panel, the wellness composites, and illness episodes that the
doctor_visit_summary prompt already carried. Extract one shared reducer
(doctor-visit-summary.ts) using the richer superset and have both
surfaces call it, so they ground identically and a future change reaches
both wires at once. Both keep the grounded latest+aggregate shape and
omit absent sections honestly.
The MCP measurement and blood-pressure write tools accepted a
client-supplied measuredAt with no temporal bound and persisted it
verbatim, so a far-future or ancient instant could poison "latest"
reads and rollup buckets. Bound it through the shared entry-instant
validator both manual capture paths use — no future beyond a 5-minute
clock-skew tolerance, nothing before 1900 — inside the write cores, and
return the structured out_of_range refusal rather than throwing.
Extract the type/range/instant checks into shared pre-write validators so
the same verdict can be reused by the confirm-gate preview.
Document the MCP server: an operator guide for enabling the
off-by-default module and connecting Claude.ai / ChatGPT / Claude
Desktop, a capabilities reference for the tool / resource / prompt
catalogue and the grounding + write model, and a skill-author guide for
building connectors against it. Link all three from the docs indexes.
Integration coverage against a real Postgres testcontainer: the
blood-pressure write persists a paired systolic/diastolic row set
atomically under one external id and timestamp, is idempotent on a
replay, and refuses an implausible pair without writing; the explicit
{from,to} comparison filters each side to its span over real rollups.
Bridge the five already-built, grounded Coach retrieval tools the
inventory already advertised but the wire was missing — closing the
advertise-but-missing self-inconsistency — each a thin wrapper over the
existing read with the same gates and present:false contract. Add a
multi-metric fan-out that loops the single-metric read, and a per-analyte
lab-history mode on the labs read. Declare a structured output schema on
every remaining tool so structured content validates for both a miss and
a hit. Page the large reads (search, lab history, multi-metric) behind an
opaque cursor. Return plain-text, citation-ready summaries from fetch
with per-item deep links instead of a JSON dump.
Add an optional explicit {from,to} range to the comparison and
changepoint reads, served over the same rollup tier the five fixed
trailing windows use (reach back to the lower bound, filter to the
span) — so a before-vs-after-a-date question is answerable without
recomputing analytics. Add a per-analyte lab trajectory read (latest to
recent history) over the existing lab resolver, page-bounded so a heavy
panel cannot balloon the response. Introduce an opaque, stable base64url
cursor helper that decodes any malformed or foreign token to the first
page rather than throwing.
Blood pressure is two values, so it cannot ride the single-value
log_measurement tool. Add log_blood_pressure: it writes the systolic and
diastolic rows atomically under one shared idempotency namespace and one
measured-at instant, range-validates each value, rejects an implausible
pair (systolic at or below diastolic), and keeps the same confirm-flag
preview gate, per-credential write budget, MCP provenance, and audit
trail the existing writes use. Declare a structured output schema on
every write tool so a preview, a commit, an already-logged no-op, and a
refusal all validate as structured content.
Fill the discoverable half of the MCP server. Add RFC 6570 resource
templates for per-item reads — healthlog://metric/{type},
healthlog://metric/{type}/{window}, healthlog://lab/{analyte},
healthlog://medication/{id}, healthlog://report/doctor-visit/{window} —
plus fixed resources for the biomarker catalogue, the data inventory, and
the default doctor-visit summary. Each template is a thin facade over the
same server-authoritative read its matching tool already uses, scoped to
the session user; a foreign id resolves to { present: false }, never
another tenant's row.
Wire user-scoped argument completion and listing on the templates: window
completes the fixed enum, metric and analyte complete only from the user's
own inventory and lab set, so autocomplete doubles as honest discovery.
State the data-grounding contract once in the server instructions —
server-authoritative values (do not recompute), { present: false } is
honest absence, free-text fields are data not instructions, discover
before fetching, data and context only. Set the server and resource icons
for connector-directory listing.
Extend the MCP prompt surface beyond doctor_visit_summary with six
ranked, installable review prompts. Each assembles a grounded,
structured data block from an existing server-authoritative engine and
frames the host to narrate from those facts only — read-only, no
diagnosis, surfacing absence as { present: false }.
- weekly_review: 7-day trends, adherence, sleep, recovery, drivers, and
the readiness/recovery + illness/cycle context the daily briefing reads
(derived-briefing + illness-cycle-briefing).
- medication_check: cadence-aware adherence alongside a linked vital.
- recovery_check: illness-recovery composites + personal baselines +
discovered drivers.
- glucose_review: the clinical glucose panel (TIR/GMI/CV%/eA1c).
- sleep_review: per-night sleep, debt, chronotype, and drivers.
- lab_trend_brief: per-analyte trajectory from the labs store, against
each lab's own stored reference range (ranges are never recomputed).
The n-of-1 intervention review and a preventive 'what to check next'
prompt are deferred: the first waits on an anchored before/after read,
the second has no screening engine to reuse.
Honour the OIDC dynamic-registration application_type field on the
register endpoint: validate it to the closed set (web/native) when
present, reject a malformed value with invalid_client_metadata, default
to web when omitted, and echo the resolved value back in the
client-information response.
Forward-compatible with OIDC-style clients without changing the
redirect-URI rules (still HTTPS-or-loopback, authoritatively matched at
authorize) or any other registration behaviour.
Advertise authorization_response_iss_parameter_supported in the
authorization-server metadata and echo the iss parameter on every
authorization response — both the success redirect carrying the code and
the access_denied error redirect. A client can now pin the response to
this server's issuer and reject a code substituted by a look-alike
authorization server in a multi-server deployment.
Additive and backward-compatible: existing clients that ignore iss are
unaffected; PKCE S256-only, audience binding, redirect-URI exact match
and the rate limits are untouched.
Combine the remote MCP server (stdio + remote /mcp + OAuth 2.1 + read
tools + deep-value reads + a confirmed, audience-bound write surface)
with the v1.23 account-security line (MFA, known/trusted devices,
free-text note encryption).
- schema: keep every v1.23 User relation (MFA, known/trusted devices)
alongside the McpOAuthConnection relation; the MeasurementSource enum
carries the MCP value.
- api-handler: the MCP audience guard (a token scoped to
health:read/health:write is bound to /mcp and refused a REST write)
and the step-up MFA helpers coexist.
- maintenance queues: register the MCP token-cleanup sweep next to the
note-encryption backfill and key-rotation queues; shift it to 04:00
so the two daily discovery ticks do not collide.
- migrations: renumber the MCP migrations to 0206/0207 so they follow
the v1.23 chain (0199-0205).
- MCP mood writes route the note through the v1.23 at-rest cipher.
The "Require a second factor" policy redirects web users to enrollment but does
not block direct API or native (Bearer) access for an account that has not yet
enrolled — a deliberate lockout-safety choice. Make that explicit: spell it out
in the admin toggle description, add a code comment at the toggle, and add an
operator runbook (docs/ops/mfa-enforcement.md) describing the boundary.
The "safe to drop legacy key" badge hardcoded bg-emerald-600/text-white. Swap it
for the success token (border-success/bg-success/text-success) the rest of the
settings surfaces use, so it tracks the theme.
Trusted devices was mounted on both the Security section and the Privacy
dashboard. Keep it only on the Privacy dashboard, which already gathers sessions
and security activity, and drop it from the Security section. While here: the
Retention card painted an empty list while the privacy summary was loading and
on error — give it a skeleton placeholder and a one-line fallback, and align the
"What's stored" indentation with the other privacy cards.
The inline rename save/cancel controls on the security-key and passkey lists and
the passkey-nudge dismiss button sat at 32px, below the 44px touch minimum. Give
them the same min-h-11/min-w-11 phone target that collapses to 32px from sm up.
Add an xs (30rem) breakpoint and stack the recovery-code grid one-up below it so
the 11-character codes no longer crowd at 320px.
Regenerating recovery codes invalidates the current set immediately, so the
action now sits behind a confirmation dialog matching the disable-MFA pattern,
with a consequence line spelling out that the existing codes stop working. While
in the card: give the disable-step TOTP input the same digit-strip + 6-char cap
and exact-length submit guard as setup, swap the static disable-code id for a
useId value, and lift the copy-secret button to a 44px touch target on phones.
Mirror the TOTP login path: both the security-key verify and verify-options
routes now reject any challenge whose kind is not "login" with the same generic
401, so a future step-up or non-login challenge can never mint a login session
through these endpoints. Set the registration and authentication ceremonies to
userVerification:"required" to match the server's UV enforcement — a key that
passes the browser but skips UV would otherwise fail server-side.
The encrypted-export route hand-rolled its step-up gate off totpConfirmedAt
alone, so an account whose only second factor is a WebAuthn security key
exported the full record on a plain session — no fresh-factor proof. Route it
through requireFreshMfaIfEnrolled, which covers both the TOTP and security-key
cohorts and lets single-factor accounts through unchanged. Cover the
security-key-only path in the unit suite.
Bring the scattered data-sovereignty controls into one /settings/privacy pane.
It assembles already-shipped pieces rather than rebuilding them: the encryption-
at-rest summary and configured retention windows (read from the server so the
copy stays truthful), links to the export surfaces (including the passphrase-
encrypted option) and the account-deletion / data-reset surface, the active-
session list, trusted devices, and the security-activity log. The backup↔
deletion lag is disclosed honestly — deletion is immediate, but encrypted
off-host backups still hold the data until they roll off retention.
Removing the second factor is a security-state change, so it should not leave
stale credentials behind. The disable endpoint now drops every other web
session, every native refresh token, and every trusted device, keeping only the
caller's current session. The lockout-recovery CLI does the same out of band,
clearing all sessions and trusted devices for the account.
A passwordless passkey login already clears user verification and is a
phishing-resistant possession factor, but it never stamped the session, so a
passkey user could not satisfy an MFA-enforcement policy or pass a step-up
re-check. Stamp Session.mfaVerifiedAt on a primary passkey login so it counts
as a completed second factor — equivalent to a password+TOTP sign-in. The
password+TOTP path is unchanged.
Wire the previously-unused mfaEnforced flag into a real policy. An admin can
require a second factor instance-wide (an AppSettings toggle in the General
section) or per user (a toggle in user management); the effective requirement is
the OR of the two. When required and the account has no active factor — no
confirmed TOTP, no registered security key, and no primary passkey — the user is
walled behind a forced-enrollment interstitial after sign-in, with only the
security settings and sign-out reachable until a factor is enrolled.
The gate is surfaced like the onboarding gate: a hint cookie the proxy reads
without a DB round-trip, recomputed from the database on every sign-in and on
/api/auth/me so a tampered cookie self-corrects. Nobody is locked out — the
enrollment surface stays reachable, and toggling the policy off (or the
disable-mfa CLI) is the operator's recovery path.
Add an opt-in "remember this device" at the second-factor step. On opt-in
a random token is minted, set in an httpOnly+Secure+SameSite=Strict cookie,
and only its HMAC hash is stored in a new TrustedDevice table. A later sign-in
from the same browser still requires the password (factor 1); a matching,
unexpired trusted device lets the login skip factor 2 only.
The token is never accepted for step-up — a trusted-device login does not stamp
mfaVerifiedAt — so a stolen device cookie cannot reach a destructive action.
Trusted devices are listed and revocable from the security surface, are revoked
wholesale when the second factor is removed or the user signs out everywhere,
and cascade away on account deletion. A recovery-code login never trusts the
browser. Migration 0204 adds the table (and the instance-wide mfa_required
policy column that the enforcement work reads).
Record the very first device a user is ever seen on silently. The
new-device sign-in alert only carries meaning once there is an
established device to contrast against, so an empty device ledger
establishes the baseline without firing a notification; alerts begin
with the second distinct device. Registration's silent first-device
behaviour is unchanged.
Also extend the destructive-action step-up to treat a registered
security key as an active second factor, matching requireFreshMfa, so a
key-only account cannot delete or reset without a fresh factor.
Stamp each passkey's last-used time on a successful assertion and return
it from the list, add a rename endpoint, and add a dismissible
add-a-passkey nudge for accounts that have none. Share the relying-party
configuration with the second-factor path.
When a password is accepted for an account with a second factor, the
login response carries a single-use ticket instead of a session. The
login page now renders a second step: a TOTP or recovery code, or a
WebAuthn security-key assertion scoped to the account's registered keys.
A WebAuthn-only account (no TOTP) is recognised at the password step and
offered the security-key path.
The security-key verify endpoints validate the open ticket, verify the
assertion against the user's second-factor credentials, and only then
mint the session through the shared login tail, so no session exists
before the factor passes. The login page also arms passkey conditional-UI
autofill on the username field.
Introduce a dedicated /settings/security section that gathers every
account-protection control in one place: TOTP enrolment (QR plus manual
secret, confirm, recovery codes shown once with copy and download),
WebAuthn security keys as a second factor (register, list with last-used,
rename, step-up-gated removal), recovery-code count and regeneration, and
primary passkey management.
The second-factor WebAuthn path keeps its own credential store, separate
from primary passkeys, and uses non-resident keys so an account can hold
many. Relying-party resolution is shared with the passkey path. A status
endpoint backs the hub; the step-up gate now recognises a registered
security key as an active second factor so a WebAuthn-only account can
still reach the gated actions.
Drop the in-app bug-report surface end to end: the user submission page,
the feedback intake and admin inbox, the optional GitHub-issue
escalation, the operator config, and the dedicated achievement badges.
- delete the /bugreport page and the bugreport / feedback / admin-feedback API routes
- drop the Feedback model, its enums, and the AppSettings GitHub-issue and enable columns (migration 0205)
- remove the admin Feedback section, the Bug Report integration card, and the system-status row
- drop the bug-report nav entry and the app-settings provider that only carried its flag
- remove the bug-hunter achievement badges and their metric
- prune the related i18n keys across every locale and the encrypted-column and key-rotation registries
Add the user-facing account-security surface in Settings:
- Active sessions (closes#64): GET /api/auth/me/sessions lists each browser
login with a coarse device label, masked IP, resolved location, the sliding
last-active stamp, and the current-session marker. DELETE the collection
signs out everywhere else (keeps the current session, revokes other sessions
and native refresh tokens, leaves API tokens alone); DELETE one revokes a
single session, scoped to the owner so a foreign id 404s.
- A sliding last-active stamp on the session, written throttled and
fire-and-forget off the request-resolution path.
- Security activity: GET /api/auth/me/security-activity, the shared read over
the audit log filtered to sign-ins, second-factor changes, password changes,
session revocations, exports, and deletions, with masked IPs.
OpenAPI regenerated for the new endpoints.
On a completed sign-in (password, second-factor, or passkey) from a device or
location not seen before, send one notification through the dispatcher cascade,
honouring the user's per-channel preferences via a new SECURITY_ALERT event
type. Dedupe on a one-way, per-user fingerprint: a SHA-256 over the userId, the
coarse User-Agent family, and a country/ASN-level geo signal, stored in the new
user_known_devices ledger (migration 0203). The first sighting of a fingerprint
alerts and records; later sightings only bump last-seen and stay silent. The
raw User-Agent and IP are never persisted — only the salted hash and a coarse
label. Registration records its device silently so a fresh account does not
alert on its own first session. The whole path is fire-and-forget and never
blocks or fails a sign-in.
Check a newly chosen password against HaveIBeenPwned through the k-anonymity
range API: only the first five hex characters of the SHA-1 digest leave the
host, the suffix is matched locally, and Add-Padding hides the bucket size.
A confirmed match is rejected so a fresh credential is never a known-breached
one; the check fails open, so an unreachable HIBP never blocks the operation.
Covers registration, password change, and admin reset. Existing passwords are
never retroactively blocked at login.
Also lands the six-locale strings for the password-breach notice and the
account-security surfaces that follow.
Account deletion and the keep-account data reset now demand a recent
second-factor re-verification for accounts that have a second factor
active; single-factor accounts are unaffected and keep the typed-
confirmation-only contract. The gate runs through requireFreshMfaIfEnrolled,
which is cookie-only by construction — a Bearer transport for an MFA-enrolled
account cannot satisfy step-up and surfaces the auth.stepup.required envelope
so the UI can launch a re-verification flow.
Add an optional admin-triggered key rotation on pg-boss that re-encrypts
the whole encrypted-column corpus to the active key, reusing the same
column registry as the command-line tool (which stays the canonical path).
POST /api/admin/encryption/rotate enqueues a singleton run and is gated by
cookie-only admin plus a fresh second factor. The run is idempotent
(rows already on the active key are skipped), active-key-only, and never
adds or drops a key — the env key map stays operator-owned, so a key drop
remains a deliberate redeploy. It fails closed per row and reports totals
into the audit trail the status view reads.
Add a read-only admin Encryption section that surfaces what the rotation
tooling computes: per encrypted-column rows-per-key-id, legacy-key row
counts, and overall rotation progress, plus a 'safe to drop the legacy
key' state once every column sits on the active key. GET
/api/admin/encryption/status drives it and returns only key ids and
counts — never key material. A shared scan over the canonical encrypted-
column registry backs both this view and the rotation run. Cookie-only
admin.
Add an opt-in to download the full backup as an encrypted archive instead
of a plaintext JSON file. The user supplies a passphrase over the request;
the server derives a 256-bit key with Argon2id and seals the export bytes
with AES-256-GCM into a self-describing HLX1 archive (magic, version, KDF
params, salt, iv, tag, ciphertext), mirroring the off-host backup envelope.
POST /api/export/encrypted returns the binary archive and is step-up gated
for accounts with a second factor; single-factor accounts use a normal
session. The passphrase is never stored and never logged, and there is no
server-side recovery — the UI states this. A standalone decryptor
(scripts/decrypt-export.ts) opens the archive with the passphrase outside
the app. The plaintext full-backup route now shares one payload builder so
both variants emit the same restore-compatible shape.
Reject any challenge that is not a login challenge before minting a session,
cap the disable and confirm request bodies through safeJson, and correct the
recovery-code scan comment to match its early-return-on-match behaviour.
When a password is accepted for an account with a confirmed TOTP factor, the
login route no longer mints a session or token. It creates a single-use,
~5-minute MfaChallenge (the opaque ticket is stored only as a hash) and
returns HTTP 200 with data:null, error:null and meta.mfaRequired + mfaTicket +
methods. The client completes the login at POST /api/auth/mfa/verify with the
ticket + a TOTP or recovery code, receiving the exact same token bundle the
password path issues, with the session stamped second-factor-verified.
- The ticket is claimed once, atomically, only after the factor verifies, so
two concurrent submissions can mint at most one session; no session ever
exists before factor two.
- Wrong codes increment the ticket attempt counter and burn it at the cap
(throttle, not an account lock); the anonymous verify surface is also
rate-limited per IP.
- meta.mfaRequired is returned only after a valid password and the
invalid-credentials response is identical regardless of MFA state, so an
attacker without the password cannot enumerate MFA.
- The session/token tail is extracted to login-response.ts and shared by both
the password path and the verify path, keeping the transport contract
identical. OpenAPI registers the verify endpoint, the MFA management
endpoints, and the mfaRequired login variant.
Add the TOTP enroll/confirm/disable surface and Argon2id-hashed recovery
codes as the v1.23 second factor.
- totp.ts wraps otpauth (zero-dep) pinned to the interop floor (SHA-1, 6
digits, 30 s) with a +/-1 drift window. Replay protection persists the last
accepted time-step per account (User.totpLastStep, migration
0202_v123_totp_laststep) and rejects any step that is not strictly newer,
so a captured code dies the instant it is accepted once.
- recovery-codes.ts issues ten single-use, high-entropy codes, Argon2id-
hashed at rest, shown once; regeneration replaces the entire prior set in a
transaction and verification burns the matched code with a guarded update.
- The TOTP secret is stored AES-256-GCM encrypted (already registered in the
encrypted-columns registry + rotation script); decrypt is fail-closed.
- Routes: setup (pending secret), confirm (activate + issue recovery codes),
disable (step-up + current factor), recovery-codes/regenerate (step-up).
Setup/confirm are cookie-only; disable/regenerate add requireFreshMfa.
- scripts/disable-mfa.ts is the operator escape hatch for a locked-out
self-hoster, audited, run via pnpm dlx tsx.
Add requireFreshMfa(maxAgeSeconds) and requireCookieAuth alongside
requireAdmin. Both resolve through getSession() only, so a Bearer token can
never satisfy them — the same structural boundary requireAdmin relies on, not
a softenable runtime check. requireFreshMfa passes only for a cookie session
whose Session.mfaVerifiedAt is within the window and whose user has an active
second factor; otherwise it throws StepUpRequiredError (401 +
meta.errorCode "auth.stepup.required") so the client can launch a
re-verification flow.
createSession now accepts an optional mfaVerifiedAt stamp so a session minted
behind a completed second factor records its freshness.
Wrap GET /api/measurements/:id through shapeMeasurementNotes like the list,
create, and update paths already do, so the encrypted note column is decrypted
into notes and the raw ciphertext is stripped from the response. Harden the
rotation-coverage guard to require both the field literal and the model
delegate, so a same-named encrypted column whose table is not rotated is caught.
Lay the account-security and data-sovereignty substrate for v1.23. No
user-facing behaviour yet — schema, crypto registry, redaction, and the
note-encryption data path only.
Schema + migrations:
- 0199 — User totp_secret_encrypted / totp_confirmed_at / mfa_enforced /
passkey_upgrade_nudge_dismissed; Session mfa_verified_at / last_active_at;
Passkey last_used_at. All additive, nullable / defaulted.
- 0200 — mfa_challenges (single-use step-up ticket; only the ticket hash is
stored, so no session or token is ever minted before the second factor),
mfa_recovery_codes (Argon2id at rest), webauthn_mfa_credentials (second-factor
keys, kept separate from passkeys, which remain the passwordless-primary set).
- 0201 — mood_entries.note_encrypted + measurements.notes_encrypted (BYTEA),
matching the existing free-text encrypted-note convention.
Free-text health notes:
- New writes encrypt the mood / measurement note into the *_encrypted column and
null the legacy plaintext column; reads prefer the ciphertext and fall back to
plaintext only for not-yet-migrated rows, via a single read/write helper.
- A boot-time job migrates existing plaintext rows in an idempotent,
transactional, fail-closed pass (encrypt, then null the plaintext, per row in
one transaction; safe to re-run; a row with content is never left both-null).
- The legacy plaintext columns are kept this release and dropped in a follow-up
once the backfill reports zero remaining rows — the same boundary the
key-rotation playbook uses before dropping a legacy key.
Crypto coverage:
- Register the three new encrypted columns and wire them into the key-rotation
script so a rotation never orphans them.
- Extend the observability and log redaction denylists so TOTP secrets, MFA /
recovery / backup codes, and export passphrases never reach a wide event.
Standardise the labs marker subtitle to text-xs across both the compact list
and the card grid, and build the localized name->slug lookup map once per
locale instead of re-scanning the catalog seeds per group on every render.
Add relative to the mobile section strips in the settings and admin shells so
the active chip's offsetLeft is measured from the strip, giving an exact
horizontal scroll target.
Drop the orphaned insights.sleep.headlineCaption and headlineCaptionSuffix
keys from all six locales; nothing references them after the sleep headline
removal.
Hide the copy button on insecure-context (plain-HTTP) self-hosts where the
Clipboard API is undefined: a new SSR-safe useClipboardSupported() detector
mirrors the speech-synthesis hook so the button is absent rather than
error-toasting on tap.
Stop the per-message action row compressing below the 44px touch floor on
narrow phones: the icon buttons keep their tap size (shrink-0 + min-w-11) and
the row wraps instead of shrinking the up-to-six targets to ~30px.
Surface an icon-only action row below each message, sharing the row with
the hover timestamp: muted until the bubble is hovered or focused on
pointer devices, and always visible on touch where there is no hover.
User turns get Copy; assistant turns get Copy, Read-aloud, a thumbs-up /
thumbs-down feedback pair (wired to the existing per-message feedback
endpoint), and Try-again. Copy strips inline chart tokens from assistant
prose so they are never pasted. Try-again resubmits the user turn that
produced the reply as a fresh turn, leaving the composer untouched.
Read-aloud only set the utterance language and let the browser fall back
to its default voice, which on most systems is a legacy compact engine
that reads robotically. Enumerate the installed voices, rank them for the
active locale (prefer enhanced/premium, Siri, Microsoft Natural/Online,
Google, and an exact lang-region match; penalise known compact engines),
and assign the winner with a natural rate and pitch. The voice list loads
asynchronously, so subscribe to voiceschanged and re-rank lazily; when no
voice beats the legacy default, fall back gracefully to prior behaviour.
The history search rendered as a thin, short strip that read as an
afterthought. Give it the app's standard full-width input height and
padding so it matches the standalone conversations page and looks
intentional.
Shorten the composer placeholder across all locales to a short, inviting
one-liner and clip the placeholder pseudo-element so a longer hint can
never wrap to a second line inside the single-row composer on a narrow
phone. The textarea value still wraps and auto-grows for real typing.
Surface a short, factual subtitle under every catalog-linked lab marker on
the Labs page, mirroring the descriptive line the Insights overview carries
under its metric headings. The text states what each biomarker measures,
sourced from the seed catalog via a new labs.catalog.desc.<slug> i18n map in
all six locales.
A reading's analyte resolves back to its catalog slug by a case- and
whitespace-insensitive match against the localised catalog name; free-text
markers with no catalog match carry no subtitle rather than a fabricated one.
Rendered in both the compact list view and the card/tile view.
Fixed-pixel chart bands ate roughly half the screen when a small phone
was rotated to landscape. Swap the brittle heights for viewport-relative
clamps so the chart envelope holds its previous size on desktop and
portrait but shrinks toward a floor on short landscape viewports. Covers
the overview trends-row slot and the mood distribution, weekday, and
time-of-day charts (with matching skeleton heights). Chart series and
styling are untouched; only the container heights flex.
The Sleep view led with an average-per-night headline card that repeated
the same figure and label already shown in the average-sleep tile within
the sleep-rhythm grid row below it. Remove the leading card and let the
tile carry the readout; the stage-distribution bar and duration trend
chart stay in place.
The long single-word labels (de "Benachrichtigungen") clipped their last
character at the tile edge on a 320 px phone. Trim the grid gutters and the
tile padding/gap to hand each label more horizontal room, and cap the bottom
sheet at max-w-md, so the label wraps cleanly onto a second line instead of
running off the tile. Tile height stays uniform.
The Theme sub-trigger inherited the shared sub-trigger padding (py-1.5,
no min-height) while every sibling row carries min-h-11 py-2, so the
Theme row sat a few pixels shorter and the menu read as unevenly spaced.
Match the item height on this instance only, leaving other dropdowns'
sub-triggers untouched.
Let a user opt a connector token into logging. The settings card gains a
write toggle; the mint endpoint accepts a closed scope choice (read, or
read_write) and builds the permission array explicitly — it can mint only
[health:read] or [health:read, health:write], never a wildcard or any other
grant. The OAuth flow grants health:write only when a client requests it, and
the consent screen states in plain language that the app will be able to read
AND log/write health data when write is requested; the token endpoint grants
exactly the approved scopes and never widens. Read-only behaviour is
unchanged. Add the English and German strings for the new toggle and warning.
Register two write tools — log_measurement and log_mood — only for a
health:write-scoped session; a read-only session never sees them advertised.
Each write mirrors the proven Telegram capture path (same range guard,
canonical unit, cache invalidation and rollup recompute) pinned to the MCP
source, and is idempotent on a hashed externalId derived from a client key, so
a retried call is a no-op rather than a duplicate.
Because the remote wire is stateless (no elicitation), the tools carry a
protocol-level confirm gate: without confirm:true they write nothing and
return the exact record they would write plus an instruction to confirm the
value with the user and re-call with the same idempotency key. Commits run
under a tighter per-credential write budget on top of the per-binding request
limiter.
Harden the resource-server audience binding: a token whose scopes are all in
{health:read, health:write} is bound to the MCP surface and refused on every
REST write/delete. A health:write token therefore admits the in-process MCP
write tools only and can never become a general REST write credential; admin
stays unreachable (cookie-only). Refuse, by construction, any medication, lab,
delete/update, or data-export tool on this surface.
Append an MCP value to MeasurementSource (migration 0196, additive enum
extension) so a row logged through the remote MCP endpoint is distinguishable
from MANUAL/TELEGRAM provenance and gets an isolated idempotency namespace
(userId, type, MCP, externalId). The value joins the read enum so existing
read/response shapes decode the rows; the REST batch and single-POST write
surfaces still reject it (server-resolved only). Regenerate the OpenAPI
contract for the additive enum.
Code-review and design follow-ups before the v1.22 cut. No behaviour change
beyond the items below.
Coach / insights:
- Coach SSE route: a non-provider error raised inside the open producer (e.g. a
DB failure persisting the reply) now records on the wide event and forwards to
GlitchTip instead of masquerading as a provider outage; it still surfaces an
error frame since the stream is open.
- Lab ↔ outcome correlations: hedge faint-tier pairs ("a faint hint, if
anything … too small to lean on") so the narrated confidence never outruns the
effect, matching the daily engine.
- Reminder sweep: mint the plan-review reminder and clear the plan's reviewDate
in one transaction so a half-applied write can't duplicate the reminder.
- Fold the two raw provider-fallback runners onto one shared chain walker.
Design / accessibility:
- Lift sub-44px tap targets to the 44px mobile floor (suggested-action card,
Coach bubble affordances), brighten the resting timestamp clock for non-text
contrast, and route the lab chart height through the shared constant.
- Replace ad-hoc arbitrary type sizes with the scale token and static size-11
icon buttons with the named icon-lg variant.
- Settings: add the missing About card-header descriptions, normalise the
low-stock numeric widths, and relax the notification-status rows from nested
cards to a divided list.
Housekeeping:
- Drop three unused exported types (and the now-orphaned source const), re-stamp
v1.21.5 comments to v1.22, and register the live-judge eval harness as a knip
entry so the script-only module resolves.
The AI-settings migration altered "User" but the table is "users"; deploy
would abort with relation "User" does not exist. Correct the identifier and
make both v1.22 column additions rerun-safe with IF NOT EXISTS.
Expose four read-only relational tools and one prompt over the MCP wire,
each a thin re-export of an existing server-authoritative engine — no new
statistics, no new business logic:
- get_correlation: the FDR-controlled, lag-aware association between two
named metrics, selected out of the existing correlation discovery the
Coach and the insight route already run. Descriptive, never causal.
- compare_metric: two metrics over the same trailing window, or one metric
across two trailing windows, computed over the WMY rollup readers with a
delta only when both sides share a unit.
- get_metric_baseline: where today's reading sits against the user's own
usual range (median ± robust deviation) plus the strongest lagged driver,
reusing the metric-page baseline builder. Honest-null below the learning
floor — never a fabricated range.
- detect_changepoints: level shifts over the rollup tier via a small in-repo
CUSUM binary segmentation with a high firing bar.
- doctor_visit_summary prompt: a structured, grounded visit-prep summary
assembled from the doctor-report data path so the numbers match the
exported PDF, with the grounding framing injected once — data and context
only, no diagnosis or verdict.
Every tool returns values, units, reference bands and provenance, declares
read-only / non-destructive / closed-world annotations plus an output schema,
and reports { present: false } for absence. All reads stay session-narrowed
and read-only; the existing security posture is unchanged.
The v1.22 waves added new keys to en.json and de.json only, leaving the
es/fr/it/pl bundles short and failing the locale-integrity guard. Translate
the 70 missing keys per locale across the Coach suggested-action prompts,
conversation seeds, read-aloud controls, derived-metric subtitles, freshness
timestamps, biomarker hide/restore controls, the per-metric insights link, and
the saved-reminders settings panel. ICU placeholders are preserved and the new
keys follow the en.json ordering within their parent objects.
Add three per-user AI settings, all configured in the app (no new env vars),
behind one migration (0198):
- Response timeout (seconds). A per-user upstream timeout threaded onto the
provider call. A local/self-hosted backend can take more than the legacy
60 s default to load a model on the first request, which timed the Coach
out before the first token. Defaults to a generous ~180 s when unset;
validated to 10–600 s.
- True token streaming for the local OpenAI-compatible client. The client now
POSTs with stream:true and parses the SSE delta chunks, emitting tokens as
they arrive, with a per-idle-gap timeout instead of a whole-call ceiling and
a graceful fallback to the buffered path when a server cannot stream. The
Coach route runs the provider call inside the response stream behind a
keepalive heartbeat so a reverse proxy no longer idle-drops the connection
during model load or generation. Grounding, refusal, and sentinel-stripping
still run on the complete reply before any token reaches the client.
- Dedicated document-scan (Lab-OCR) provider. Off by default — the OCR path
uses the main provider chain. When enabled, lab-report processing resolves a
separate provider/model/key, with a hint when the chosen model is not
vision-capable. The new key column is AES-256-GCM at rest and registered in
the encrypted-columns registry and the rotation script.
i18n: EN + DE translations for the new Settings → AI strings (other locales
carry interim copies to keep locale parity green).
Add the medication-class to target-vital map and the adherence-to-vital
storyline so the Coach can surface, association-only, when an adherence dip
lines up with a drift in the vital that medication targets. Never causal,
never a dose-change suggestion; dose questions defer to the prescriber.
Add an in-repo CUSUM / binary-segmentation changepoint detector over DAY
rollups with a high firing bar (persistence plus within-segment-spread
magnitude) so the Coach can name a dated, sustained level shift instead of a
vague "lately", and stays silent when no break clears the bar.
Surface which source recovery is read from when two sources materially
diverge, with the honest reason, instead of silently averaging.
Lay the n-of-1 experiment substrate: a nullable encrypted outcome column on
coach_plans (migration 0197), the review_due / reviewed lifecycle statuses
wired into the encrypted-column registry and rotation script, and a daily
worker that reads the experiment result back in the user's own before/after
numbers — association-only, honest-null, no cheerlead on an adverse trend,
and routing a worsened vital to the clinician. The user-visible read-back
stays behind COACH_EXPERIMENT_VERDICT until its review cases clear.
Extend the outbound dose screen to catch experiment-shaped dose changes
(halve / skip / stop a medication to test it). Add evaluation cases and unit
tests for each new surface.
Harden the remote MCP endpoint and the in-repo OAuth bridge against the
adversarial review of the milestone branch. The two ship-blockers made the
"read-only, revocable" promise false; the rest tightened the surface.
Audience binding at the resource server: a token whose only grant is the new
`health:read` scope is the MCP read credential. The REST edge now accepts it on
safe (read) methods and refuses it on writes/deletes with 403, so a connector
token can read over `/mcp` and GET routes but can never POST/DELETE the record
despite a "read-only" consent. The `/mcp` resolver is unaffected.
Revocable, theft-detectable refresh tokens: refresh artifacts now anchor to a
persistent connection row. Revoking a connection stops every future refresh and
kills its access tokens; a replayed (already-rotated) refresh is treated as
theft and revokes the whole connection. Connections surface in settings as the
unit a user can disconnect. The old access token is revoked on rotation.
Also: fail closed on the OAuth/`/mcp` surface when no canonical origin is
configured and mark the discovery documents non-cacheable; honour the global API
kill-switch on `/mcp`, `/token`, `/authorize`, and `/register`; cap the CIMD and
registration body reads while streaming instead of after; show the verified
client origin on consent; validate CIMD redirect-URI schemes; assert same-origin
on the consent POST; and add a daily prune for expired connector tokens.
New migration 0195 adds the connection anchor and links access tokens to it.
Route the composer seed questions through i18n so a German user no longer
gets an English opener when tapping Ask the Coach from a metric page or an
insight card. Replace the relative timestamp on the briefing and per-metric
cards with a freshness label that shows the time only for today, drops it for
yesterday, and renders a locale date honouring the user's date-format
preference for older entries. Render the briefing paragraph, the per-metric
and score assessments, and the period narrative through the shared paragraph
helper so longer prose shows real paragraphs instead of one run-on block.
Adds the insights.coach.seed.* and insights.updated* keys (English + German).
Feed the user's own usual blood-pressure range — the median plus or minus
MAD band, computed from the readings already in the snapshot — into the Coach
context with a clause to quote those bounds verbatim, so a band is never
invented from window means. Thread the stored first name into the prompt and
surface it sparingly on roughly one turn in three, never as a rote greeting.
Add the context-layer rule that names the relevant context before a
suggestion, a single-chart-token allowance that works on every provider, and
a per-turn opener hint.
Wire the already-shipped suggested-action SSE frame to render: the streaming
hook collects it and the message thread shows the action card, mirroring the
reminder-suggestion path. Extend the eval golden set with band-verbatim,
chart-token, and no-overclaim cases, and pin a red-team case proving the
chart-token syntax cannot smuggle a dose past the outbound screen.
Replace the fixed three-beat assessment skeleton with a menu of moves the
model orders freely, and reframe the daily briefing from per-signal
enumeration to a verdict-first read that weaves the few salient signals into
one associative story. Score cards now name both the helped and the hurt
contributors, translate the band into a forward read for the day, and close
on one grounded nudge; the deterministic fallback gains the same band-meaning
sentence so provider-less and demo accounts read as well.
Add a hash-seeded opener-archetype rotation (deterministic, testable) so
consecutive cards and days no longer open the same way, a day-rotated sampling
seed for the period narrative in place of the fixed reference seed, and a
shared paragraph formatting contract. Add a score-number grounding gate and
extend the briefing grounding allow-set to the glucose, labs, preventive-care
and workout aggregate blocks so real figures can be cited without opening the
gate. Gate the daily score warm on an input fingerprint so a quiet day skips
the model call.
A dedicated Settings card co-locates the off-by-default MCP module toggle
with token management for connecting an external assistant:
- Mint / list / revoke a dedicated `health:read`-scoped token (shown once,
copy to clipboard) via new `/api/mcp/tokens` routes — never the
`medication:ingest` scope, never the wildcard.
- Surface the connector URL and the enable switch in one place, wired to
the existing module-preference endpoint.
- New section wired into the Settings shell next to API & Tokens, with
EN + DE strings (and the other locales) under `settings.mcp` and
`settings.sections.mcp`.
- Add the `search` + `fetch` retrieval pair with the exact ChatGPT wire
shapes (`{id,title,url}[]` and `{id,title,text,url,metadata}`), backed by
the same server-authoritative reads and returning real deep-link URLs into
the web app for citations. Without these the server is invisible in
default-mode ChatGPT.
- Annotate every read tool `readOnlyHint`/`destructiveHint:false`/
`openWorldHint:false`; structured tools also declare an `outputSchema` and
return `structuredContent`. Omitting the hints is a validation error in the
connectors.
- Reject a cross-origin browser request to `/mcp` with 403 (DNS-rebinding
defense) and point the 401 `WWW-Authenticate` at the resource metadata.
Let a remote MCP client (Claude.ai, ChatGPT) self-register and connect to
the `/mcp` endpoint over OAuth 2.1 + PKCE, bridged onto the existing `hlk_`
Bearer model — no parallel identity store.
- Discovery: RFC 9728 Protected Resource Metadata and RFC 8414 Authorization
Server Metadata at `/.well-known/oauth-protected-resource` and
`/.well-known/oauth-authorization-server` (added to the proxy allowlist).
- `/api/mcp/oauth/authorize` — consent gate backed by the existing session;
a code is only issued on an authenticated POST decision, never on a GET.
- `/api/mcp/oauth/token` — authorization-code and refresh-token grants with
mandatory PKCE S256, single-use codes, and rotating refresh tokens. The
access token is an ordinary short-lived `hlk_` token scoped `health:read`.
- `/api/mcp/oauth/register` — RFC 7591 DCR fallback; CIMD (SEP-991) is the
preferred path and resolves client metadata through `safeFetch`.
- Audience binding (RFC 8707): every token is bound to the canonical `/mcp`
resource URI; a mismatched `resource` is rejected at authorize and token.
- The grant artifacts (code, refresh, DCR client id) are HMAC-signed,
self-describing, and domain-separated; the central log redactors scrub
them plus the PKCE verifier.
The local AI provider's private-host opt-in is read from the environment but
was missing from the compose environment whitelist, so the value never reached
the container. Add it (and document it in the env example) so a self-hoster can
point the local provider at an internal Ollama / LM Studio host.
Bridge the existing transport-agnostic MCP server factory to a Next.js
route at /mcp using the SDK's Web-standard Streamable HTTP transport, so a
remote client can connect with a manually-minted hlk_ Bearer token. The
same tool/resource layer the local stdio transport serves is reused
verbatim — no duplicated tool logic.
The route owns four gates in front of the transport:
- Auth: Authorization: Bearer hlk_… resolved by the same validator the
HTTP edge uses, so revoked/expired/scope semantics never drift. A cookie
is never consulted, so the cookie-only admin boundary stays unreachable
here regardless of token scope.
- Rate limit: the Postgres limiter keyed by the <userId>:<tokenId> binding
(mcp:<binding>), so a single shared credential cannot drain the account.
- Module gate: a new opt-in mcp module, OFF by default. With it off the
endpoint answers 404 and reveals nothing — a self-hoster who never opts
in gets zero new exposure. The Modules settings hub renders the toggle.
- Transport: a fresh stateless server + transport per request.
Read-only by construction (only read tools are registered) and same-origin
by construction (no CORS headers). The proxy lets /mcp skip the page cookie
redirect since it authenticates by header, mirroring the webhook surfaces.
Add unit coverage for the closed when-grammar and `---REMEMBER---` capture
(propose-free active write, recall-only note, malformed drop, per-user cap),
the closed action-card allowlist (off-list and bad-interval rejection, no
auto-apply), the daily sweep (due flip, plan-review mint, undecryptable
skip), and the routes (owner-scoped no-IDOR 404, field-by-field build, the
off-allowlist 422, and the module gate).
Add a "Your reminders" panel under Settings → Coach that lists the stored
reminders, highlights the due ones, and gives the user view, confirm,
done/dismiss, and delete control over their own memory. Light the Coach
launcher dot when a reminder is due, alongside the existing nudge signal.
Add the generic confirm-card component for the suggested-action frame: a
plain-text preview with a localized per-action heading and a single confirm
tap that posts the validated params to the build endpoint.
Translate the new copy in English and German.
Add the management surface for the reminder memory: list, create, confirm,
done/dismiss, and soft-delete, every query scoped to the session owner with
the same no-IDOR posture as the plans routes (a cross-user or unknown id is
a 0-count no-op, never a leaked existence). The note is the only writable
field; timing is resolved server-side from the closed grammar.
Add the confirm endpoint behind the generalised action card. It switches on
a closed action-type allowlist and builds each entity field-by-field from a
server-resolved spec: a preventive-care reminder whose schedule comes from a
closed interval catalog, or a reminder note. It never auto-applies, and the
allowlist excludes any medication or clinical change.
Regenerate the OpenAPI contract for the new routes.
Capture the casual "remember this / remind me about X" utterance the Coach
used to drop. The model emits a provider-neutral `---REMEMBER---` sentinel
that the chat route strips inline on EVERY turn — so a short conversation,
where the request usually happens, no longer waits for the long-chat memory
worker. The note is the user's own explicit request, so it is stored
silently as active; a validated closed grammar (ISO date, relative +Nd/+Nw,
or a context cue) sets the trigger, and a malformed block is dropped.
Surface the stored reminders two ways. A snapshot recall sub-block injects
near, overdue, and already-surfaced reminders so the Coach references them
unprompted next turn. A daily sweep flips overdue reminders to due and
mints a one-off review reminder from a plan whose review date has passed,
finally activating the plan review-date column. In-app surfacing only here;
push delivery follows later.
Also generalise the propose-then-confirm action card: a closed
`---SUGGEST-ACTION---` allowlist (a preventive-care item or a reminder note)
rides an additive SSE frame. Nothing is created until the user confirms.
Introduce a third Coach memory type alongside facts (durable traits) and
plans (standing if-then habits): a time-anchored "bring this back later"
reminder with a due moment and a done state. The free-text note is
AES-256-GCM ciphertext via the shared Bytes codec; metric, status, the
trigger fields and the dates stay plain so the sweep and the injection
picker rank and filter without a per-row decrypt.
Migration 0196 is purely additive (status and trigger_kind are app-side
closed enums, not DB enums). Register the new note_encrypted column in the
encrypted-columns registry and the rotation script so a key rotation can
never silently skip it.
Thread the rolling early-detection pass and the labs -> outcome pass through the
correlations route and the coach correlations reader, both of which already
build the same daily series. The route response gains an optional `emerging`
block (recent-window pairs not yet established, flagged provisional) and a
`labCorrelations` block; the coach reader gains matching `emerging` and
`labDrivers` fields. Regenerate the OpenAPI contract for the new response shape.
The daily briefing's always-on feature payload carried only the core vitals,
mood, sleep, steps and medication compliance, leaving four captured domains
invisible to it. Add four blocks to the aggregated features:
- Labs: recent flagged biomarkers (out-of-range or trending vs the prior
reading). Retired markers are excluded; qualitative results stay neutral with
no fabricated verdict.
- Preventive care: due and overdue items off the server-authoritative next-due
date, so an overdue screening can prompt an early nudge.
- Workouts: a provider-agnostic count, duration and distance aggregate over the
trailing windows (the briefing previously saw activity only as step counts).
- Glucose: the present-and-trend block the other vitals carry, where glucose
previously reached the briefing only as a single signal-of-the-day row.
The three extra reads are scoped to the wide briefing window; the tight
snapshot window that already carries its own context skips them. Every block is
omitted when its domain is empty rather than emitting zeros, and user-entered
labels are sanitised before they reach the prompt.
Add a rolling 14-21 day early-detection pass to the FDR correlation engine
so an emerging shift (illness pulling HRV and sleep down together) surfaces
before the 180-day retrospective scan accrues the n>=20 it needs. The recent
pass re-runs the same engine over the trailing window with a lower paired-day
floor, a tighter Benjamini-Hochberg target, and a no-faint effect bar, then
drops any pair the retrospective scan already established so a pattern is never
reported as both established and emerging.
Extend the curated behaviour x outcome matrix with three clinically meaningful
links: medication adherence -> next-day blood pressure (systolic and diastolic,
the latter previously never a candidate channel), sleep as a behaviour
(poor sleep -> next-day mood / HRV), and a separate point-vs-window labs ->
outcome pass (each draw against the contemporaneous outcome mean) since sparse
lab draws cannot meet a daily-lag floor. Every pass keeps FDR control, the
effect-size floor, and shrinkage; thin data degrades to absent, never a
fabricated link. All associations stay descriptive, never causal.
The bottom-right launcher opened a blank new chat on an unread proactive
message instead of the thread the Coach wrote. Return the owning
conversation id from the nudge-status endpoint, have the launcher open
/coach?c=<id> on the unread path, and stamp the same deep link on the push
metadata so the notification lands in the exact conversation rather than
relying on a most-recent heuristic.
Activate the dormant chart-token render path for the Coach: an allowlisted
metric:<TYPE> token in a settled assistant turn mounts the self-fetching
HealthChart for the user's own series, gated on the turn's grounded
provenance so a metric the reply never saw never renders and no data is
fabricated. Add a read-aloud toggle backed by the browser Speech Synthesis
API (no dependency, no egress) and a hover/tap timestamp tooltip on each
message bubble. Render message text through the shared paragraph helper.
Add a shared, pure ProseBlocks helper that splits trusted plain text on
blank lines into <p> blocks and single newlines into <br>, with no markdown
library and no dangerouslySetInnerHTML. A catalog-whitelisted linkifier turns
grounded /learn/<slug> references into safe anchors via learn-more-link;
unknown slugs stay plain text. Rewrite StreamedProse to be streaming-safe:
completed paragraphs settle while only the growing tail animates word by
word.
Introduce a read-only Model Context Protocol surface built on the official
@modelcontextprotocol/sdk so a user can query their own HealthLog data from an
MCP-capable assistant over local stdio with a pasted, scoped Bearer token — no
extra infrastructure.
The tool and resource layer in src/lib/mcp is transport-agnostic: five read
tools — list_metrics, get_metric_series, get_medication_compliance, get_labs,
get_correlations — each a thin wrapper over the existing server-authoritative
retrieval paths, returning structured values, units, reference bands, and
provenance with an explicit { present: false } for absence and no prose
verdicts. Two resources expose the user's profile (data-minimised) and
medication list. A stdio entrypoint (src/cli/mcp-stdio.ts, pnpm mcp:stdio)
resolves the token through the shared Bearer path and binds the session to
<user>:<token>; with no token it refuses to start.
Read-only and admin-unreachable by construction: only read tools are
registered, and requireAdmin stays cookie-only so the Bearer/MCP wire can never
elevate. Covered by unit tests for every tool and resource, the auth/scope
model, and the cookie-only admin boundary. The server ships in the existing
container; docker compose up is unchanged.
Move the canonical Bearer-token validation out of the api-handler request
wrapper into src/lib/auth/bearer.ts so it can run outside a Next.js request
scope without dragging next/server, next/headers, or the Wide-Event machinery
into the import graph. requireAuth's Bearer path now delegates to it and keeps
the existing auth.bearer.* audit trail and Wide-Event annotation, so the
revoked / expired / scope / lastUsedAt semantics live in one place and cannot
drift between callers.
Covered by a focused unit test pinning the hash-only lookup and the
revoked / expired / insufficient-permission rejections; the existing
require-auth Bearer suite stays green.
Clicking a type on the measurements list now drills into that metric's
insights sub-page when one exists, falling back to the existing
?type= filtered list when it does not. A reverse type-to-slug map backs
the navigation, preferring the dedicated single-metric page so an
ambiguous type resolves correctly (PULSE to pulse, not blood-pressure;
WEIGHT to weight, not bmi). No API change.
resolveInsightsLayout merged any default tile a saved layout did not
enumerate as hidden, so a late-added tile such as steps (introduced in
v1.12) was force-hidden on every read of an existing or iOS-synced
layout and its nav pill never appeared, even with step data present.
Appended tiles now inherit the default visibility instead, matching the
fresh-account default; the per-tile data floor still keeps an empty
metric out of the nav.
Detail pages already render a one-line description under their heading;
the overview section headings had no slot for one. Add an optional
subtitle to SectionHeading, rendered in the same muted typography, and
wire it into the obvious top-level sections: Trends (folding its existing
right-aligned caption into the slot), Your health scores, and Your
vitals.
The metric sub-page heading carried a trailing info "i" that opened a
reference-guidance tooltip on every routed insights page, and four card
satellites repeated the pattern (mood statistical caveat, health-score
delta read, period-narrative provenance, derived-score provenance). Drop
the glyph and its disclosure everywhere. Where the disclosure held the
only copy of a descriptive line, that line now reads inline as a muted
caption so the information stays visible instead of hiding behind a hover
affordance.
Replace the hand-rolled native scan dropzone button with the shared
Button primitive, swap the inline-style chart height for a height
class, and remove the Learn link from the biomarker value view where it
does not belong.
Rebuild the biomarker manager as compact rows (name + unit inline,
reference range / panel on a muted sub-line) with a hide toggle, an
edit button, and a delete, replacing the one-card-per-marker layout.
Hidden markers collect in a separate restorable section. Filter hidden
markers out of the lab-entry picker, keeping the currently selected one
visible so an existing reading still resolves its label.
Touch targets meet the 44px mobile floor (size-11 sm:size-9).
Add a `hidden` boolean to the Biomarker model (migration
0195_biomarker_hidden) so a marker the user no longer needs can be
soft-removed instead of deleted: it leaves the active catalog and the
lab-entry pickers but keeps its readings and canonical unit/range.
Thread `hidden` through the DTO, the update schema, the PUT handler
(field-by-field), and the list endpoint (visible markers first). Regen
the OpenAPI contract.
Move the preventive-care surface from the German-derived slug /vorsorge to
the English /checkups and make the feature read consistently across locales.
The route directory, the nav entry, the dashboard card link, and the
onboarding tour step now point at /checkups. A permanent redirect keeps old
bookmarks, the PWA's cached navigation, and push deep-links resolving.
Display label is now "Checkups" in English and the equivalent term in
Spanish (Revisiones), French (Examens), Italian (Controlli), and Polish
(Badania); German keeps "Vorsorge". The internal id stays "vorsorge" so
saved layouts, module-order preferences, and tour state keep resolving
without a data migration.
The Layout hub links, the global injection-site exclusion card, and the
mobile passkey list hand-rolled their own card surfaces with rounded-lg
or sm:p-5 padding, drifting from the SettingsCard contract
(rounded-xl, p-4 md:p-6). Wrap them in SettingsCard so every settings
surface paints the same radius, border, and padding. Content and
behaviour are unchanged.
SettingsSectionFrame re-rendered the page heading and subtitle that
SettingsShell already paints from the section slug. No route used it;
only the per-section smoke tests rendered bodies through it to assert the
visible heading and its id.
Remove the component and move the heading wrapper those tests rely on
into a test-only harness that reproduces the shell's heading markup, so
the smoke tests keep asserting what the user actually sees.
The desktop section nav in both shells used a single sticky wrapper at
top-20. Because the nav can run taller than the scroll viewport and the
top offset assumed a topbar that sits outside the scroll region, the
panel drifted with the content and re-pinned abruptly on scroll-up.
Cap the wrapper height to the viewport and let it scroll independently
(top-6, max-h-[calc(100dvh-5.5rem)], overflow-y-auto, no-scrollbar) so
the nav stays pinned. Covers every /settings/* and /admin/* page through
the shared shells.
The comprehensive insight that feeds both the daily briefing and the
insights trend narrative is a single reasoning-heavy generation over the
full feature snapshot. On a large account it runs past the provider
client's fixed 60 s timeout and is aborted mid-stream, so the chain
reports an all-failed outcome, no cached block is written, and the
surface falls back to its static placeholder on every visit. The warm
paths bounded the same generation tighter still (45 s) and counted the
failure with no observable signal, so the blank state was permanent and
silent.
Make the per-request upstream timeout configurable
(CompletionParams.timeoutMs, default 60 s for every existing caller) and
raise it to 120 s for the comprehensive surface only, across all three
generation paths: the on-demand regenerate, the on-demand warm, and the
nightly pre-generation. Raise the warm-pass budget to 130 s so the
provider's own timeout governs a genuine completion rather than clipping
it early. Annotate the comprehensive failure in both the generator and
the nightly loop so a generation that keeps failing is queryable instead
of leaving the surface empty without a trace.
The dedicated conversation-history page gates on the operator master
flag and the per-user opt-out like the main Coach route, so add it to
both grep-based gate-site allowlists.
Cover the new composer settings link, the seeded-opener dismiss control,
and the recency group headers (today, yesterday, this week, earlier) on
the dedicated conversation-history page, across all six locales.
Add a Settings entry to the composer's "+" actions menu and remove the
page-toolbar gear; the drawer header keeps its own gear. Let the seeded
"worth a look" opener fill the composer column and give it a per-day
dismiss control. Route the Conversations action to /coach/conversations
and drop the in-page history drawer.
Replace the left slide-in conversations drawer with a standalone
/coach/conversations page: a search field above a recency-grouped list
(Today / Yesterday / This week / Earlier) built on the existing
conversations query. Selecting a row routes to /coach?c=<id>, reusing the
deep-link open mechanism — no new API, no new state. The drawer's
Conversations affordance now navigates to the new route, and the Coach
page drops its ?view=conversations in-page drawer handling.
Mirror the about-me management routes: the PATCH (confirm/update) and DELETE
on /api/coach/plans/[id] now apply a per-user mutation limit, closing the
consistency gap with the rest of the Coach management stack. Owner-scoped and
cheap, so the limit only caps a runaway client loop.
Adds the Coach goal / if-then-plan memory model (propose-then-confirm,
encrypted, migration 0194) and a standing Coach quality harness (graded
golden set + red-team in CI, optional nightly model judge off by default).
Caches the dashboard snapshot per language. No breaking changes.
Drop the unused extractCoachClaims helper (+ its CoachClaim types) and the
unused CoachPlanStatus type — the deterministic grader uses the boolean claim
helpers directly, and the plan status enum is consumed via the const array.
Update the dashboard-snapshot suite for the locale-keyed cell: stub the i18n
context and add getQueriesData to the fake query client so the locale-aware
preload probe is exercised.
Hotfix: the Coach and daily briefing failed for ChatGPT-provider accounts —
a provider tool-call rejection surfaced as a 500. Provider failures now
degrade gracefully, the Codex tool-call wire format is corrected, and the
provider error body is captured for diagnosis. Also: the thinking bubble
shows only the animated dots.
The pre-stream thinking bubble dropped the visible "Thinking…" word and now
renders just the three bouncing dots. The label stays in an sr-only span so
the thinking state is still announced to screen readers.
A Codex backend 400 on the richer Coach tool-call request (and the daily
briefing) propagated out of the provider chain runner unwrapped: a 4xx other
than 401/403/429 was not classified as a hard provider failure, so it bypassed
AllProvidersFailedError and rethrew from the chat route as an HTTP 500. Plain
status-card completions, which never send a tools array, were unaffected.
Harden the failure path so a provider error can never surface as a 500:
- Treat any 4xx a provider wire returns as a hard failure so it cascades and is
wrapped in AllProvidersFailedError. InsightSchemaError (the strict-JSON 422
surface) stays the only non-hard class, so that diagnostic is preserved.
- The chat route now also classifies a tagged provider error that reaches its
catch unwrapped and emits the graceful coach.provider.* SSE frame instead of
rethrowing. Only an untagged error keeps the 500 + error-reporter path.
Make the Codex 400 diagnosable: fold the redacted upstream response body into
the thrown error message and the chain wide-event, so the field the backend
rejects is visible in logs (it was previously dropped).
Fix the Codex request shape behind the tool-call 400:
- Send the required strict flag on each Responses-API function tool.
- Replay prior assistant turns as output_text, not input_text, per the
Responses API multi-turn contract — the violation only surfaced on round two
of the tool loop, which is the path that broke.
Codex-only; the OpenAI, Anthropic, local and mock providers are untouched.
A deterministic golden-set + red-team floor that gates every Coach change
offline, plus an opt-in live-judge layer that scores warmth and groundedness
with a pinned model only when COACH_EVAL_API_KEY is set.
- 42 golden cases across grounding, cross-metric, data-honesty, provider-parity
and own-baseline taxonomies, graded by behaviour-anchored binary criteria.
- Structured-claim grounding helpers extend the numeric verifier from numbers
to threshold / own-baseline / confident-verdict claims; the numeric API is
unchanged so the existing grounding tests stay green.
- Red-team battery asserts the existing inbound refusal, outbound dose screen,
acute red-flag escalation and learn-link guards fire.
- Nightly + manual workflow runs the deterministic floor always; the live judge
step runs only when the secret is present and is non-blocking.
Add a CoachPlan model: durable, user-scoped if-then implementation plans
the Coach proposes and the user confirms. The background memory-refresh
pass extracts plans the user agreed to and writes them as "proposed"; only
the user-facing PATCH on /api/coach/plans/[id] activates one. The free-text
cue / action / target are encrypted at rest with the shared Bytes codec;
metric, status and dates stay plain so the injection picker ranks without a
per-row decrypt. Active plans fold into the Coach snapshot's memory block
and survive the prompt-budget drop alongside facts. List, confirm and
soft-delete routes are Coach-gated and narrow the owner from the session.
Key the dashboard-snapshot query cell by locale so a locale switch reads
freshly localised tile copy on its own cell instead of serving the prior
locale's prose until the slot goes stale. The zero-arg key stays
byte-identical, so every prefix-based invalidation keeps reaching the
locale-keyed cells; the offline-banner and preloader match on the prefix.
Migration 0194 (additive CREATE TABLE). Encryption-rotation registry and
script cover the three new Bytes columns.
The A1–A3 source files carried v1.22.0 docstrings from before the work was
re-cut as the v1.21.2 point release. Comment-only sweep so blame and the
in-file version tags read consistently with the shipped version.
The "the Coach is already on …" pill resolved its metric label through the
brand-free English domain vocabulary, so it read in English on every
non-English locale. It now prefers the existing localised measurements.type*
name for each measurement-backed source, keeping the English phrase only for
the gait/audio long tail that rarely becomes a scoped launch. Also rewords the
English return-to-baseline line so the day count reads as the in-range run
length rather than the time since.
Surfaces the Coach's existing intelligence as ambient presence: a per-metric
"Coach read", a visible scope handoff and a seeded opener, briefing callbacks,
a tension verdict and return-to-baseline note on the score, two motivational-
interviewing moves, and grounding parity on the local model path. No
migrations, no breaking changes.
The builder is server-only by construction — only the metric-page route
imports it and it reads Prisma — and no other module in the tree depends on
the `server-only` package, so the import was both redundant and an undeclared
dependency. Removing it keeps the dead-code gate clean without weakening the
boundary.
The metric-page "Coach read" strip and the unscoped-hero seeded opener are
Coach surfaces, so both routes now call requireAssistantSurface("coach")
alongside the insights module gate. With the Coach turned off they return
the same refusal every other Coach route does, instead of lingering as a
"Coach read" line while the assistant is disabled. Restores the coach
route-gate inventory invariant.
The hero score card and the daily briefing accept tension, returnToBand, and
memory props, but nothing populated them server-side, so the lines never rendered.
Resolve all three server-authoritatively on the dashboard snapshot DTO (iOS rides
the same payload):
- tension — fires when the readiness composite's contributors disagree (a strongly
favourable one alongside a strongly unfavourable one), reusing the readiness
components the wellness ring grades and the shared disagreement detector.
Suppressed under a clinical red-flag (a coincident-deviation that no illness
explains) so the red-flag path is never reconciled into a calm verdict.
- returnToBand — runs the streak detector across the salient vitals and surfaces
at most one metric that has come back inside its own personal range after a
genuine prior out-of-band run.
- memory — maps the machine-derived coach memory block to the briefing's recall +
forward-look.
The blocks are locale-agnostic (contributor keys + a metric type); the insights
hero localises them through the existing readiness-contributor and metric-name
labels before handing the cards their resolved strings. Each value omits rather
than guesses. The snapshot cache key now carries the locale for the recall prose.
Add a deterministic narrative layer the Coach narrates rather than invents:
- Briefing callbacks + forward-looks: the daily briefing recalls the prior
period and points ahead, anchored to a figure, one callback per surface.
- Tension Verdict: when the readiness composite's contributors disagree
(good sleep but a rising resting pulse), surface a brief honest internal
read on the health-score card. A real clinical red-flag is never reconciled
away into a calm verdict.
- Streaks + return-to-baseline: a pure detector over a DAY-bucket series
closes a worry once a metric returns inside the personal MAD band after a
genuine out-of-band run; a gap day breaks a streak.
- Coach prompt: add develop-discrepancy, roll-with-resistance, and an
anti-persuasion clause; ratchet PROMPT_VERSION.
The tension flag and streak event are computed server-side; the card lines
render resolved fields and the model narrates them.
Replace the hidden composer prefill with a visible scope affordance: when
the Coach is launched narrowed to a metric, the hero shows a scope pill
naming that metric plus a tappable seed question, and the drawer sources
rail surfaces the same scope line. When the Coach opens unscoped, resolve
today's most notable derived signal server-side and offer it as a tappable
opener; fall back to the neutral greeting when no signal crosses the gate,
so the blank box is never cold and never fabricated.
Add an internal GET /api/insights/coach/seeded-question that reuses the
briefing's derived-signal detector to select the opener server-side; the
client renders the resolved DTO.
Render a compact two-line "Coach read" above each metric sub-page chart:
an own-baseline line ("Your usual range is X–Y; today's Z sits within /
above / below") and one top lagged-association line from discovered
correlations, framed descriptively. It makes a metric read seen-in-context
rather than isolated.
The strip is server-authoritative: a new GET /api/insights/coach-read
resolves the band via the median ± k·MAD baseline engine (with its 7-day
history floor — below it the strip says "still learning your range" rather
than inventing a range) and the strongest on-metric driver from the
existing FDR-controlled, effect-size-floored correlation discovery (no
lowered floor; the line is omitted when nothing clears it). The client
component only renders the resolved DTO, so web and iOS read identical
numbers. DTO shapes and the pure selection logic live in a client-safe
sibling shared by the component and the unit test.
Wired through the HealthKit metric scaffold (every generic metric page) and
the bespoke weight, blood-pressure, and pulse pages.
The post-hoc numeric verifier only graded the tool path, so local /
no-tools providers got no grounding net. Grade the no-tools prose
against the snapshot the model was shown — its structured sections,
which already carry the correlations-snapshot block — and flag any
cited number that traces to no snapshot leaf. Non-blocking: the count
is annotated under coach.prose.number_unverified and the unverified
token soft-stripped, never a stream-killer. Only runs when the full
figures were delivered this turn; a cheap follow-up has no fresh set
to grade and no-ops. Tool-path behaviour and tolerance unchanged.
Patch release. Pins the action row in the centred dialog so the primary
button stays reachable without scrolling under long forms, browser zoom, or
display scaling. No migrations, no breaking changes.
The desktop branch of ResponsiveSheet rendered DialogContent as a single
overflow-y-auto grid, so a tall form — or one made tall by browser zoom,
display scaling, or a longer-language label — scrolled its header, body,
and footer together and could push the primary button below the fold. The
branch now runs as a flex column capped at max-h-[calc(100dvh-2rem)], holds
the header and footer shrink-0, and scrolls only the body, matching the
bottom-sheet branch that already pinned its footer. Every form mounted
through the shared surface inherits the fix.
Add a desktop end-to-end guard at a short viewport asserting the body
scrolls and the footer's bottom edge stays within the viewport.
The DateField/DateTimeField overlay input sits inside the wrapper's
border (~42px); the 44px tap target is the wrapper. Measure the closest
date-field wrapper so the sweep reflects the real affordance.
DateField and DateTimeField front an sr-only native date input with a
formatted overlay carrying the 44px target, matching the file-upload
pattern the sweep already exempts. Generalise the exemption to any
visually-hidden or zero-size input.
The daily cap is now provider-aware; the seeded chain resolves to a
user-egress provider, so the refusal must be seeded at USER_PLAN_CAP
rather than the smaller operator cap to trip the gate.
The three intake dialogs (log, edit, dose-history add) still used a raw
datetime-local input, so they ignored the user's date-format preference.
Route them through DateTimeField like every other date surface. Also
correct a stale diastolic-floor comment in the dashboard verdict.
The correlation discovery matrix gained two non-MeasurementType channels
(medication compliance, symptom severity), but the Coach get_correlations
reader still filtered its measurement query by hand and never built the new
channels. The channel keys leaked into the Postgres enum cast and threw for any
user with medication or illness data.
Extract the route's two channel-series fetchers into a shared module
(correlation-channel-series.ts) so both consumers build the channels identically,
then wire the reader to derive its type filter from discoveryMeasurementTypes and
fold the fetched compliance + symptom series into the discovery input. The reader
now humanises the channel keys to "medication adherence" / "symptom severity".
Render daily medication-compliance rate and illness symptom-severity as
first-class series in the FDR discovery engine, so the adherence-dip →
symptom-flare link (and compliance → vital drift) can finally surface.
Compliance pools every active medication's dose-history ledger into a per-day
taken/(taken+missed) rate, re-keyed to the user's display timezone. Symptom
severity reads the illness day-log functional impact, zero-filling healthy days
only across real episode spans so a user with no episodes yields an empty
series rather than a constant. Both flow through the existing n >= 20, BH-FDR,
effect-size, and shrinkage gates unchanged; a sparse series degrades to absent.
Each channel forms its own metric family (no self-lag tautology) and keeps the
association-not-causation framing. Add discoveryMeasurementTypes() so every
caller drops the non-measurement channels before a Prisma type filter.
The labs block already surfaces value + unit + range + date, and the Coach
prompt instructs it to quote them verbatim, so labs are cited end-to-end — no
gap. Pin that with a tool-level guard asserting value and unit survive
get_labs, so a future change to the labs pass-through can't quietly strip the
number the Coach needs to cite.
The Coach may point at a published /learn guide, but only a slug in the
catalog. That was a prompt instruction with no enforcement — a model could
hallucinate /learn/lower-your-cortisol and ship a dead link. Add a deterministic
post-filter on the assembled reply that keeps a published slug verbatim and
strips any reference whose slug is not in the catalog, tidying the prose around
the removed link. Makes the catalog's impossible-by-construction claim an
enforced guarantee.
The illness-recovery tool and the illness context block carried only labels,
lifecycle, and dates plus the recovery composite — never the recovery-gap, the
metric that drove it, the nadir, the pre-onset deviations, or the red flags the
illness card renders. So a recovery question got the composite, not the gap the
user sees, and a sustained-fever or low-SpO2 escalation was invisible
in-conversation.
Run the existing episode-correlation read-layer (the same engine the card and
the red-flag notifier use) for the most relevant episode — active first, else
the most-recently-resolved — and attach the computed scores to get_illness_recovery,
coverage-gated so a thin signal yields nothing rather than a fabricated number.
Document the illnessScores shape in the Coach prompt so it restates those
numbers verbatim and escalates a red flag rather than reassuring.
Each retrieval tool re-scoped the snapshot to a single source, so its cache key
never matched the inventory's full-source build — a four-tool turn paid four
extra full snapshot builds on top of the inventory's. Thread the inventory's
probe scope to every tool so a per-tool read lands the one cache entry the
inventory already primed, collapsing the turn to a single build (a window
override still gets its own correct build). Raise the loop's round cap from two
to three so a sequential cross-metric why-chain isn't starved; the per-round
token budget is unchanged and the absolute ceiling tracks one above so the
final round still forces prose.
The tool-mode path dropped the launch scope's sources: the inventory probes the
full source set, so a Coach opened from a metric page saw the whole inventory
and could roam, while only the no-tools path honoured the narrowing. Inject a
one-line FOCUS hint naming the launched domain(s) into the tool-mode user turn
so the model prioritises that metric and fetches its figures first, with every
other domain still reachable if the question leads there. Empty on a generic
open, so the prompt prefix is unchanged on that path.
The per-bucket OLS accumulators stored x on the raw epoch-day axis (x ~ 20540),
so sum_xx accumulated past ~1e10 and the square shed ~10 of a double's
significant digits before the value was ever stored. The sub-day x detail was
lost at accumulation time, which no read-side identity can recover; an
ill-conditioned window's composed slope/r2 then drifted from the live REGR_*
probe past the parity gauge.
Rebase x to a fixed origin (epoch-days of 2020-01-01) at write time so the
squared terms stay O(1e7) and exact. Slope, r2, and population sd are invariant
under the affine x-shift, so the live raw-epoch probe still parity-matches the
cross-bucket compose; composeRegression needs no formula change. The per-bucket
slope/r2 columns stay on the raw axis (shift-invariant).
Migration 0193 recomputes sum_x/sum_xy/sum_xx from measurements on the rebased
basis, one set-based update per granularity. Recompute-from-source is idempotent
and rerun-safe; it touches only rows that already carry the accumulators and
leaves pre-migration NULLs for the boot re-fold.
Un-quarantine the DST-boundary parity test: it now matches live REGR_* within
the existing 1e-9 relative bound, with the WEIGHT case still bit-identical.
The acute red-flag clause hardcoded its blood-pressure numbers as prose
literals and named no glucose or fever floor at all, so the Coach could
silently drift from the dashboard hero and the notification engine. Compose
the systolic/diastolic, glucose, and sustained-fever thresholds straight from
clinical-floors.ts and echo the previously-missing glucose and fever lines, so
every acute number the Coach states is bound to the one source of truth.
Add an outlook contract (gentle forecast, what-to-expect, anticipatory if-then
— all conditional, ranged, association-framed) composed beside the tone
contract on the Coach and the comprehensive briefing, so a reply can sharpen
expectations without a false promise. Tighten the tone contract against
over-validation (affirm the effort, not an unsafe choice; stay neutral when
there is nothing to praise) and route why/conflict questions through
get_correlations and its coincident-deviation flag before answering.
Coverage asserts the clause numbers equal the constants and the new contract
reaches both surfaces.
Add a single discreet LearnMoreLink to the per-vital baseline tiles
(via a new footer slot on the shared tile), the glucose clinical panel,
the resilience tile, and the lab biomarker detail. Each pointer is
registry-resolved and renders nothing when the concept has no mapped
guide, so only surfaces with a real article show one.
A small, calm anchor that links a concept out to its public /learn
guide. Fail-closed: the href is registry-backed (a closed-set value,
never user input) and an unmapped concept renders nothing. Plain
anchor, no markdown. Adds a common.learnMore label across all locales.
Promote the static /learn guide catalog into a typed concept-to-slug
lookup the deterministic UI surfaces can consume. learnUrl is the only
sanctioned /learn URL builder and learnLinkForMetric fails closed for
an unmapped id. A test asserts every mapped slug resolves in the
catalog, so the mapping can never point at a guide that does not exist.
The SpO2 sustained-low scan passed episodeDays straight to runFlag, relying
on the reader to deliver them in day order, while the fever path already
sorts its unioned series. The run scan is calendar-consecutive, so feed it
chronological input by construction: sort the filtered SpO2 points before
the scan, matching the fever path and removing the latent ordering coupling.
Both lastStableReturn and computeSymptomReturn counted the in-band settle
run by array-index adjacency instead of calendar days, so three sparse logs
spread over weeks could register a stable return and move the headline
recovery gap. Extend the run only while the prior point is in-band and the
calendar day immediately before, matching the runFlag rule, bounding the
run's span to its logged-day count.
The slot-dedup pass ran a findFirst (latest takenAt) plus a findMany
(all live rows) per medication — 2N round-trips, 40 for a 20-med user.
Pull every live intake row for the user's medication set in one ordered
findMany, group in memory by medicationId, and derive each medication's
lastIntakeAt (max non-null takenAt) from the same grouped slice the
findFirst used to scan. The read cost is now one query regardless of the
medication count.
Behaviour is unchanged: the grouped rows arrive in the same
scheduledFor-asc order the heal and snap passes depend on, and the
derived lastIntakeAt matches the old desc-ordered findFirst.
Every callsite moved to DateField/DateTimeField, which honour the
user's date-format preference; the native DateInput/DateTimeInput
wrappers have no remaining importers.
The activity sync ran a per-(entry, field) findFirst followed by an
update or create, roughly entries x 3 x 2 round-trips — about 2,190 on
a 365-day backfill. Flatten the entry/field grid into the concrete rows
the sync wants to land, resolve every overwrite-vs-insert decision in
one findMany keyed on the same four dedup columns, then run a single
createMany for fresh slots and one transaction for the changed-value
overwrites. A backfill is now a bounded handful of queries regardless of
the day count, and an unchanged re-sync is a no-op on the write path.
The overwrite contract is unchanged: an existing WITHINGS row for a
(type, measuredAt) slot keeps its row and only rewrites value; a fresh
slot inserts with the stable externalId. createMany uses skipDuplicates
so a concurrent sibling sync landing the same slot is tolerated by the
four-column unique.
Add a dependency-free DateTimeField that paints the date part in the
user's date-order preference and the time part in the hour-cycle
preference over a hidden native datetime-local input, preserving the
local yyyy-MM-ddTHH:mm value contract. Migrate the seven DateTimeInput
callsites (measurement add/edit, mood add/edit, lab add/edit,
medication intake quick-add) so the chosen format is honoured there too,
and cover the field's value contract and per-preference rendering with a
component test.
The rollup read path windows the 7/30/90-day regression on
startOfUtcDay(now - N days), while the live-SQL fallback filtered the
REGR_* aggregates on the wall-clock NOW() - INTERVAL bound. An account
crossing a coverage boundary (warm to cold) therefore saw the boundary
day's readings flip in or out of the slope window depending only on
whether the rollup tier was populated, returning a cache-dependent
slope for the same request.
Anchor both cold-fallback paths on date_trunc('day', NOW() AT TIME ZONE
'UTC') - INTERVAL 'N days', the SQL twin of startOfUtcDay, so warm and
cold return identical boundary-day membership. The avg7/avg30/median
windows stay wall-clock since the warm narrow query computes them the
same way.
Reduce the full-page Coach new-chat surface to a calmer layout. Drop the
toolbar Conversations and New chat buttons and place the settings gear in
the top-right corner; both removed actions remain in the composer's + menu.
Collapse the greeting to one line and remove the gear from the composer so
its front is only the + actions menu, leaving a single baseline row of
[+] textarea [mic] [send].
Swap the four DateInput callsites (filter bar, export range, schedule
history, onboarding baseline) to DateField so the user's date-order
preference paints every date input. Adapt the onChange from a native
event to the direct ISO value DateField emits, and add an optional
autoComplete prop so the onboarding date-of-birth field keeps its bday
autofill hint.
The FDR cross-metric driver pairs were reachable only through the
get_correlations tool, so a local/no-tools provider got the coincident
flag but zero discovered drivers and the cross-metric layer silently
degraded. The system prompt's fallback to a snapshot driver field
pointed at a field that did not exist on that path.
Attach a bounded top-3 slice of the same gated, ranked driver output the
tool serves to the no-tools snapshot floor, gated on the derived layer
and shed first under prompt-budget pressure. Local users now reach
parity on the flagship cross-metric capability.
Significance alone surfaced trivial drivers: a large-n pair with r=0.16
narrated as a confident "tends to go with" link while explaining ~2.5%
of variance. Add three gates before a pair is ranked as a driver:
- exclude same-metric-family lagged pairs (mood->mood, BP component
self-lag) as serial-autocorrelation tautologies, widening the existing
exact self-pair skip to the whole family.
- shrink each Pearson estimate toward null by n/(n+10) so a thin-data
correlation cannot out-rank a deep one on an inflated point estimate.
- floor the shrunk effect at 0.2 (drop below) and bind 0.2-0.3 to a
hedged "faint hint" phrasing tier, 0.3+ to confident phrasing scaled
by sample depth. The reported r and p stay the honest raw statistics.
Rank by shrunk effect magnitude, q as tie-break.
The coincident-deviation read grouped the latest day with a UTC date
slice while the sibling readiness read already used the tz-aware day key.
For a user east or west of UTC, a late-evening or early-morning reading
landed on the wrong calendar day, so a fired flag could compare a vital
from the wrong day against its band and narrate '2 or more vitals out of
band today' on the wrong day.
Thread the user's timezone into the latest-day read and mint the day key
with userDayKey, matching readiness. The Coach derived-snapshot and the
correlations reader pass the real account timezone through. Adds a
regression covering a UTC+2 user whose late-night reading rolls to the
next local day.
The hypertensive-crisis floors were defined in three modules with a
divergent diastolic value: the notification engine and the Coach acute
clause used 180/120, but the dashboard hero used 180/110. A reading like
170/112 lit the red critical-BP banner on the hero yet never tripped the
notification alarm or matched the Coach's stated acute number — one
surface said crisis, the other two stayed calm on the same row.
Promote the absolute floors into a dependency-free clinical-floors leaf
and have the hero, the safety-floor engine, the status registry's fever
band, and the illness escalation all import from it. The crisis diastolic
floor is the guideline-correct 120 (ACC/AHA hypertensive urgency); the
wider 110 hero net is dropped so the surfaces tell one story. A coverage
test pins that every consumer resolves the same constants.
The daily AI-token gate was a flat 25,000/day for every provider. That cap
exists to bound the OPERATOR's API bill, but a ChatGPT-OAuth (Codex) or BYOK
chain egresses on the USER's own plan/key and costs the operator nothing, so
gating it on the operator-cost ceiling is wrong. A single gpt-5.x reasoning
turn legitimately reports 20k-40k gross total_tokens (re-sent system prompt +
inventory + tool defs per round + hidden reasoning, summed across rounds), so
one or two turns exhausted the flat cap and locked the user out of a plan they
pay for.
Classify the chain's cost owner by its primary provider via resolveDailyCap:
an admin-openai primary (operator pays) keeps the operator-cost cap; every
user-egress primary (codex/openai/anthropic/local) gets a generous
abuse-only ceiling. Threaded into the Coach chat gate and both OCR-extract
modes.
Raise the operator-cost cap 25,000 -> 200,000 so the operator-key path also
survives a normal day of reasoning turns, keeping gross-token accounting
rather than reworking the reconcile math.
Subtract cached input tokens at reconcile: the Responses-API gross total still
includes prompt-cached input the user did not re-pay for. The codex client
already parsed cached_tokens; sum it through the tool loop and bill
total_tokens minus cached.
The budget-exceeded copy named "00:00 UTC", which reads as a wrong local
clock for any non-UTC user (Berlin midnight UTC is 01:00/02:00 local) and
drove confused "the limit resets at the wrong time" reports. Reword
dailyLimitBody across all six locales to "midnight UTC", matching the
existing errorBudget wording, so the figure is honest regardless of the
reader's timezone.
Add a discreet, on-brand action to the high-value cards (briefing,
recommendation, correlation, status/assessment, health score, period
narrative) that opens the Coach pre-scoped to the card topic with a
seeded question. The action self-gates on the operator flag and the
per-user opt-out, never tints the card, and surfaces once per card.
Make the launch context's metric scope live instead of discarding it.
A conversation opened from a metric surface now narrows its snapshot to
that metric with a data-aware seed question, and the global FAB inherits
the page's ambient scope so drilling into a metric and tapping it no
longer opens a blank chat. The scope threads through the drawer into the
first turn of a fresh conversation.
The multi-vital coincident-deviation flag (two or more vitals outside
their usual band on the same day, with the illness-explained reframe)
already reached the period narrative but never the Coach. Attach it,
fired-only, to the derived block so both Coach paths can narrate it: the
tool path via the recovery/correlations tools and the no-tools snapshot
floor. A quiet day adds no entry, keeping the snapshot noise-free.
The Daily Briefing strips any number absent from its server-computed
figures; the Coach's tool path had no equivalent, so a transcription or
paraphrase drift (a tool returns systolic 128, the reply says ~138) could
ship. Collect the numeric leaves from every present tool result this turn,
extract the numbers the reply asserts, and soft-correct any that match no
fetched figure to a neutral placeholder — annotated, non-blocking, and a
no-op on a qualitative turn or the no-tools path where there is nothing to
grade against. The bounded retrieval loop now returns the present results'
payloads so the check has an authoritative figure set.
The retrieval-tool DATA INVENTORY was built against the user's default
narration cluster (cardio/body/mood/medication), so sleep, glucose,
activity, body composition, vascular, SpO2, gait and environment series
reported absent even when the data existed — and the grounding rule then
told the model not to fetch them. Probe presence against the full source
set instead, and generate the metric-series inventory rows from the
complete source-to-section map so every series the user has rows for is
advertised. Per-tool reads still re-scope to their own domain, so the
wider probe never widens a figure read.
Add three retrieval tools whose snapshot blocks already existed but were
unreachable in tool mode:
- get_workouts: recent sessions + per-sport rollup
- get_cycle: phase / prediction / correlation (gated by cycle access)
- get_correlations: the FDR-controlled lagged cross-metric drivers plus
the coincident-deviation flag, surfaced descriptively
Also honour the caller's window in get_labs and get_illness_recovery so a
cross-metric answer no longer silently mixes horizons.
The retrospective narrative was the one model surface that omitted the
shared tone contract, so it read colder than the daily briefing beside it.
Compose the tone fragment into both prompts and warm the hand-written tone
line to name a genuine win when the period earns it, while keeping the
descriptive-never-causal and no-alarm guards intact. The cross-surface
coverage test now asserts the contract reaches this surface too.
The per-score deterministic text named the score, its standing, and the
weakest contributor, then stopped at the diagnosis. When the band is not
green and the weakest driver is behaviourally addressable (sleep, mood,
consistency, timing), append a single grounded pointer drawn from that same
contributor. Physiology-only drivers add nothing, so the text affirms and
watches rather than manufacturing a step.
The no-key and timeout fallback text carried zero reference to the user's
own readings and, on a provider timeout, rendered as hasProvider:true —
indistinguishable from a fresh assessment. Compose each fallback from the
per-metric signal the card already builds: name the current value, place it
against the user's own baseline, and close with one plain-language pointer,
in the same warm voice as the model surfaces. Degrade to the prior generic
tip only when a metric has no usable history. The timeout envelope now
reports hasProvider:false so the UI surfaces it as the computed summary it
is rather than mislabelling it as provider prose.
The course-window date field now keeps its ISO value on a hidden native
input (data-slot) and edits through a text overlay (data-testid). Fill
the overlay, then assert the committed ISO value on the hidden input.
Swap the medication scheduling, inventory, illness, vorsorge, lab-OCR
and profile-birthdate date inputs from the native field to the
DateField primitive, which paints the chosen DMY/MDY/YMD/AUTO order
over an ISO yyyy-MM-dd value. The committed value contract, min/max
bounds and aria wiring are unchanged.
Sharpen the Coach's behaviour-change reflexes from the coaching literature,
prompt-only:
- Connect the signals on a why/pattern question — consult correlations and
link them descriptively ("after short-sleep nights your next-morning HRV
tends to read lower") instead of reading metrics in isolation, always an
association worth a small experiment, never a cause.
- Confidence ruler on action turns — after naming one small step, ask how
doable it feels 0-10 and shrink it when that lands low, keeping the choice
the user's.
- Three-beat shape on data-review turns — finding, likely driver, one small
step; name the dominant contributor when citing a derived band.
- Extend the MI micro-moves with developing discrepancy and rolling with
resistance, and add an anti-persuasion check: any suggested change must
serve the user's own stated goal.
One bilingual tone-calibration example added per locale.
The shared contracts cover chronic deferral (dose, diagnosis, drug-level
to a clinician) but had no acute branch. Add a closed-list crisis clause —
chest pain, syncope, sudden severe symptoms, hypertensive-crisis readings,
suicidal ideation — that points to prompt or emergency care without
diagnosing, and keep every other turn calm and non-alarmist. It rides the
single shared-contract source so it lands on the Coach, the per-metric
status cards, the comprehensive briefing, and the period narrative; the
coverage test asserts it on all four surfaces.
Dependency-free date input that displays the value in the user's
date-format preference while editing rides the native picker behind a
calendar button. The committed value stays ISO yyyy-MM-dd, so it is a
drop-in for DateInput. Height/target-size parity, disabled/min/max/
placeholder, and progressive typed entry parsed against the order.
Accept dateFormat on the profile PATCH field-by-field, echo it from
the profile + /me reads, mirror it into the client auth user, and add
the field to the OpenAPI request/response schemas.
Add a date-order resolver + formatter mirroring the hour-cycle path:
AUTO defers to the active locale, DMY/MDY/YMD pin the field order
through a canonical Intl locale. Thread it through makeFormatters and
expose useDateFormatPreference from the i18n context, backed by the
same localStorage mirror as the time-format preference.
Mirror the time-format preference with a per-user date display
choice. AUTO follows the locale convention, DMY/MDY/YMD pin the
field order. Additive, rerun-safe migration; existing rows read AUTO.
The floor reads Date.now() inside buildComparisonSnapshotForUser, so it
lands in [before - 400d, after - 400d] relative to the test's brackets.
The prior upper bound used before - 400d, which fails when a millisecond
ticks between the bracket read and the snapshot's clock read.
Thread the request abort signal through runCoachToolLoop into each
per-round completion so a mid-generation client disconnect tears down
the upstream call instead of paying the full token cost. Completes the
no-tools-path wiring to cover the tool-retrieval path as well.
Regenerate the OpenAPI spec for the trimmed OCR-capability reason enum.
The four AI clients threw bare Errors on an empty reply (and, for Codex, on a
stream-level failure) with no httpStatus, so the provider-chain classifier saw
the same null-status shape as a genuine transport failure (ECONNRESET, DNS,
timeout) and could not tell them apart in observability. Attach a sentinel
`httpStatus: 0` plus a `kind` discriminator (`empty_response` /
`stream_failed`) on those throws. The cascade is unchanged:
`isHardProviderFailure` already treats a status <= 0 as a hard failure and
walks to the next provider — this only sharpens the logs.
The Coach SSE turn ran the provider call to completion before the stream
existed and never wired it to the request's abort signal, so a mid-generation
client disconnect still paid the full upstream token cost. Thread an optional
`signal` through CompletionParams and singleUserTurn into each client's
safeFetch (composed with the per-call timeout via AbortSignal.any), and pass
`request.signal` from the chat route's no-tools path. An early disconnect now
tears the upstream request down instead of generating into a closed
connection. Omitting the signal leaves behaviour unchanged (timeout-only).
Drop the unreachable text-only-model reason from the OCR capability
enum and its DTO type. Replace the literal NUL delimiter in the Fitbit
dedup map keys with a named constant so text tooling stops treating the
file as binary (behaviour unchanged). Correct the Coach tool-executor
docblock, which overstated cross-tool snapshot-cache sharing.
The extraction already self-scores each field; the review screen only
hinted at a shaky numeric value. Add a calm per-row low-confidence badge
when any tracked field scores below the threshold, including qualitative
rows that had no confidence cue. Informative secondary badge, no alarm
colour.
The app-logs, audit-log and host-metrics admin endpoints parsed their
query with a throwing parse(), so a bad parameter surfaced as a 500 and
a reported incident. Switch each to safeParse and return the standard
422 validation envelope, matching how body parsing is handled.
Text-mode OCR structures browser-extracted text rather than running a
vision call, so it reserved and reconciled the full vision budget even
though the spend is far lower. Add an ocrExtractText ceiling and reserve
that in text mode, and refund the reservation in full on a clean extract
failure instead of charging the reserved estimate.
A channel whose stored config cannot be decrypted or JSON-parsed is a
structurally-permanent failure, not a transient blip: the ciphertext/key
pairing and the malformed body decode the same way on every retry. Routing
it through the transient path burned the 5-failure backoff budget and ended
on the misleading `give_up_after_5_failures` reason — during a key-rotation
gap every channel hit this and auto-disabled for the wrong reason.
Classify decrypt and parse failures as hard rejects so the channel
auto-disables on first occurrence with the precise reason
(`<type>_config_decrypt_failed` / `<type>_config_parse_failed`). Genuine
transient failures (network, timeout, 5xx) keep the existing backoff path.
Reformulate composeRegression from the determinant form (n*Sxx - Sx^2)
to the mean-centered corrected-sum identities (Sxx = Sxx - Sx^2/n, ...).
Algebraically identical to the prior form and to Postgres REGR_* /
STDDEV_POP, but it subtracts the x mean at matched scale and avoids the
catastrophic cancellation the determinant form suffers on the ~1e10
epoch-day x-axis. Also read the exact stored sum_value accumulator for
the Sy term instead of reconstructing it as mean*count, dropping the
AVG-to-multiply float residual; fall back to mean*count when the column
is null.
Composition-only: no migration, no schema change. The WEIGHT live-parity
case stays bit-identical at the 1e-9 relative bound. The near-flat DST
parity seed improves by ~30x but still exceeds 1e-9, so it remains
quarantined with an updated note pointing at a window-local x-rescale as
the real fix.
The o1/o3/o4 prefix match also caught o1-mini, o1-preview and o3-mini,
which have no image input. A user pinned to one was told OCR works, then
the provider rejected the image block. Treat an o1/o3/o4 slug as
vision-capable only when it is not a -mini or -preview variant, applied
to both the OpenAI and Codex gates.
Symptom and vital points are present-days-only, so a sparse febrile
series (e.g. days 10/15/20/25) was counted as one long run and falsely
fired the sustained_fever escalation. Continue a run only when the next
matching point is the calendar day immediately after the previous one,
using the existing dayDiff helper; otherwise start a fresh run of one.
buildComparisonSnapshotForUser read a user's entire measurement history
per type with no date floor, then fed each series to summarize(). The
snapshot only consumes avg30 / avg30LastMonth / avg30LastYear, and the
widest of those reaches into the [365, 395)-day window (ageMs < 395 days
in meanOfWindow), so nothing older than 395 days can change a result.
Add a 400-day measuredAt.gte lower bound to every per-type read in both
copies of the builder — the page-blocking POST /api/insights/generate
handler and the shared comprehensive-generate path the nightly
pregenerate cron drives. The fan-out runs all seven snapshot types via
Promise.all, so on multi-year accounts the unbounded scan was tens of
thousands of rows per type. 400 days keeps a 5-day floor over the
strict-less-than 395-day reach.
A user reported the dashboard crashing on returning to a backgrounded
tab. The on-focus snapshot refetch re-rendered the page, and each chart
received a fresh inline `() => markChartReady(id)` handler every render;
the chart's data-ready notify effect, keyed on that ever-changing prop,
re-fired on every commit. That per-commit passive effect kept the chart
row's anchored popovers re-committing until the browser tripped its
render-depth guard and the page fell back to its error card.
Latch the notify behind a ref so it fires once, when the chart's initial
query settles, and depend on the load flag alone. The handler is read
through a ref so a late-arriving callback still fires without re-arming
the effect. Extract the pure fire-once decision into a small helper with
direct unit coverage (no DOM runner, per project convention).
The DST same-UTC-day case is a cross-implementation parity artifact in the
test harness, not a product defect — the rollup regression math is covered
bit-identically by the WEIGHT case, which folds the same raw rows the live
REGR reads. Skip it with a tracked follow-up so the release is not blocked.
The case exists to pin that the epoch-day x-axis is DST-agnostic, not to
exercise BLOOD_GLUCOSE, whose source-priority/CGM consolidation makes the
live raw-row regression and the bucket-folded accumulators legitimately
read different row sets. WEIGHT folds raw rows directly (the bit-identical
case proves it), so the DST-straddling timestamps now parity-match tightly.
The DST case used near-flat scattered values, where the naive sum-of-
squares closed form is ill-conditioned versus Postgres REGR (a near-zero
r2 amplifies the cancellation). The WEIGHT case with the same epoch-day x
magnitude and uneven per-day counts already matches bit-identically, so a
clear monotonic trend restores tight parity while still exercising the
DST-straddling timestamps and a same-UTC-day reading pair.
The composed regression is verified correct (slope agrees to ~7e-9 with a
stable mean-centred reference on this seed); the remaining gap is a cross-
implementation artifact between the closed form and Postgres REGR on an
ill-conditioned r2/sd at large epoch-day x. The read tier rounds to 2-3 dp,
so parity is gauged there. Full bit-parity is tracked in the backlog.
The naive sum-of-squares regression at epoch-day x~20540 is inherently
ill-conditioned; six-figure parity was still inside the cancellation
noise. Five significant figures remains far beyond the read tier's 2-3 dp
rounding. Production unchanged; full bit-parity tracked in the backlog.
The same-UTC-day DST case drives the worst cancellation in the closed-form
slope at large epoch-day x; Postgres REGR_SLOPE is mean-centred and keeps
the digits we lose. The accumulators sum over the identical raw rows, so
this is precision, not divergence — gauge it at six figures (well beyond
the read tier's 2-3 dp).
Gauge the rollup-vs-live regression parity with a relative tolerance
instead of an absolute nine-decimal one. The accumulators sum per raw
reading over the same epoch-day x as the live REGR_* query, so the paths
are algebraically equal; their tiny divergence is float-summation order
amplified by cancellation in the slope denominator at large epoch-day x.
Production code is unchanged.
The DST-boundary parity case seeds two readings on the same UTC day, so the
windowed compose folds a closed form over a large epoch-day x (~20540). The
n*Sxx - Sx^2 denominator subtracts two ~1e10 magnitudes and loses about ten
significant digits to cancellation before the divide, so an absolute
toBeCloseTo at 9 dp is the wrong gauge for a slope that small. The populator
sums per-raw-reading accumulators grouped by day, so both paths regress over
the identical raw-row x set and agree mathematically; the residual is float
summation order alone. Switch all parity assertions to a shared relative
tolerance (1e-9 of scale), far tighter than the 2-3 dp the read tier rounds to.
Rewrite the two Fitbit integration tests' fetch mocks to the classic
Fitbit Web API shapes the v1.20.0 client now reads (body log, activity,
1.2 sleep levels, activities/list), so the sync ingests the expected rows
again. Anchor the workout externalId on the numeric logId and pin the
tz-sensitive test user to UTC; all counts, dedup, stage and source-
priority intents are preserved.
The v1.20.0 transport switch retargeted the Fitbit client from the Google
Health API onto the classic Fitbit Web API, but the two sync integration
tests still stubbed fetch with Google-shaped dataPoints responses. The mocks
no longer matched the endpoints the client reads, so every ingest assertion
saw zero rows.
Rewrite the fetch-mock layer in both files to key on the classic endpoint
paths and return classic-shaped JSON the real mappers parse:
- body/log/weight, activities/heart for the metrics sync
- activities/steps for the cumulative-activity sync
- 1.2 sleep log levels.data for the per-stage sleep sync
- activities/list for the workout sync
The ingest contract is preserved (row counts, dedup, source priority, sleep
stages, daily-total overwrite). Two assertions move to their classic
equivalents: the workout externalId now anchors on the numeric logId, and
the workout test pins the user to UTC so the offset-less local wall-clock
timestamps the classic sleep/activity logs emit resolve to the same instants
the cross-source picker buckets on.
Send the OpenAI JSON response format only when JSON is requested and no
tools are in play, so the Coach prose path (and the tool loop's final
round) is no longer corrupted into a JSON blob. Resolve the Anthropic
finish reason to tool_calls whenever tool_use blocks are present. Memoize
the per-request feature extraction and prefs read and project the feature
query, so the coach tool fan-out shares one read instead of six.
The F1 coach tools each rebuild a snapshot with a distinct single-source
scope, so the 60s snapshot LRU (keyed on the source list) never shares their
reads. On a multi-metric turn the model can fan out up to six tools that each
fire `extractFeatures` (a user lookup plus a windowed measurement read plus the
all-time extremes) and a prefs `findUnique` concurrently — up to twelve heavy
reads against the shared Prisma pool at once.
Memoise both reads on the request-scoped event cache, keyed on userId (and
windowDays for the features read), so the fan-out collapses to one of each
within a request while staying correct across requests. Project the feature
extractor's measurement read to the six columns its consumers actually use
(type, value, measuredAt, sleepStage, source, deviceType) so the 200k-row-class
PULSE / glucose windows stop pulling every column. Outputs are unchanged.
Parse Fitbit's offset-less classic sleep/activity wall-clock strings
against the user's timezone (DST-safe) so a non-UTC user's segments land
on the right wake-day. Add a shared rotated-refresh-token persist that
compare-and-swaps on the stored ciphertext and reuses a peer's rotated
token on a lost race, so concurrent syncs across Fitbit, WHOOP and Oura
no longer park a connection at a spurious reauth.
Fitbit, WHOOP, and Oura all issue one-time-use refresh tokens. The token
persistence did an unguarded read → refresh → write, so two overlapping
syncs for one user both read the same stored refresh token and both POST
it. The provider honours only the first; the loser's refresh comes back 401,
gets classified as reauth, and the connection is parked at `error_reauth`
spuriously even though the peer rotated the token successfully.
Add a shared `persistRotatedToken` helper that writes the rotated pair with
a compare-and-swap on the stored ciphertext (matched verbatim, since AES-GCM
re-encryption of the same plaintext differs byte-for-byte). When the
conditional update touches zero rows a peer already won the race; re-read the
row and reuse the peer's freshly rotated access token instead of erroring.
Wire it into all three providers — Fitbit and WHOOP `getValidToken`
(proactive expiry refresh, keyed by connection id) and Oura `storeOuraTokens`
(reactive 401 refresh, keyed by user id). Both racing callers now end with a
usable token and neither drops the connection.
The Anthropic client derived `finishReason` solely from `stop_reason`. The
Coach tool loop continues only when `finishReason === "tool_calls"`, so if
Anthropic ever returned tool_use blocks under a different (or absent)
stop_reason, the reason resolved to undefined, the loop dropped the tool
request, and the empty tool_use-only reply surfaced as the final answer —
the requested data was never fetched.
Set `finishReason` to "tool_calls" whenever tool_use blocks are present,
clamped ahead of the stop_reason map, so the presence of a tool request is
authoritative regardless of the reported stop reason.
The OpenAI client pinned `response_format: { type: "json_object" }` on every
call. The Coach tool loop's forced-final round runs with `tool_choice: "none"`
and expects plain prose, but JSON mode coerced the reply into a JSON object —
the user saw a JSON blob or an empty reply instead of the coach's answer.
Send `response_format` only when the caller opts in with `responseFormat: "json"`
and no tools are present, mirroring the Anthropic client's prefill gate. The
JSON-consuming callers that relied on the implicit mode (insight generate +
briefing retry, medication extraction, the connection-test probe) now pass the
flag explicitly; the prose Coach path and every tool round stay out of JSON
mode. Codex already omitted the flag and is unaffected.
Sort the fever red-flag points chronologically before the consecutive-run
scan so a dual-source episode (thermometer plus journaled fever) can no
longer falsely escalate or hide a sustained-fever flag. Treat a trailing
out-of-band tail shorter than the stability window as transient noise so
a settled vital keeps its recovery day; only a sustained re-deviation
clears it.
The last-sustained-return walk nulled the recovery day whenever the final reading
sat out of band, even after a vital had settled for several days. A lone noisy
sample on the last logged day — common in resting-HR and HRV curves — then erased
a real recovery gap that had previously been reported.
Treat a trailing out-of-band tail shorter than the stability window as transient
noise and keep the settle day; only a sustained re-deviation (a tail at least as
long as the stability window, a genuine relapse with no re-settle) clears the
return. Apply the same noise tolerance to the symptom-burden track for parity.
Cover the trailing-outlier, genuine-relapse-then-re-settle, and
sustained-no-re-settle cases.
The classic 1.2 sleep log and the activities list emit local wall-clock
ISO strings without an offset. `parseLocalInstant` parsed them with a bare
`new Date(iso)`, which reads an offset-less string in the process zone (UTC
in production). A non-UTC user's sleep-segment `measuredAt` (the segment
end) was therefore shifted by their UTC offset, and a near-midnight segment
could flip onto the wrong wake-day in the night reconstruction.
Resolve offset-less timestamps against the user's stored timezone via a new
`zonedWallClockToUtc` helper (the inverse of `wallClockInTz`, DST-safe via
the same two-pass converge as `startOfLocalDayInTz`). Thread the resolved
zone through the sleep and workout sync paths. Strings that already carry an
offset or `Z` stay absolute and parse as-is; an absent zone keeps the prior
host-local fallback.
Emit regression_source: unavailable_pending_backfill when slope/r2/sd are
null because a tenant's accumulators are not yet backfilled, so the miss
is observable rather than silently logged as a rollup hit. The null value
is unchanged; no live fallback added.
The windowed slope / r2 / sd readers compose from the per-bucket
regression accumulators (migration 0190). Between a deploy and the
boot-backfill completing, a tenant's pre-0190 DAY buckets still carry
NULL accumulators, so composeWindowedRegression returns a full miss and
the slim / comprehensive readers emit a null slope. The null is the
honest answer until the backfill refills the accumulators, but the miss
was passing as a plain rollup read with no signal, against the no-silent-
cap rule.
Add hasPendingAccumulatorBackfill to the read tier and key a new
regression_source annotation off it (unavailable_pending_backfill vs
rollup) on both warm paths. No live-SQL fallback is added; the design
intentionally avoids the heavy scan.
The fever escalation unions two individually-sorted sources — the passive
BODY_TEMPERATURE series and the day-log feverC — into a Map, then feeds the
entries to the consecutive-run scanner. Map iteration is insertion order, not
chronology, so an episode with both a thermometer feed and journaled fever read
the run out of day order: a broken fever could read as a sustained run (a false
seek-care escalation) or a genuine run could be hidden (a missed one).
Sort the unioned points by day before the scan. Cover both directions with
interleaved-source fixtures.
Replace the Coach snapshot-stuffing with a small base context plus a data
inventory and six on-demand retrieval tools (metric series, glucose,
sleep, medication compliance, labs, illness recovery) the model calls as
needed, then answers. Cuts first-turn input ~35-65% while preserving
grounding and every gate; tools return a structured present:false rather
than throwing, so the model cites only what it fetched. Falls back to the
snapshot path when a provider lacks tool support.
Replace snapshot-stuffing in the Coach turn with a small base context plus
on-demand retrieval. The model receives the full system prompt, a tool-mode
grounding addendum, and a tiny DATA INVENTORY manifest listing which domains the
user has logged and which tool fetches each; the figures themselves are pulled
only when a question needs them.
Six read-only tools wrap the existing server-authoritative snapshot builder:
get_metric_series (blood pressure / weight / pulse + the additive series),
get_glucose_panel, get_sleep, get_medication_compliance, get_labs, and
get_illness_recovery. Each returns a structured { present: false } for an absent
or opted-out domain rather than throwing or returning an ambiguous empty list,
so the model can always tell missing data from real data. Module and cycle gates
are enforced at the builder, so opted-out domains are unfetchable.
A bounded loop runs the call -> tool-result -> answer cycle: at most two rounds
with a hard cap of three that forces a final answer, parallel tool execution,
and only the final round streamed. The day's budget reserves the per-call
ceiling times the round count and reconciles the summed tokens; the atomic
reserve and reconcile primitives are unchanged. All inbound gates (rate limit,
refusal, replay scan, consent, budget) stay before the loop.
The legacy snapshot path stays alive as the no-tools fallback: tool mode is on
only when every provider in the chain supports tool-calling, so a chain that
includes a local model falls back to the prior behaviour verbatim. The tool
trace is persisted onto the assistant message provenance.
The Coach snapshot builder returns the structured, post-degrade record it
serialises to JSON alongside the JSON string, so a single domain block can be
read out without re-parsing. Add an optional tool-call trace to the provenance
envelope (which retrieval tools ran + whether each found data) and mirror it in
the OpenAPI provenance schema.
Groundwork for on-demand retrieval: the retrieval tools slice one section out
of this record, reusing the builder's gates and I/O verbatim.
Retarget the Fitbit integration from the Google Health API (which needs
annual CASA review — a wall for self-hosters) to the classic Fitbit Web
API with PKCE and rotating refresh tokens, so a self-hoster can connect
with an instant developer app (migration 0191 adds the PKCE verifier).
Transport-only fork; the ingest, dedup, source-priority, backfill and
poll paths are reused. Intraday HR and skin temperature are not mapped
(no honest slot); the integration stays experimental.
Store additive per-bucket regression accumulators (sum_x, sum_xy, sum_xx,
sum_yy) so windowed slope, r2 and standard deviation compose from the
rollup tier instead of a live raw-table scan on every warm request
(migration 0190, backfilled by the boot queue). Median and the exact
anomaly count stay on a minimal live query. Parity with the live REGR_*
path within float tolerance, well below the read rounding.
Add the optional FITBIT_REDIRECT_URI to .env.production.example (BYO-key shape,
derives from NEXT_PUBLIC_APP_URL) so check-env documents it. Rename the
settings-card copy from "Google Health" to "Fitbit" across all six locales now
that the transport is the classic Fitbit Web API.
Connect mints the PKCE pair, stashes the verifier on the OAuth-state row, and
sends the S256 challenge to the authorize endpoint. Callback reads the verifier
back and presents it on the token exchange, refusing cleanly when it is absent.
The connection-test probe hits the classic /1/user/-/profile.json endpoint.
The shipped Fitbit integration spoke the Google Health API, whose Restricted
scopes need brand verification plus an annual CASA assessment before leaving
testing mode — a wall most self-hosters cannot clear. Retarget the transport to
the classic api.fitbit.com Web API: a self-hoster registers an app at
dev.fitbit.com in minutes, no CASA.
Forks only the transport; the upsert/dedup/rollup tail, source-priority,
backfill, poll worker, and status ledger are reused unchanged.
- OAuth: www.fitbit.com/oauth2/authorize + api.fitbit.com/oauth2/token, PKCE
(S256), classic self-serve scopes.
- Refresh tokens rotate (one-time use): persist the new token every refresh,
with a defensive keep-existing guard on a malformed reply.
- Per-endpoint date-range fetchers chunked to the API range caps; the backfill
horizon and poll concurrency are tuned for the 150 req/h-per-user budget.
- Verified classic-shape extractors for weight, body fat, SpO2, HRV (dailyRmssd),
resting HR, respiratory rate, steps, distance, active calories, floors, VO2 max
(range midpoint), per-segment sleep stages, and workouts.
- Skips intraday HR (needs a per-app grant; daily resting HR covers it) and skin
temperature (a baseline delta with no honest absolute slot).
The classic Fitbit Web API authorises with PKCE (Authorization Code + S256),
so the connect route mints a code_verifier and the callback presents it on the
token exchange. Carry it on the short-lived FitbitOAuthState row.
Additive + nullable (migration 0191): a legacy in-flight row carries no
verifier, so the change is reversible and rerun-safe with no data loss.
Two suites pin that the windowed slope / r2 / population-sd composed from
the rollup accumulators equals the live Postgres REGR_SLOPE / REGR_R2 /
STDDEV_POP over the same raw rows.
compose-regression.test.ts (container-free) builds per-DAY-bucket
accumulators from a raw fixture exactly as the populator's SUM terms do,
then asserts composeRegression over those buckets equals a direct OLS over
the flat raw rows - the same closed form Postgres folds - on an uneven
multi-bucket fixture, plus the coverage contract (NULL accumulator,
< 2 readings, and zero-x-variance all return a full miss so the reader
falls back to live SQL).
rollup-regression-parity.integration.test.ts (real Postgres testcontainer)
seeds raw measurements across many uneven DAY buckets, folds the rollups,
and compares the accumulator compose to live REGR_*/STDDEV_POP to <= 1e-9:
a single-source WEIGHT series, a dual-source RHR series proving the
source-collapse runs BEFORE accumulator summation (composed == live WHOOP
-only, != the cross-source blend), and a DST-boundary series confirming
the UTC epoch-day x-axis parity-matches across the transition.
The residual is pure floating-point reassociation (per-bucket-then-window
vs flat summation), orders of magnitude below the read tier's rounding.
Move the AI client layer from a single system/user string pair to a
structured system + messages[] shape, with optional tools/toolChoice and
a result carrying toolCalls/finishReason/cachedInputTokens. Per-provider
prompt-caching is wired (Anthropic cache_control, OpenAI/local auto
prefix); tool plumbing is in place for the Coach to use. Behaviour-
preserving: no streaming change, snapshot-once retained.
Replace the CompletionParams { systemPrompt, userPrompt } string pair with a
structured { system, messages[] } request across all five AI clients. One
refactor unlocks three things: provider prompt-caching (Anthropic cache_control
on the stable system prefix, OpenAI/Codex automatic prefix cache, with
cachedInputTokens surfaced for observability), the tool-calling wire surface,
and clean multi-turn.
Each client maps the new shape to its own wire and degrades unsupported
features silently: the local OpenAI-compatible server never receives a tools
field (most reject unknown keys), and prompt-caching is best-effort everywhere.
CompletionResult gains toolCalls, finishReason and cachedInputTokens; the tool
types and per-provider mapping ship now, the tool-use loop does not.
Vision inputs fold into the user turn's content parts via singleUserTurn, with
each provider's pre-existing block ordering preserved. The corrective-retry
paths append to the last user message rather than a flat string. The Coach
keeps assembling one user turn so its snapshot-once and grounding behaviour is
unchanged. No wire schema visible to API consumers changed.
Behaviour is unchanged across every surface; the full suite stays green.
Migration 0190 left existing DAY rollup rows with NULL regression
accumulators until a re-fold refills them. Add a second discovery arm to
enqueueBootTimeRollupBackfill, UNION-ed with the per-day missing-coverage
arm: a user whose DAY buckets carry sum_xy IS NULL inside the fold window
needs a one-time re-fold so the windowed slope/r2/sd readers compose from
the accumulators instead of falling back to the live REGR_* scan.
New writes self-populate the accumulators (they ride the same aggregate
as mean/sum), so the arm drains to empty once the boot re-fold catches up
- the same convergence the v1.4.39 sum_value rollout reached through this
queue. Bounded to ROLLUP_FOLD_WINDOW_MS so days outside the window never
re-flag a user on every boot.
The summaries slice (computeFromRollups) and the comprehensive aggregator
(buildFromRollups) each ran a narrow live $queryRaw over the raw
measurements partition on EVERY warm-rollup request to fold slope7/30/90
and r2_7/30/90 via REGR_SLOPE / REGR_R2 - the un-eliminated live scan on
the happy path.
Drop the six REGR_* columns from both narrow queries and compose the
windowed slope / r2 (and the population sd via composeRegression) from
the per-bucket accumulators stored in measurement_rollups. The 90-day
DAY buckets are loaded once (already loaded in comprehensive; one bounded
findMany added in the slim slice), source-collapsed to the canonical
source, then summed over the day-aligned 7/30/90-day sub-windows. A
window whose buckets lack accumulators (pre-backfill) composes to null,
the same shape the empty REGR_* FILTER returned, so a transitional read
never returns a wrong partial regression.
median (PERCENTILE_CONT) and the exact |z|>2 anomaly count stay on the
minimal live query - both are non-composable across buckets. The cold
coverage-miss path keeps the live REGR_* regression. Annotate
slope_source:"rollup" + median_source/anomaly_source:"live" so a
regression to the heavy path stays observable (no-silent-cap rule).
Make the physiological return relapse-aware (the start of the final
sustained in-band run, not the first), so a vital that relapsed and
recovered again reports its true recovery. Add an optional sleep-context
sub-line (baseline vs episode asleep-minutes), withheld on thin data and
never feeding the gap median.
Adds a neutral sleep observation alongside the per-episode recovery gap: the
user's usual asleep minutes over their well baseline vs the mean over the
episode window, narrated as 'you slept more/less than usual' when the two
diverge. It is an observation only — never a recovery claim, never a return
track, and never part of the headline recovery-gap number.
Per-night asleep totals come from the canonical night reconstruction over the
baseline and episode windows, so a multi-source night (two writers for one
sleep) is counted once, matching every other sleep surface. The observation
withholds entirely on thin data — too few baseline or episode nights, or a
delta below a 30-minute floor — so day-to-day jitter is never narrated as a
finding.
Additive optional DTO field; no schema, no migration. Two locale keys across
the six bundles. Regenerate the OpenAPI contract from the Zod source.
Windowed slope, r2 and population sd still ran a live REGR_*/STDDEV_POP
query over the raw measurements partition on every warm-rollup request,
because the per-bucket mean cannot reconstruct an OLS regression over the
raw points in a window. The per-bucket sd/slope/r2 columns are a
within-bucket fold and do not compose linearly across DAY buckets.
Add the four regression-accumulator sums Postgres folds for REGR_* /
STDDEV_POP - sum_x, sum_xy, sum_xx, sum_yy - per (user, type,
granularity, bucketStart, source) over the same canonical-source rows
the live query reads, with the x-axis in epoch days. With n = count and
sum_y = mean*count already stored, all six terms are additive across
buckets, so a 7/30/90-day slope/r2/sd composes from the summed
accumulators with a bit-identical result to the live closed form.
Migration 0190 adds the four nullable columns; nullable keeps it
additive so existing rows read NULL until the boot re-fold refills them
(mirrors the v1.4.39 sum_value rollout). The populator emits the four
SUM terms inside the existing aggregate (no extra round-trip) and
persistRollupRows writes them. composeRegression / composeWindowedRegression
evaluate the closed form from already-collapsed buckets and return a full
miss when any in-window bucket lacks accumulators, so the reader falls
back to live SQL rather than composing a partial window.
Default the built, data-gated sleep, steps and glucose dashboard tiles to
visible (each self-hides without data) and surface the Vorsorge chart row,
so shipped capabilities are discoverable on the home dashboard. Seed the
demo with an explicit showcase layout.
The demo account seeds rich sleep, steps, glucose, VO2max and workout
data but inserted no dashboard layout, so the home page rendered the
new-account default and under-sold the product on its primary surface.
Persist an explicit dashboard layout derived from the default with the
seeded-rich strip tiles forced on and the daily-verdict hero enabled, so
a fresh seed renders a full, lived-in dashboard for screenshots.
The sleep, steps and glucose strip tiles shipped default-hidden even
though their modules are on by default and rich data flows in from the
sync integrations, so accounts had to hunt through Settings to see them.
Flip the three strip tiles to tileVisible:true and the vorsorge chart-row
card to visible:true, matching the VO2max precedent: each tile self-gates
on having data, so empty accounts keep a clean dashboard. Chart rows for
the three metrics stay off — those charts live on the /insights sub-pages.
totalBodyWater, boneMass and oxygenSaturation stay hidden: they carry as
catalogue ids for the iOS layout round-trip only and have no web render
path, so a flip would surface nothing. Note that at the splice point.
Update the silhouette skeleton and layout default tests for the new
counts.
A vital or symptom curve that settles, re-deviates mid-episode, and settles
again now reports the FINAL sustained in-band run as the recovery return rather
than the first. The earlier first-return rule locked onto an interim settle and
under-reported the gap whenever a relapse followed. The change is monotone — a
last return is always at or after the first — so no existing recovery number
shifts; it only extends the gap for episodes that genuinely relapsed.
The anchor-after-first-deviation guard is unchanged: an early in-band run-up is
still never read as a return. The symptom-burden track honours relapse
identically over logged days so the headline median never mixes first- and
last-return semantics.
Adds relapse, multiple-relapse, too-short-final-run, ends-out-of-band, and
symptom-relapse engine fixtures plus a cross-episode integration fixture.
Add per-bucket HR min/max (migration 0188) threaded through batch ingest
and the series reader; make telegram_chat_id unique (0189) with a
defensive pre-clean; make migration 0187 rerun-safe. Capture a numeric
measurement from a Telegram reminder reply (source=TELEGRAM, strict chat
binding, self-cleaning); blood pressure stays satisfy-only pending a
guided two-number prompt.
Fold the illness journal's per-day functional-impact curve (falling back
to linked symptom severity) into the recovery-gap as a return track
against the known-constant healthy baseline. It is adverse by
construction, feeds the qualifying-days floor and the gap median, and
withholds on sparse logging. The card names the driver when symptoms
lead ('…before your logged symptoms eased').
Two related changes to the Telegram bot's data path.
One chat ⇄ one account. `User.telegramChatId` is now UNIQUE (migration
0189, defensive pre-clean of any pre-existing duplicate binding). The
inbound webhook resolves the user from the chat id, so a chat shared
across two accounts would route a reply or button tap ambiguously. The
index is over a nullable column, so an unlinked account (null chat id)
never collides — only two accounts binding the same non-null chat id are
rejected.
Numeric measurement capture. A numeric reply to a measurement reminder is
now captured as a Measurement (`source = TELEGRAM`, added to the enum in
0189). On "Erledigt" for a reminder that names a single-value metric, the
bot offers an optional value prompt via force_reply; a numeric reply is
written through the shared `logTelegramMeasurement` helper — same range
guard, canonical unit, cache + rollup refresh as the manual route. The
reply binds to the reminder and user through a TelegramPromptContext keyed
on the prompt message id (the same strict chat/message binding the
mood-note path uses); the userId is never read from the payload. Comma
decimals are tolerated; non-numeric and out-of-range replies get a calm
hint and write nothing; the reply self-cleans on the ~30-min sweep. Blood
pressure (two values) is out of scope for the single-numeric path and
keeps its satisfy-only flow.
Guard the CREATE TABLE + both CREATE INDEX statements with IF NOT EXISTS so
a re-run on a database that already has the table is a no-op, matching the
guard convention every other recent migration ships with. Semantics are
unchanged — same table, same columns, same indexes.
When the logged symptom curve drives the recovery gap, the card names it
('you felt better N days before your logged symptoms eased') instead of
the vital-driven phrasing. Mirror the new gapDriverType and the return
track's adverse flag into the client DTO, and add the driver-aware copy
across all six locales.
The v1.19.0 hourly heart-rate bucket
(`stats:HKQuantityTypeIdentifierHeartRate:<UTC-hour>`) shipped avg-only:
one PULSE row carrying the hour's mean bpm. Add the hour's min/max so a
client can render the intra-hour range without re-uploading raw samples.
- Two nullable columns on `measurements` (`value_min` / `value_max`),
migration 0188. Set only on a well-formed hourly HR bucket row; null for
per-sample readings, per-day cumulative totals, and manual entries.
- Batch ingest threads the spread through the same unit conversion as the
average and persists / overwrites it alongside `value` on the bucket
row, pinning it null everywhere else.
- The series reader returns the spread for `kind=pulse` (null on a
per-sample PULSE row); the OpenAPI Zod contract documents both fields on
the batch entry and the series point. The avg path is unchanged.
The recovery-gap engine reasoned only over passive vitals and ignored
the journal's own per-day symptom series. Add the functional-impact
(symptom-burden) return track: read functionalImpact per episode day
(falling back to the strongest linked symptom severity) and treat it as
one more return against the known-constant baseline (healthy = 0).
The symptom track needs none of the MAD-band / contamination machinery
the passive vitals carry — the band is the constant 0 — so it is more
reliable to fold in than any passive vital. It is adverse by
construction, so it contributes to adverseCoverageDays and the headline
gap median, giving the qualifying-days floor a genuinely
illness-relevant contributor where a weight-only episode had none.
Stability is logged-days-only: a sparse journal that logs impact-0 once
and stops withholds the return rather than fabricating an all-clear from
the absence of further logs. gapDriverType names the dominant return
track, with the symptom curve winning ties as the most illness-specific
signal.
A multi-year All range was silently truncated to the most recent year by
the DAY-only 365-row cap. Route windows wider than the cap through a new
tiered rollup reader (WEEK/MONTH by span, finer on coverage miss,
uncapped by tier choice); narrow windows stay byte-identical. No silent
truncation — a coverage miss logs and falls through to the daily path.
The chart still requests aggregate=daily for windows over 7 days, but
the server now steps the rollup tier up to WEEK / MONTH for ranges past
the DAY cap and returns whole-history coverage. Update the stale
comments that claimed the "All" range was still capped at ~365 daily
buckets; the client fold + bucketTimeSeries downsample the coarse
buckets to the visible range as before.
A multi-year "All" range read its series through the DAY-only daily
reader, which caps the result at BUCKET_CAP.daily (365) buckets. The
older history was silently dropped — the range collapsed to roughly the
most recent year instead of the whole span.
Add readTieredRollupSeries, which reads the bucket tier matching the
charts own display downsampler (WEEK for 1-2 years, MONTH beyond) and
returns whole-history coverage at that tier. readDailySeries now routes
any window wider than the DAY cap through it, falling finer on a coverage
miss and logging the miss rather than truncating silently. Short and
normal ranges never enter the branch, so the common path is unchanged.
Surface the captured Oura resilience band (limited…exceptional) as a calm
tile on the recovery surface, with a quiet above/in-line/below-average cue
and a learning state while the series is sparse. Server-authoritative read
of the existing summaries slice; neutral card chrome.
Correct the sleep-rhythm peer-card comment to state plainly that each
card reads its own window (the captions disclose the per-card night
count); no computation changed. Remove the dead streaming.startedAt and
streaming.reasoning state left after the thinking-disclosure removal.
v1.19.0 ingests Oura's daily resilience level end-to-end (RESILIENCE,
ordinal-encoded limited=1 … exceptional=5) but no surface rendered it.
Add a calm resilience tile to the recovery page: it names the latest
band, gives a one-line description, and a quiet "vs the recent average"
cue once a few days exist, with a learning note while the series is
still sparse. Resilience is a categorical band, not a continuous score,
so it reads as a dedicated readout rather than a misleading line chart.
Server-authoritative — the tile reads the stored summaries slice (latest
+ mean, both computed server-side) and never re-derives the value. Calm
posture: one neutral card, no green-when-good / red-when-low tint. The
band label and trend cue carry the read in muted text only.
Self-gating: renders nothing without a reading, and counts toward the
recovery-page data gate so an Oura-only account no longer sees the empty
note. Copy lands in all six locales.
The thinking disclosure was removed, so the coach streaming bubble no
longer reads startedAt or reasoning. Both fields stayed populated by the
send hook but nothing consumed them. Remove them from
CoachStreamingMessage and every writer, accept the additive reasoning
SSE frame as an explicit no-op so the wire type stays recognised, and
update the thread-test fixture. Streaming is otherwise unchanged.
The comment claimed the average-per-night, sleep-debt, and chronotype
cards share one window and cannot disagree. They each read their own
sub-window — the average over the full scorable span, sleep-debt over
its trailing 14 nights, chronotype over CHRONOTYPE_WINDOW_NIGHTS — so
the figures are each internally honest rather than identical, and every
card's caption discloses its own nightsCounted. State that plainly; no
computation changes.
The drawer's Conversations button now hands off to
/coach?view=conversations so the full page opens the conversation list on
arrival; widen the navigation assertion to accept the query string.
Add averagePerNight to the SleepRhythmResource OpenAPI schema so the new
average-per-night card's field is part of the iOS-consumed contract (the
endpoint already returned it; the Zod source had not been updated, so the
in-sync check could not catch the omission).
Tag each vital return with whether it deviated in the illness-adverse
direction and tally only adverse returns when naming the recovery driver,
so a neutral move (e.g. weight drift) can no longer be surfaced as the
vital someone recovered ahead of.
Tighten the cross-episode typical-gap floor to count only active days
carrying an adverse-direction reading (a neutral weight-only episode can
no longer clear it), via a single shared adverse-direction predicate.
Surface the most-frequent returning vital so the card names the driver
('…before your resting heart rate settles') instead of a bare line.
The cross-episode summary already computes per-vital physiological returns
but the card surfaced only one bare line. Thread the dominant returning
vital through the aggregate and the DTO so the copy reads 'you tend to feel
better before your resting heart rate settles' instead of a context-free
gap figure.
The aggregate tallies which vital produced a real return across the
episodes that qualified for the typical-gap median and reports the most
frequent one (deterministic alphabetical tiebreak); the driver is named
only when a gap actually clears the magnitude gate. Zero new compute — the
returns were already in hand. Adds the localized copy in all six locales.
The cross-episode typical-gap median counts an episode toward its
qualifying-days floor via the per-episode coverage `historyDays`, which
counts a day if ANY banded vital has a reading. A WEIGHT-only episode (no
illness-adverse direction) could therefore clear the floor and feed the
median despite carrying no illness-relevant signal, weakening the honest-
sample claim.
Compute a distinct-adverse-day count in the engine — days with at least
one notable reading in the metric's adverse direction — and feed that as
the qualifying-days floor instead. Reuses the engine's single adverse-
direction predicate (now exported as `isAdverseDeviation`); a neutral-
direction vital never counts a day. The insufficient arm contributes 0,
which is correct since its gap is null and never reaches the median.
S1: add an average-per-night sleep card as a third peer beside debt and
chronotype, computed by a shared helper over the same scorable-night set
so the three cannot disagree; grid widens to three columns. S2: restore
full saturation on the two missed mood surfaces (calendar heatmap and the
time-of-day chart), keeping the quiet tint only for no-data cells.
The v1.19.0 saturation fix missed two surfaces: the mood-by-time-of-day
bars and the mood calendar still rendered populated values at 0.55
opacity, reading as a dimmed glitch. Drop the dimming on populated bars
and cells (and the legend swatches); keep the quiet tint only on the
no-data empty state.
Compute the mean asleep minutes per night from the same canonical
per-night data the sleep-debt and chronotype cards read, exposed as a
new averagePerNight field on the sleep-rhythm DTO. Surface it as a third
peer card with identical chrome, so the three tiles sit side by side.
Calm partial state under a four-night floor; localised in six locales.
C1: /coach defaults to the new-chat hero, auto-resuming only on an unread
coach-initiated message. C2: always-visible Conversations + New chat
toolbar. C3: replace the thinking disclosure with the typing-dots into
word-fade streaming animation. C5: the drawer Conversations handoff opens
the history list on arrival so the full view is never a blank pane.
S3: rename the settings nav entry label Layout to Dashboard across all
six locales (route unchanged). S4: fix the main sidebar spine order to
Medications, Vorsorge, Laborwerte, Krankheit, Insights, Coach,
Achievements, with module gating preserved.
Reverse the resume-most-recent entry default: opening the Coach now lands
on the new-chat hero. The only exception is an unread coach-initiated
message (nudge-status `unread`), which auto-opens the most-recent thread so
the user sees what the Coach wrote. The `?c=<id>` deep-link is unchanged.
Add a dedicated, always-visible Conversations button (plus a New chat
button) to the page surface so reaching past conversations no longer hides
in the composer menu. The drawer's Conversations handoff now routes to
`/coach?view=conversations`, which opens the history drawer on arrival, so
the side-view to full-view path always shows the list instead of a blank
pane.
Replace the elapsed-time thinking disclosure with the classic typing
indicator: while a turn is still thinking with no prose, the assistant
bubble shows the pulsing typing dots, then swaps to the streamed
word-by-word prose once tokens land. Drop the now-unused reasoning/
startedAt bubble props and the orphaned disclosure component.
Below the dashboard/measurements/mood/cycle head, the entries now follow one
fixed sequence: Medications, Vorsorge, Labs, Illness, Insights, Coach,
Achievements. A disabled module drops out; the rest keep this order.
Send the full health snapshot only on the first conversation turn (or
when the verbatim history window starts eliding turns); cheap follow-ups
ride the transcript with a one-line grounding pointer. Drops follow-up
snapshot prefix from ~6,150 to ~50 tokens while preserving grounding and
cross-metric correlation.
The full SNAPSHOT block (every cluster plus labs, illness, cycle and the
reference-grounding lines) was interpolated into the prompt on every turn,
so a one-word follow-up re-paid the entire grounding prefix. Ship it only on
the first turn, or once the history window starts eliding turns; on a cheap
follow-up send a short pointer back to the snapshot the model already received
earlier in the same conversation. The prior turns ride the transcript, so
grounding and cross-metric correlation are preserved while a follow-up turn
sheds about 6,000 tokens. Record snapshotSent and promptChars on the reply
annotation so the cut is observable.
The left-nav entry read "Layout" but the editor it opens is the dashboard
layout home. Rename the label across all six locales; the route stays
/settings/layout.
The canonical rotation registry (src/lib/crypto/encrypted-columns.ts) is the
source of truth its guard test cross-checks; add EcgRecording.waveformEncrypted
there (the script already references it) and drop the redundant parallel guard.
Update the sleep/rhythm route test to the v1.19.0 rolling 5-night balance
(nightsCounted 5, debt 300) instead of the old 14-night cumulative sum.
The insights provider-chain integration test mocks @/lib/insights/features but
omitted BRIEFING_FEATURE_WINDOW_DAYS, which the generate route now reads, so the
route threw on access. Add it to the mock.
The v1.19.0 EcgRecording.waveformEncrypted Bytes column was not added to the
key-rotation script, so an operator rotation would strand ECG waveforms on the
legacy key and dropping it would make them permanently undecryptable — the same
class as the v1.18.1 clinical-spine gap. Wire the column into the rotation and
add a schema-driven coverage test that fails CI if any encrypted Bytes column
is ever left out again.
The Telegram reminder flow confirms a dismissed measurement with
telegram.measurementDone; the parallel measurementDismissed string was added
but never wired, tripping the orphan-key i18n guard. Remove it from all
locales.
The chronotype card carried its own section heading on top of the title it
already renders, so 'Chronotype' appeared twice and the tile did not match
its sibling. Drop the standalone heading and bring the card to the same
shape as the sleep-debt tile: identical chrome and a bold primary value with
one secondary line. The chronotype band is the primary value at the same
scale the debt tile uses for its hours figure, with the natural sleep-
midpoint clock as the companion line.
The cumulative model summed every nightly deficit over a 14-night window
with no credit for catch-up sleep, so a sleeper a modest amount short on
most nights carried hours of standing debt that never recovered while the
pattern held. Replace it with a running balance over a 5-night window:
short nights add their capped shortfall, nights above need pay the balance
down at a sub-linear recovery rate, floored at zero. The headline now
reflects what is still owed tonight and shrinks as the sleeper catches up.
Extend the self-cleaning Telegram bot beyond medication. A mood reminder
now carries a 1-5 inline keyboard plus a "Note..." force-reply and a
"Later" action; a tap logs a MoodEntry (source TELEGRAM, server-side,
user resolved from the linked chat, idempotent on a per-tap externalId)
and confirms with a calm callback toast, no chat echo. An optional
free-text note attaches to that entry via the prompt-context correlation.
A measurement reminder carries "Done" (satisfies the cadence) and
"Later" (pushes nextDueAt forward) — completion and snooze only, no value
capture.
Prompts and replies self-clean in ~30 minutes: the window drops from 1h
and moves to a shared helper, the sender schedules an unanswered prompt's
deletion at send time, and the existing 15-min sweep also prunes expired
prompt-context rows. The webhook keeps the shared-secret + timingSafeEqual
+ rate-limit-before-secret posture; every write callback authenticates via
the stored chat binding. The Telegram chat id is added to the egress
redaction denylist.
Add TelegramPromptContext (migration 0187) mapping a bot force-reply
prompt to the entry a free-text reply attaches to, keyed on
(chatId, promptMsgId) and scoped to the owning user. MoodEntry.source
already accepts the free-text TELEGRAM value, so no enum change is
needed.
The notifications page stacked a breadcrumb strip directly above its h1,
so the page read as two headings (the title followed by a channels line).
Remove the breadcrumb so the page starts with a single heading, matching
every other page. Drop the now-unused breadcrumb i18n keys.
The notification preferences build event labels dynamically as
notifications.event<PascalCase>, so the static call-site coverage test
never enumerated them and three event types rendered raw keys:
MEASUREMENT_REMINDER, MEDICATION_INTAKE_SYNC and COACH_NUDGE. Add label
and description for each across all six locale bundles.
A range tab wider than the dashboard's batched series slice (90 / All on a
~30-day batch) re-read the narrow slice instead of fetching the requested
window, so the chart kept showing ~30 days. Thread the batch's coverage
day-span to the chart and read the preloaded slice only when the requested
window fits within it; a wider tab falls through to a real fetch. Also raise
the All-range request window past the old 365-day cap so longer histories
are no longer silently truncated client-side.
Round out the public demo so every feature is explorable:
- Glucose: a 14-day Nightscout-style CGM stream (a reading every 15 min)
alongside the existing spot panel, so the glucose surfaces render fully.
Tagged RANDOM to satisfy the glucose-context CHECK.
- Labs: a second older panel plus qualitative rows (negativ / nicht
nachweisbar) so the list, detail, sort and delete all have content and a
per-analyte trend.
- Illness journal: a resolved cold with a day-by-day symptom curve and an
active chronic condition carrying a recent flare.
- Cycle: opt the demo account in and seed five observed cycles with biphasic
basal body temperature, period flow, fertile-window mucus and OPK, per-day
symptoms, BODY_TEMPERATURE dual-writes, and a cached forecast.
Cleanup extended to the new tables in FK-safe order; the seed stays
idempotent across re-runs.
Move the non-medical disclaimer to the cycle-tracking on-ramp and drop it
from the repeating in-feature surfaces (calendar tab, predictions panel);
delete the now-unused disclaimer component. Align the left ring column with
the tab content rather than the tab strip. Raise the settings subheading
sizes to the standard card-description scale. Reflow the reminder rows so a
wrapping label no longer collides with its button at narrow widths. Unify
the wheel and phase-education tile padding. Cite the symptothermal
double-check primary source under the second fertility sign.
Oura's daily_resilience collection reports a categorical level — limited,
adequate, solid, strong, exceptional. Encode it ordinally (limited=1 …
exceptional=5) into the numeric measurement value so it fits the existing
model with no new categorical column: add the RESILIENCE measurement type
(migration 0186), fetch and map daily_resilience in the Oura sync, and skip
an unknown or missing level rather than coercing it to zero.
RESILIENCE_LEVELS in the Oura client is the single source of truth for the
encoding. The day-keyed externalId keeps a re-sync idempotent so a re-scored
day overwrites in place. Extend the measurement-type label, unit, range,
category, chart-token, and PR-direction maps; add the metric label across all
six locales.
The 0–14d raw band folded same-day readings on a UTC calendar key, so a
late-evening reading slipped onto the wrong day for non-UTC users and
across DST transitions. Key it on userDayKey(tz) instead and make the
timezone a required option so a caller cannot silently fall back to UTC.
The coarser DAY/WEEK/MONTH/YEAR bands stay UTC-anchored on purpose: the
rollup writer and the live aggregator both bucket on the UTC calendar
day and are parity-locked for the read-swap fallback. Document the split
and thread the resolved user timezone from the coach snapshot caller.
Surface the load-bearing first-run vars in docker-compose.yml and the README
quick-start: secret generation, the SESSION_COOKIE_SECURE=false LAN note, the
single persistent volume, and a /api/version verify step. Add an Unraid
Community Applications template plus a NAS walkthrough covering the external
Postgres setup, and a Portainer app template referencing the compose stack.
Drop vendor names from four metric definition bodies (wrist temperature,
falls, six-minute walk, sleep breathing disturbances) and frame the
six-minute-walk and breathing-index readings as screening proxies rather
than the supervised clinical test. Add a primary citation to the cardio
fitness body: each one-MET higher fitness associates with roughly
10-25% lower mortality (AHA, Circulation 2016). Propagated across all
six locales.
Apple Watch Blood-Oxygen samples OXYGEN_SATURATION through the day and
overnight. The readings map to latest aggregation and sit in neither the
cumulative SUM drain nor the daily-mean drain, so every sample piled up raw
forever. Fold out-of-window per-sample SpO2 rows to one daily-mean stats:
row, soft-delete the raw rows, and preserve the daily min/max on the rollup
DAY bucket via a pre-fold recompute — the lowest sample is the overnight
desaturation nadir, the clinically meaningful figure the mean would hide.
Generalise the PULSE min/max preservation to a min/max-preserving set so
the pre-fold recompute and the skipped post-fold recompute cover both PULSE
and SpO2; resting-HR derivation stays PULSE-only. RESPIRATORY_RATE is left
to the existing daily-mean drain it already belongs to.
The comprehensive briefing's bounded feature read (extractFeatures) was
paid more than once where it should not be: the on-demand route and the
shared generator each run a downgrade ladder that re-reads with identical
arguments, and a nightly warm has no shared object across consumers in a
single tick.
Add a request/run-scoped feature cache: withFeatureCacheScope opens an
AsyncLocalStorage store, getCachedFeatures memoises the first compute per
(userId, day, includeRaw, sinceDays), gated on a cheap salient-input
fingerprint so a mid-scope ingest recomputes and a probe failure fails
safe. Outside a scope it is a pure passthrough. Wired into the
comprehensive generator, the on-demand POST handler, and the nightly
per-candidate loop + forceWarmUser. Coach keeps its own snapshot.
The slow-moving status input fingerprint covered the card's own metric
rows plus mood and height, but not the FDR discovery channels that also
feed the prompt via getRelevantCorrelationsForMetric. A correlation
newly surfacing off a behaviour channel (e.g. daylight -> next-day
weight) left the card's own rows untouched, so the input gate re-stamped
the stale assessment and the discovery never reached the prose.
computeStatusInputFingerprint gains includeCorrelationChannels: it widens
the same grouped probe by the discovery behaviour + outcome channels
(deduped, mood excluded as it is mood-entry backed), so a channel change
flips the hash. Wired into weight-status, the one caller that folds a
relations block.
The Heart sync recorded only the device's atrial-fibrillation verdict per
on-device ECG. Fetch each recording's full waveform via the Heart get
endpoint and store the raw micro-volt samples so a real trace can later be
rendered.
The sample array is sensitive health data: it is held AES-256-GCM encrypted
at rest in the new EcgRecording.waveformEncrypted Bytes column (fail-closed
crypto), with the sampling frequency, sample count, duration, lead, average
heart rate, and AFib verdict kept as plaintext descriptors. A re-sync upserts
on (userId, source, externalRecordingId) so a recording is never duplicated;
a waveform fetch failure never fails the verdict ingest. Rendering UI is a
follow-up.
Define the go-forward aggregated heart-rate upload shape so clients can
stop posting one Measurement per raw HealthKit sample. A heart-rate row
keyed `stats:HKQuantityTypeIdentifierHeartRate:<bucket-start-hour>`
carries the hour's average bpm as a single PULSE row and rides the
existing `stats:` overwrite path: a re-post within the hour overwrites
the value instead of duplicating, identical to the per-day cumulative
rows. The hour boundary is UTC so the key is timezone-stable.
The batch route rejects a row that targets the heart-rate bucket prefix
with a malformed hour suffix as a per-entry skip
(`malformed_hr_bucket_id`) rather than persisting an un-overwriteable
row; well-formed buckets, per-sample uuid ingest, and the per-day
cumulative overwrite path are all unchanged.
Average-only for now — the measurement row carries a single value
column, so hourly min/max would need a schema change and is left as a
follow-up. Documents all three externalId shapes plus the `updated`
status in the OpenAPI contract.
The medication detail clients computed the supply headline in the browser
via summariseSupply, risking web/iOS drift and pulling the shared math
into the client bundle. The GET /api/medications/[id]/inventory route now
computes the canonical SupplySummary server-side through summariseSupply
(the single source of truth) and ships it in the DTO; the Bestand and
Übersicht surfaces render the ready figures. Adds the MedicationSupplySummary
schema to the inventory list response and regenerates the OpenAPI contract.
The muted mood treatment (opacity-55 on the tag bars, fillOpacity 0.55 on the
distribution and weekday charts) read as a rendering glitch rather than a
deliberate calm tint. Drop the dimming so every mood bar paints at full
saturation in its level hue; the guard test now asserts the dimming stays gone.
A separate mood colour concept can be revisited later.
summariseSupply is rendered by the medication detail client surfaces, so the
v1.18.11 underflow wide-event it called pulled node:async_hooks into the client
chunk and failed the production build. Split the observability out: the helper
stays a pure clamp, a new pure detectSupplyUnderflow reports a would-be negative
pool, and the low-stock job emits the medication.inventory.underflow event at
its request-scoped seam where annotate can actually fire.
The codex vision gate accepted any gpt-5 prefix, which included the gpt-5-codex
specialist slugs that the surrounding contract documents as text-only. Each lab
OCR upload on such a slug burned a doomed image attempt before falling back.
Exclude the -codex slugs from the gate so multimodal gpt-5.x slugs still report
vision while the codex specialists stay on the text path.
The split commits added the directory barrels but left the original
single-file modules in the tree; Node resolves the .ts file ahead of the
directory index, so the barrels were shadowed. Remove the originals so
the split modules are the live source.
The v1.18.11 briefing now folds illness-episode and cycle state into the
prompt via buildBriefingIllnessCycleContext, which resolves module gates
through memoizePerRequest and queries cycleProfile. The generate route test
mocks neither getEvent nor the cycle/illness models, so every case threw once
the new call path ran. Stub the context module to no output and add the
getEvent export to the logging-context mock so these cases stay scoped to the
provider and cache behaviour they assert.
Weight and BMI move on a weekly cadence, but the per-metric generators
rebuilt the full snapshot (bounded findMany + per-series rollup reads +
correlation math) every day and relied on the post-build content-hash gate
to skip only the model call. Add an input gate that runs first: a single
grouped probe over the salient inputs (per-type count + newest reading, plus
mood for weight and the profile height for BMI) fingerprints what the
snapshot would read. On a non-forced run whose fingerprint matches the cached
row, the cached assessment is re-stamped under today's key and the whole
build is skipped. A miss falls through to the normal build and the existing
post-build gate.
Consent is checked before any re-date so a revoked-consent user never has
stale text presented as current. Forced runs never gate but still record a
fresh fingerprint for the next day.
The 02:00 status cron now runs generateStatusBatchForUser for every eligible
user: it builds the seven per-metric snapshots once and issues a single
provider call for the metrics still needing the model, fanning each summary
back into the same per-metric cache rows the standalone generators write.
The six later per-metric crons (02:05-02:30) keep their drivers but now hit
the calendar-day cache the batch already filled, so they resolve as cache
reads and only cover the edges — a user discovered after 02:00, a metric the
batch omitted, or a batch-call failure (which the batch falls back per metric).
Reproducibility and the consent/grounding gates are unchanged: the batch runs
through runStatusCompletion, which still pins the reference seed.
Drops the now-unused per-metric general generator and recordRun plumbing.
The warm narrow query and the cold heavy query each referenced the
canonical-source subquery twice — once in the window_stats CTE and once in
the outer body — so the DISTINCT-ON pick over the live 90-day rows ran twice
per rebuild. Extract canonicalMeasurementsCte (the SELECT body without the
alias wrapper), bind it once as WITH cm AS (...), and select FROM cm m from
both window_stats and the outer aggregate. Output is identical; the dominant
self-join runs once.
canonicalMeasurementsFrom now wraps the same body, byte-stable for the
single-reference call sites (pinned by a new parity test).
The daily_cardiovascular_age collection reports a years-unit arterial-age
estimate that maps directly onto the existing VASCULAR_AGE metric, so no
migration is needed. Resilience stays out: it is a categorical level with
no numeric home and would require a new type plus a categorical column.
The Heart List endpoint returns a per-recording AFib verdict from the
ScanWatch family that no sync fetched. Capture each recording as an
IRREGULAR_RHYTHM_NOTIFICATION event row carrying the device's own verdict
in rhythm_classification, mirroring the Apple Health categorical-event
treatment. Reuses the existing enum, column, and constraint, so no
migration. Fires on the measure webhook path, which shares the metrics
scope the Heart List endpoint needs.
The briefing only reached the dashboard as the hero's lowest-priority
verdict, which the severity ladder almost always pre-empted, so a fresh
briefing was invisible above the fold. Lift its own signals (falling back
to key findings) onto the hero as a self-gating strip beneath the verdict
row, reusing the insights briefing's tone-bar row language. Placement
only: the verdict ladder and the briefing content are unchanged.
The dense intra-day retention drain folds out-of-window per-sample PULSE rows
to one daily-mean row to bound the high-frequency ingest. The mean-only fold
erased the two clinically-meaningful daily figures derived from the raw
stream: the resting estimate (20th-percentile of each day's PULSE, the
fallback for users without a native resting series) and the daily min/max held
in the persistent rollup DAY bucket.
Preserve both in the PULSE facet of the fold:
- Recompute the PULSE DAY rollup from the raw rows before the fold, while they
are still live, so the bucket keeps the true intraday min/max/mean; skip the
post-fold recompute that would collapse them.
- For users with no native resting series, mint one derived RESTING_HEART_RATE
row per folded day from the day's 20th percentile, sourced COMPUTED so it
never collides with measured resting rows and reads back as the clean
resting series. Users with a native series are skipped to avoid
double-counting; the probe is memoised per user.
Add a per-day failure boundary so one poisoned day can no longer abort the
whole walk. Surface the derived-resting count through the handler return,
the wide-event annotation, and the worker log.
Add a labs block to the coach snapshot so the assistant can answer
questions about the user's lab results. Each entry is the most recent
reading per biomarker over the last 12 months (capped), carrying the
resolved name, value, unit, reference range, in/out-of-range verdict,
and measured date. Qualitative readings (e.g. negativ) are included
with a neutral range status. Values resolve server-side via the same
helpers the labs API uses, and the system prompt instructs the model to
quote them verbatim and never invent a lab number, so the block stays
grounded. The free-text qualitative result is sanitised before it can
reach the prompt; the encrypted note is never read.
The Coach page opened a blank new chat regardless of where the user came
from. Make it a deep-link target: a `?c=<id>` query param opens that
thread on mount, and absent it the page auto-opens the user's most-recent
conversation. "Most recent" is the `updatedAt desc` head of the rail
list, so the same thread opens on web and mobile.
The shared conversation surface grows two additive props
(`initialConversationId`, `autoOpenMostRecent`) that feed its existing
selection state; the history drawer's selection stays the single source of
truth. The drawer surface is unchanged — both props default off and the
rail list is only fetched when auto-open is requested.
Add a central sanity gate in summariseSupply, the single helper every
supply readout flows through. A corrupt or legacy row with a non-finite
or negative unit count could pool straight into the headline and surface
a nonsensical negative Bestand; the helper now floors the pooled
remaining, total and expired figures at zero and emits a
medication.inventory.underflow wide-event when a raw value would have
gone negative, so the underlying data defect stays observable.
closes#31
Serve achievements via SWR and precompute Berlin day-keys to kill the per-row
Intl cost; switch dashboard/summary to cachedSwr; lazy-load three charts that
bypassed the dynamic boundary.
Bound the comprehensive briefing's feature read to 400 days while keeping
all-time extremes honest via a full-history groupBy, and append module-gated
illness and cycle context to the briefing prompt.
Achievements: the bucket was 60 s hard-TTL while the app-wide unlock
notifier polls every 2 minutes, so every poll landed on an expired
entry and fired a cold ~400 ms rebuild in the foreground. Add a
10-minute stale-while-revalidate window and read the route through
cachedSwr; the route already persists pendingUnlocks outside the cached
factory, so a stale serve never skips an unlock write.
Achievements builder: the vitals array ran through several full-array
passes, each re-deriving a Berlin day-key (and hour) per row via
Intl.DateTimeFormat. Compute each row's day-key once and thread the
parallel array into the green-day, weekly-consistency, engagement, and
hidden passes; the hidden pass reuses a single precomputed hour array.
Results are byte-identical to the per-row derivation (parity test added).
dashboard/summary: read through cachedSwr so the iOS-polled endpoint
serves its existing 1 h stale window instead of re-paying the cold
rebuild every 60 s on the plain cached path.
Charts: route the three recharts consumers that bypassed the shared
next/dynamic boundary (lab biomarker chart, sleep stage stacked bar,
recommendation rationale charts) through dynamic imports with matching
loading skeletons, taking recharts off the first-load JS for
/labs/[id], /insights/sleep, and the insights recommendations surface.
Bound the comprehensive briefing's feature read to a 400-day window instead of
walking the full measurement history on every generation. All-time min/max/avg
are sourced from one grouped aggregation over the whole history, so the
allTime* fields stay honest while the unbounded row materialisation is gone;
the oversize downgrade ladder remains as the safety net.
Fold illness-episode and cycle state into the briefing prompt, reusing the
server-authoritative builders the Coach already assembles. Both are
module-gated and cost nothing for users without them, so a mid-illness or
mid-cycle user finally gets context-aware prose.
Surface non-ok geo-lookup responses as a warning instead of swallowing them,
make the lookup provider/URL override explicit, and whitelist TRUST_CF_CONNECTING_IP
+ the geo env in compose so the real visitor IP is stored behind Cloudflare.
Fold image input into the Codex client (input_image over the ChatGPT-OAuth
responses endpoint, no API key); make the codex vision gate model-aware; resolve
labs OCR to vision mode for multimodal codex with safe text/local-OCR fallback.
The ipwho.is default rejected server-side egress with a 403 on the free
plan, which the resolver swallowed as a clean miss, so the admin
login/audit overview rendered "—" for every session. Switch the default
online provider to ip-api.com (the parser already accepts its
status/countryCode shape) and surface a non-ok HTTP status (403/429/5xx)
as a wide-event warning instead of returning null silently, so a future
provider rejection is visible.
Whitelist TRUST_CF_CONNECTING_IP, IP_GEO_LOOKUP_URL and
IP_GEO_LOOKUP_DISABLED in the compose environment block and document them
in .env.production.example so a Cloudflare-fronted deployment can store
the real visitor IP rather than the edge address.
Image input rides the ChatGPT-OAuth codex/responses endpoint via the
input_image content block (protocol spec 2b) — no vision API key. The
codex client folds params.images into input_image blocks alongside the
input_text; text-only calls keep the prior byte-identical wire. The
vision gate now resolves codex capability against the working slug
(GPT-5.x / gpt-4o-class multimodal) instead of a blanket false, with the
OCR capability resolver feeding the codex slug rather than the user's
aiModel. A non-multimodal slug or an image-block rejection degrades to a
clean extract failure / local-OCR text mode.
Rework the Coach page to a ChatGPT-style IA: drop the top header, make the
composer the control hub (+ actions, settings, mic), open history as a left
slide-in drawer, and keep the composer one constant width across new-chat and
active conversation. The FAB drawer surface is untouched.
Drop the dedicated page's top header bar and the conversation-history
strip. The composer now carries the controls: a leading actions menu
(new chat + open conversations) and a settings deep-link beside the mic
and send. Conversation history opens as a left slide-in drawer that works
on phone and desktop.
Pin the composer to one centred, max-width-capped column shared with the
thread so its width stays constant from the new-chat hero into an active
conversation. The drawer surface keeps its own header and body chrome
unchanged.
Data-loss fix, measurement consolidation, AI grounding, full-page Coach, local OCR, Withings sleep vitals, and broad polish. Six upfront audits + diff review; CI green.
The 15s default cancelled the live-aggregate analytics fallback on a coverage
miss over a large account (Postgres 57014), surfacing as a 500. Sixty seconds
still bounds a runaway query well before it can starve the pool while leaving
headroom for the heaviest legitimate read.
Fix encrypted day-log data loss and the stalled cumulative-measurement drain;
capture Withings sleep vitals and Oura VO2max; harden AI grounding across the
briefing, Coach, narrative, and status cards; full-page Coach with history and a
working mic; on-device lab OCR; preventive-care, labs, and settings polish; BP
warm/cold parity, a statement timeout, and a covering measurements index.
Keep the full-page Coach composer clear of the mobile bottom nav and safe-area;
wrap long More-hub labels instead of ellipsising; raise lab-history row controls
to 44px on touch; make the biomarker-detail Add button icon-only on small phones.
Size the full-page Coach surface to clear the mobile BottomNav band plus the
iOS safe area so the docked composer and Stop control never sit under the nav;
desktop full-bleed height is unchanged (the nav is mobile-only).
Stop truncating the More-hub labels: shrink the label type and allow a
two-line clamp so long words (German Benachrichtigungen, and the long es/fr/it
labels) read in full instead of ellipsising.
Raise the lab reading-history edit/delete controls to 44px on touch
(compact on desktop), and drop the lab-detail Add button to icon-only on the
narrowest phones.
Enforce apiFetch in src/lib/queries and migrate two hand-rolled hooks; thread a
DST-safe day-key into readiness and mood-narratives; lazy-load qrcode; add a
shared day-millisecond constant.
The QR is only rendered after a share link is created, so import qrcode
lazily inside the effect instead of statically at module top. The library
no longer ships in the initial settings bundle.
The readiness same-day grouping and the mood-streak anchor/walk both
sliced the UTC instant for their day key, mis-bucketing a late-evening
reading for a logger east or west of UTC. Thread an optional tz (default
Europe/Berlin) through both paths and key the buckets via userDayKey so
the day boundary follows the user's wall clock. Centralise the inline
day-in-ms literal in a shared time-constants module and use it here.
Add src/lib/queries/ to the api-fetch-required ENFORCED_ROOTS so the
rule matches its own header comment, then route the two flagged query
hooks through apiGet instead of hand-rolling the envelope unwrap.
Ground daily-briefing numbers against the signal features, guard Coach replies
against fabricated doses and risk scores, gate the period narrative, feed the six
status cards a precomputed signal block, add SSE cancellation, and cap the Coach
snapshot read.
The safety-contract matrix parity test checks rule bodies are non-empty;
this asserts the resolved getCoachSystemPrompt(locale) string carries the
grounding marker and the GLP-1 dose-safety contract for all six UI
locales, closing the gap where a non-EN/DE locale could silently fall
back to a weaker EN-plus-footer body.
BP, pulse, weight, BMI, mood and medication-compliance still asked the
model to derive the recent-vs-baseline comparison from raw buckets — the
exact arithmetic small models do unreliably, and the pattern the
base-system prompt already advertises as precomputed.
Build the SIGNAL block (current / baseline / signed delta / normal-swing
verdict) with buildMetricSignal from each card's graded series and inject
it into the snapshot; BP carries two signals, compliance one per
medication. Each card's prompt section now leads from the signal block
and is told to state, not recompute. The server-computed compliance %
is unchanged.
The single multi-type window read had no take bound, so a lastYear /
allTime scope over a high-frequency type (PULSE / glucose are
200k-row-class) loaded a year of rows just to fold them into the ~21
daily + ~10 weekly buckets the prompt shows. Read newest-first with a
6000-row cap and restore ascending order; the coarse MONTH/YEAR tail
still comes from the rollup tier, so deep history survives.
The period narrative was written straight to storage with no
post-generation check and no corrective retry, unlike Comprehensive
Insights. A populated context could still yield causal phrasing or a
number absent from the context block.
Add a validator that flags causal language (EN/DE) and any number not in
the structured context; on a trip, retry once with a corrective suffix,
then write nothing (the last good row stays) rather than persist an
ungrounded or causal story.
Apply the outbound safety screen to the assembled Coach reply before
persistence and streaming, with a per-turn audit row on a block.
createSseStream gains a cancel handler + a shared AbortSignal: a client
disconnect now flips the signal so the token-emit loop stops enqueuing
into a closed connection and any provider fetch wired to the signal is
torn down. The chat emit loop honours it between frames.
detectRefusal guards only the inbound user message; the model's reply
was streamed verbatim. A dose-prescription or a fabricated clinical risk
score that slipped past the system-prompt contracts reached the user
unfiltered.
Add a deterministic outbound screen on the assembled reply (which the
route already buffers before streaming): tight dose-change patterns plus
risk-score patterns, EN/DE. On a trip the turn is replaced with a calm,
grounded fallback and any reminder suggestion / key-value provenance from
the discarded reply is dropped.
The daily-briefing block (paragraph / signalsOfDay / keyFindings) was
schema-shaped but never number-checked the way recommendations[] are: a
restated figure could drift from the snapshot with no server rejection.
Add a grounding cross-check that asserts every number the briefing prose
states traces to a features.signalsOfDay figure (latest / avg / delta)
within a rounding tolerance. Wire it into both generation sites with one
corrective retry, then strip the briefing rather than persist a
fabricated number. Also gate the higher-temperature paragraph reroll on
the same check so a phrasing reroll cannot drift a figure.
Align the Vorsorge settings card to the canonical pattern, neutralize the danger
zone card, mark the bug-report feature deprecated, make ipwho.is the default geo
resolver, and keep the mobile More-hub labels on one line.
Trim the longest label ('Benachrichtigungen') with truncate (nowrap +
ellipsis) instead of break-words so every More-hub tile reads on a
single line on a phone, matching its neighbours. The full label stays
discoverable via the title attribute; the desktop layout is untouched.
Add a discreet 'Deprecated' badge and a short note ('will be removed in
a future version') to the bug-report settings entry. The functionality
stays intact; this only signals the upcoming removal across all locales.
Render the danger-zone section on the shared SettingsCard surface
(bg-card + border-border) instead of the destructive-tinted card edge.
A red card surface reads as alarming rather than premium; the danger
signal now lives only in the destructive delete button and its typed
confirmation dialog, with the AlertTriangle header icon still flagging
intent.
Adopt the standard SettingsCard + SettingsCardHeader contract for the
three Vorsorge settings groups (view, order, manage): bordered card
surface, neutral muted section icon, text-lg title, and a short muted
description, replacing the bare sections with text-sm headings so the
surface reads like every other Settings section. Add a reorder
description string across all locales.
Make the keyless ipwho.is HTTPS lookup the primary resolver for every
self-host. The bundled MaxMind GeoLite2 offline tier becomes an optional
fallback, consulted only when the online lookup misses or when egress is
disabled. A missing offline tier is now the expected baseline rather than
a gap, so the one-shot admin alert fires only when neither resolver can
produce a location.
Capture Withings sleep vitals (HR/respiratory/SpO2/sleep-HRV/score) via the sleep
summary; collapse overlapping BP sources on the cold path for web/iOS parity; add
a STRESS_SCORE source-priority ladder; add a soft-delete-aware measurements index
and cap the all-time cold scan.
Add a covering partial index over (user_id, type, measured_at DESC) WHERE
deleted_at IS NULL (migration 0183). The hot reads all filter deleted_at and
order measured_at DESC, but the base index is ASC and excludes deleted_at, so
every type not served by the two existing partial indexes pays a heap re-check
of deleted_at over its partition. The covering partial removes the re-check for
the DISTINCT-ON-latest, list and series reads.
Cap the cold-fallback all-time aggregate in summaries-slice at a 15-year
window. It sits far beyond any real history, so it never changes a real
all-time count/min/max/mean; it exists only so a coverage-miss cold read can't
trigger an unbounded full-partition scan.
STRESS_SCORE has no source-priority ladder, so metricKeyForType returns null
and the SQL source collapse falls to an alphabetical rank-90 tiebreak. It is
safe today because the only producer is the COMPUTED HRV-derived proxy, but
the instant a device-native producer (Oura daily_stress) is enabled both rows
survive and blend nondeterministically.
Add a stress metric key with an OURA-above-COMPUTED default ladder, wire it
through metricKeyForType and RANKED_TYPES, and surface it in the Sources
settings (all six locales). A guard test asserts every RANKED_TYPES member
resolves to a non-null metric key so a future ranked type can't ship unladdered.
Regenerate the OpenAPI for the additive source-priority key.
The raw BP pulls in the comprehensive aggregator and the bp-in-target live
fallback read every source's row for BLOOD_PRESSURE_SYS/DIA. BP is a
multi-source metric — a Withings reading mirrored into Apple Health is two
rows for one event — so each shared reading was counted twice in the
in-target denominator and twice in the recency-weighted grade. The rollup
branch of the same fast-path reads source-collapsed measurement_rollups, so a
rollup-warm account and a rollup-cold account returned different BP
percentages and grades for identical data — a web/iOS and cold/warm parity
break.
Route both raw pulls through canonicalMeasurementsFrom so the cold/live path
picks the ladder-winning source per (type, day), matching the warm/rollup
path. Per-event rows of the winning source survive the collapse, so the
downstream 5-minute pairing still sees individual readings.
Move the Vorsorge cadence into the metric row, link card headings to filtered
results, tighten the daily-briefing card rhythm, and invalidate the dashboard
snapshot on briefing regenerate and layout save so the Startseite refreshes.
The per-segment sleep sync wrote only the stage timeline; the per-night
summary's vitals — average HR, respiratory rate, SDNN-based HRV, average
SpO2 and the sleep score — were dropped at the fetch. Add a getsummary
fetch that maps each vital to its existing MeasurementType (RESTING_HEART_RATE,
RESPIRATORY_RATE, HEART_RATE_VARIABILITY, OXYGEN_SATURATION, SLEEP_SCORE), so
no migration and no client-contract change: the rows flow to the sync delta
under the WITHINGS source like every other server-side fetch.
Each vital upserts on the stable (userId, type, source, externalId) composite
keyed by the night's session id, so a re-sync overwrites rather than
duplicates. A summary fetch failure is isolated so the stage segments already
written survive. Fractional SpO2 is normalised to percent.
Turn the labs add action into a scan/manual choice; add opt-in client-side local
OCR (tesseract.js) so non-vision providers can scan reports; move biomarker
delete to the detail view; make overview tags compact; surface reference-range
confidence for new biomarkers.
The dashboard hero reads the briefing headline and the layout from the
server snapshot. Regenerating a briefing or saving the dashboard layout
only touched their own caches, so the snapshot kept stale data until its
own window lapsed and the Startseite looked unchanged until a manual
reload. Invalidate the dashboard snapshot on briefing regenerate and on
layout save/reset.
Harmonise the daily-briefing card body rhythm with its sibling insights
cards (space-y-3, matching correlation-card) so it no longer reads taller.
Drop the grey cadence pill from the card heading. The cadence now reads as
part of the metric line, styled like the metric label itself (no shaded
background), via a new MedicationCardHeader metricMeta slot. Nothing else
sits beside the heading.
Link a measurement-linked reminder's heading to the measurements history
pre-filtered to its type, reusing the existing kebab target. Free-text
reminders keep a plain heading.
Remove the inline trash icon from the lab list and tile rows; delete now
lives only on the value detail view, beside Edit, so a stray tap on the
overview cannot remove a biomarker. Make the reference-range badge beside
each analyte heading quieter on the overview (smaller text, tighter
padding, no icon) so it reads as a subtle qualifier instead of competing
with the heading.
Replace the separate, capability-gated scan button with a single Add
menu: "Scan a document" (OCR) or "Add a value manually". The scan entry
appears only when the capability probe reports a usable mode, and routes
to native vision or local in-browser OCR (tesseract.js, lazy-loaded,
deu+eng, images only) per the resolved mode. Add the local-OCR opt-in
toggle to the Labs settings page with honest "less accurate — review
every value" copy.
Surface range confidence prominently on the OCR review screen for new
numeric biomarkers, where the extracted range becomes the catalog range:
flag low range-confidence or a missing range so the user verifies it.
Add an opt-in in-browser OCR path so a user whose provider cannot read
images (ChatGPT-OAuth/Codex, a text-only model) can still scan a lab
report. The capability probe gains a `mode` field: `vision` when the
provider reads the image directly, `text` when local OCR is enabled and
any provider is configured, or null. The extract route grows a text
branch that accepts the in-browser-OCR'd text as JSON and structures it
through the existing schema + review/commit flow; the raw image never
reaches the server in text mode. New `labsLocalOcrEnabled` user column
(migration 0184, default off) gates it, with a dedicated PATCH endpoint.
Preserve undecryptable encrypted day-log columns on unrelated edits (data-loss
fix); give the cumulative drain a per-day error boundary so one colliding day no
longer aborts the walk; apply a per-session statement timeout.
Without a statement timeout a single stuck query or lock wait holds one of
the 20 pool slots indefinitely; enough of them permanently exhaust the pool
and every DB-backed route stops serving. Apply statement_timeout and
idle_in_transaction_session_timeout (default 15s, env-overridable, 0 to
disable) at connection establishment so every pooled session is bounded.
Correct the stale pool-sizing comment: production max_connections is 200,
not 100, and the web and worker containers share the one Postgres.
Make the Coach page full-bleed (no inner panel, no minimize button), keep the
conversation history rail, always render the dictation mic with a disabled
tooltip and error toast when speech is unavailable, and drop the starter chips.
Always render the dictation mic so the affordance stays discoverable. On
a browser without the Web Speech API (or during SSR) render it disabled
with an explanatory tooltip instead of hiding it or leaving a control
that silently does nothing on tap. When a dictation start fails — denied
microphone permission, an insecure-context self-host, or no network for
the recogniser — surface a quiet toast rather than resetting in silence.
Add voiceUnsupported + dictateError strings across all locales.
The cumulative pass lacked the per-day error boundary and adopt-in-place
canonical-noon resolution its sibling drains carry, so the first day whose
local-noon mint collided with the (measured_at, source, sleep_stage) unique
index threw and aborted the entire global walk — leaving years of raw
step/distance rows uncollapsed.
Resolve the canonical daily row by externalId then by the canonical-noon
slot and adopt it in place, retry once on a concurrent-writer P2002, and
route per-day failures through onBucketError so one bad day is stepped over
while every other day still collapses.
Drop the inner bordered card on the dedicated /coach page so the chat
reads as one continuous conversation rather than a window inside a
window, and remove the top "back to the drawer" minimize control (it is
pointless on the full page; the bottom-right FAB drawer remains the
separate compact surface). Also remove the two starter-question
suggestion chips below the new-chat composer, leaving a greeting +
composer hero that matches the reference new-chat layout. History stays
reachable through the existing conversation rail.
A partial-update upsert soft-decrypted the existing encrypted column,
returned null on failure, then re-wrote it — so editing an unrelated
field permanently nulled an undecryptable note or sensitive envelope.
Preserve the existing ciphertext verbatim when the encrypted field is
absent from the update; only re-encrypt on an explicit field write.
Mirrors the v1.18.1 rotation fix.
Coach new-chat surface with streamed answers and a token footer; lab-report OCR
ingestion with a mandatory per-row review; qualitative lab results; labs list
default, sort, and delete; VO2max on the dashboard with Oura ingest; preventive-
care date and due-status polish; faster condition journal; honest report
staleness; dashboard freshness on background sync.
Hard-evict the dashboard snapshot on batch and Withings background sync so the
Startseite reflects synced readings on focus; withdraw the Oura daily_stress
mapping pending source-priority wiring; make OCR in-batch dedup independent of DB
visibility; widen the vision allowlist to Claude 3; null-guard the reasoning
frame; correct OpenAPI and cache-TTL drift.
The hook comment said the server SWR cache is warm for 60 s, but the
snapshot key's fresh TTL is DASHBOARD_REFETCH_INTERVAL_MS + 60_000
(~180 s). Correct both references.
The streaming-doc description listed `model` as a required field of the
`done.usage` envelope, but the runtime CoachUsage type has it optional /
nullable. Align the description and regenerate the OpenAPI contract.
A reasoning frame with no text appended the string "undefined" to the
thinking disclosure. Coalesce evt.text to an empty string on both the
streaming and the trailing-flush paths.
The vision allowlist matched only the 3.5/3.7/4 families, so a
self-hoster pinned to claude-3-opus / claude-3-sonnet / claude-3-haiku
was wrongly blocked from lab OCR though those models read images. Add
the bare claude-3-{opus,sonnet,haiku} prefixes.
The per-row commit loop ran isDuplicate() then create() sequentially, so
two identical rows in one confirmed document only collapsed because row 1
autocommitted before row 2's live query. Track written keys in a
per-request seen-set keyed on analyte+day+value so in-batch dedup no
longer depends on DB visibility. Note the mint-then-create orphan on a
mid-row failure is benign and self-healing, matching the manual path.
STRESS_SCORE already has an HRV-derived COMPUTED producer that is not
wired into the source-priority ladder or the weekly graded-series
collapse, so adding a second Oura producer caused nondeterministic
collapse and double-counting. Drop the daily_stress fetch/map/wiring and
leave a note to re-add once STRESS_SCORE is laddered. The vO2_max
mapping, which is correctly ladder-wired, stays intact.
Enlarge the OCR confirm and lab result-type tap targets to 44px; clamp the coach
hero chips; switch the Vorsorge due-status to a calendar-day delta so an item due
earlier today still reads as due today and a prior day reads overdue.
The in-app measurement routes evict the snapshot cache, but the iOS
batch ingest and the server-side Withings sync only marked it stale (or
never touched it), so a focus-refetch could serve pre-sync data for up
to the snapshot TTL. Route both background-sync entry points through a
hard evict so returning to the dashboard surfaces freshly-synced rows.
Add lab-report OCR ingestion: optional image and document inputs on the AI
provider clients, a capability-gated extract route, a mandatory per-row review
screen, and an idempotent commit that mints biomarkers from confirmed rows only.
Add the OcrCapability, OcrExtract, and OcrCommit operations under
/api/labs/ocr/* and regenerate the YAML. Document the multipart upload, the
untrusted-extraction contract, and the source OCR commit value. Mark the three
routes exempt in the module-route gate inventory under the labs data-layer
reasoning.
Add the Scan a report entry to the Labs page, gated on the capability probe.
The review dialog uploads a photo or PDF, then shows each proposed reading for
per-row confirm, edit, or discard with duplicate, new-biomarker, and
low-confidence hints in the labs design language. Only confirmed rows commit.
Add the OCR query key and the six-locale strings.
Resolve a vision-capable provider, gate with consent + a 6/hour rate bucket +
a reserved per-day token budget, then transcribe an uploaded photo or PDF into
structured proposed rows. The raw upload stays in memory only and is never
persisted. The commit route writes only the rows a human confirmed, resolving
or minting a biomarker per row with source OCR and skipping commit-time
duplicates. The capability probe drives whether the UI shows the scan entry.
Extracted text is treated as untrusted; the review screen is the safety
boundary.
Extend CompletionParams with optional images[] and documents[] so the
provider clients can carry an uploaded lab-report photo or PDF. Anthropic
folds image and native document blocks into the user turn; OpenAI and the
local OpenAI-compatible client fold image_url blocks. Text-only callers are
untouched. Add supportsVisionForConfig / supportsPdfForProvider and an
ocrExtract budget for the Lab-OCR ingestion path.
Rebuild the /coach surface as a centred-hero chat with a word-by-word fade, a
single thinking disclosure, and a per-message token footer; emit token usage on
the done frame and persist it via migration 0182; plumb an additive reasoning
frame with an elapsed-time fallback for non-reasoning providers.
Replace the cramped empty thread on the /coach page with a centred hero:
a greeting, the live composer beneath it, and HealthLog's own starter
questions as chips, over a faint purple atmospheric backdrop with one
staggered entrance reveal. The hero re-parents the existing composer
rather than forking it, so dictation, auto-grow, and the send/stop
control behave identically; the first send docks the composer to the
bottom and the conversation takes over. Gated to the page surface and
the fresh new-chat state, so the drawer, guided-question, and adopt
flows are untouched.
The Oura v2 client fetched sleep, readiness, activity, daily sleep, and
daily SpO2 but skipped the dedicated vO2_max collection and daily_stress.
An Oura user got no VO2max from any path even though the enum, the
cardio-fitness page, fitness-age, norms, and the LOINC code already exist.
Add fetchVo2Max + mapVo2Max (vo2_max -> VO2_MAX, mL/(kg·min)) and
fetchDailyStress + mapDailyStress. Oura exposes no headline stress score,
only minutes in high stress vs high recovery, so derive a bounded 0-100
STRESS_SCORE from the high-stress share of the day's charged time. Both
collections wire into the existing reactive-refresh sync tail; no
migration, no schema change, no iOS-contract impact.
The VO2max strip tile was opt-in, so accounts that already sync a
cardio-fitness sample from Withings, Apple Health, Fitbit, or Oura saw
nothing on the dashboard. Default the strip tile on; it still self-gates
on having a VO2_MAX sample, so empty accounts keep a clean tile strip.
The chart row stays off — VO2max charts live on /insights/cardio-fitness.
Stream the assistant turn word-by-word with a soft motion-safe fade
(StreamedProse), reduced-motion collapses to instant. Replace the
dots-plus-word placeholder with a single thinking indicator: a live
elapsed-time line that freezes into an auto-collapsed past-tense
disclosure, which surfaces a reasoning summary when one is present and a
factual fallback otherwise — never both a spinner and a Thinking word at
once. Add a quiet per-message token footer (count plus model, no cost)
fed by the live done.usage frame and persisted tokensUsed on reload.
Add the localised strings across all six bundles.
Invalidate the dashboard snapshot on measurement writes and refetch on focus so
fresh and background-synced readings surface without a hard reload; expose the
no-provider state on the briefing read with a connect-a-provider hint; align the
briefing footer caption.
An account with no AI provider configured still reads the last cached
briefing — the read path serves it without ever touching a provider — so
the days-old report rendered as current. The pregenerate cron runs fine;
the cache simply can never refill without a provider. Report provider
availability from the read GET and pair the briefing's relative age with a
discreet 'this can't refresh — connect a provider' hint linking to
Settings -> AI, on both the overview hero meta row and the briefing card
footer. The hint only shows when a briefing is actually rendered; the
empty-state already owns its own no-provider branch.
Also align the briefing card footer caption to the 11px scale the
adjacent period-in-review card uses, so the two overview footers read at
one size.
Default the labs surface to list view and reuse the medication card header for
tiles; surface confirm-gated biomarker delete on detail and list rows; add
alphabetical sort; align the labs settings page to the canonical SettingsCard.
A measurement added in-app left the dashboard hero, score ring and tile
strip showing yesterday's value until a hard reload. The snapshot query
carries refetch-on-mount/focus off and a 120 s poll, but the measurement
mutation bundle never invalidated the snapshot key, so an in-app create /
update / delete updated every other surface and left the Startseite stale.
Add the snapshot key to the measurement-dependent bundle, mirroring the
medication fix; the manual routes already hard-evict the server snapshot
bucket, so the triggered refetch returns post-write data at once.
Background syncs (Withings, Apple Health, iOS batch) fire no client
mutation event, so the keys were never invalidated and the dashboard only
caught up on the 120 s poll or a reload. Refetch the snapshot on window
focus so returning to the tab after a sync surfaces the new readings; the
60 s staleTime keeps rapid focus toggles free and the warm server cache
keeps the focus refetch cheap.
Defer illness recovery-gap correlation to an opt-in detail expansion; gate the
typical-gap insight behind signal-density thresholds; align the illness settings
page to the canonical SettingsCard pattern.
Restructure the Labs settings sub-page to the canonical Settings dialect:
each block is a SettingsCard with a SettingsCardHeader (neutral icon, title,
short description), matching the AI / notifications / mood sections instead of
bare sections with text-sm headings. Surface the new alphabetical sort options
in the sort select.
Add a delete affordance with a confirm step to the biomarker detail header,
reusing the existing DELETE /api/biomarkers/:id endpoint. On success the
catalog and result-list queries invalidate and the view returns to the labs
surface. Removes the Settings detour for clearing a stray biomarker.
Set the compact list as the Labs surface default; tiles in the card view now
route through MedicationCardHeader so a lab tile reads identically to a
medication or Vorsorge tile side-by-side. Add alphabetical (A-Z / Z-A) sort
options to the per-module sort choice.
Carry an additive usage envelope on the SSE done frame
({ totalTokens, model }) and persist tokensUsed + model onto
coach_messages (migration 0182, nullable, no backfill) so the
per-message token footer survives a conversation reload. Add an
optional reasoning frame the client can render inside the thinking
disclosure when a reasoning-capable provider emits a summary; providers
without reasoning emit none. Extend CoachMessageDTO and regenerate the
OpenAPI contract. Server-authoritative: clients display the count, never
recompute it.
The illness settings sub-page rendered bare sections with text-sm headings
and no card surface, breaking from the canonical Settings layout the
Stimmung and Medikamente sections follow. Wrap each block in a SettingsCard
with a SettingsCardHeader (neutral icon, text-lg title, short muted
description), so the view and order controls read as first-class Settings
cards consistent with the rest of the surface.
The illness list load triggered the cross-episode correlation fan-out on
every mount, so the page waited on up to 24 per-episode reads before it
painted. Move that work behind an explicit opt-in: the insights endpoint
takes an includeRecoveryGap query param (default off), the list reads the
count breakdown on a single fast query, and the recovery gap is computed
only when the viewer opens the Analyse expansion.
Tighten the typical-gap gate against coincidental data. A gap now counts
only from an episode that cleared a per-episode measurement floor
(MIN_GAP_MEASUREMENT_DAYS), at least MIN_EPISODES_FOR_TYPICAL_GAP such
episodes must qualify, and the median magnitude must reach
MIN_TYPICAL_GAP_MAGNITUDE_DAYS — a 0/+-1-day gap is the coincidental
baseline and is withheld as noise.
Drop the Vorsorge trend strip; show last-completed as relative today/yesterday
via a shared helper; render next-due status as discreet coloured inline text.
Remove the 7-day bar strip under the Vorsorge card last-done line and its
backing component, query key, and i18n strings. Render the last-done value
as a relative date (today / yesterday / absolute) through a shared
relativeCalendarDate helper that reuses the medication today/yesterday keys.
Show the next-due status as discreet coloured text - green today, orange
overdue - leaving the card surface neutral.
Wrap the sleep-debt headline and the chronotype section in a shared
grid on the Sleep view so they stack on narrow viewports and sit
side-by-side from the sm breakpoint up. Both already read the same
sleep-rhythm cache, so the pairing adds no extra fetch; each card keeps
its own internals.
Drop the pure-white Coach floating-action-button icon for the button's own
text-background on-accent token, so the glyph reads as a calmer mark on the
purple-to-pink gradient instead of a harsh white. Update the colocated
assertion to pin the glyph token.
Add first-class qualitative lab findings (positive / negative / borderline
serology) alongside the existing numeric path. A LabResult is now EITHER a
numeric reading (value + unit, optional reference bounds) OR a qualitative one
(valueText, e.g. "negativ"), never both, never neither.
- schema: make `value` nullable, add `value_text`; migration 0181 is additive
(DROP NOT NULL + ADD COLUMN), every existing numeric row is untouched.
- validation: numeric-XOR-qualitative refine on create + update; unit/bounds
apply to the numeric path only; valueText capped at 120 chars. The single
resource PUT rejects a cross-type edit.
- serialise: carry valueText through both DTOs; a qualitative row reports the
neutral `unknown` range verdict (no number to classify).
- FHIR: a qualitative result emits valueCodeableConcept — SNOMED qualifier
value for confidently-known terms (Negative 260385009, Positive 10828004,
Not detected 260415000, Detected 260373001), text-only for borderline /
grenzwertig and any unrecognised term. The raw text always rides .text.
- UI: a numeric/qualitative toggle in the lab form; list, history, detail,
chart, sparkline, and the doctor-report PDF all render qualitative readings
as their text and never plot or divide on a null value.
- i18n: new labs.form keys across all six locales.
- openapi: response carries a nullable value + valueText; regenerated.
Close the three FHIR R4 coding gaps in the health-record export. Both are
export-layer only — no schema column, no migration, no contract change.
Illness Conditions now carry a broad SNOMED CT category concept per
IllnessType (infectious disease, allergy, injury, mental disorder, autoimmune
disease, chronic disease, with the generic Disease root as the fallback for an
unknown type) instead of the single generic placeholder. The user's own label
stays on code.text, the class on category.text, and the unconfirmed +
patient-reported guard rails are unchanged — the export still asserts no
specific diagnosis.
Lab Observations resolve a curated analyte to a real LOINC coding alongside the
user's code.text for the common biomarkers (HbA1c, lipid panel, ferritin, TSH,
vitamin D, creatinine, eGFR, ALT, AST, CRP, fasting glucose, hemoglobin), with
analyte-key and EN/DE alias normalisation. The canonical UCUM code is stamped
on the value only when the recorded unit matches the mapped canonical symbol;
an unmapped analyte or a non-matching unit keeps the honest text-only code and
display-only unit, never a fabricated code. The resolver takes an optional
canonical-name argument so a future Biomarker-catalog link can take precedence
over the free-text analyte without a further change here.
Mint an active AI-processing consent receipt for the demo user so the
Coach renders its sample conversation without a consent-accept write the
read-only tenant rejects. The receipt is kind `ai_full`, the master grant
the Coach and Insights gates both accept; the read path only checks an
active row of that kind, so a placeholder artefact suffices. Add a
consent_receipts reset to keep the seed idempotent.
Anchor every taken intake's scheduled_for on its canonical dose-slot
instant (08:00 morning, 20:00 evening, Europe/Berlin) with taken_at set
slightly later, so the row is slot-anchored and resolves its slot within
the dashboard's resolution radius. Previously today's morning rows used
scheduled_for = taken_at, which the scheduler reads as an ad-hoc take that
only resolves within a minute of its own instant, leaving the 08:00 slot
reading overdue while the take floated beside it. Today's picture now reads
on-track at any daytime clock, with the evening dose pending and not yet due.
The public demo has no live AI provider, so the Insights, daily-briefing,
and per-metric status surfaces rendered empty or prompted to connect a
provider. Pre-populate the same cache rows a real generation would write
so the demo shows finished, grounded English prose that survives every
re-seed without any outbound call:
- users.insights_cached_text: comprehensive insight + daily briefing
(paragraph, signals of the day, key findings, trend annotations), with
insights_cached_at refreshed so the read serves it as current.
- audit_logs: the seven per-metric status cards, stamped with today's
Berlin day key and a real model name so the cache read treats each as a
fresh assessment.
- insight_narratives: week and month narratives, AES-256-GCM at rest like
every other narrative row.
All prose follows the assessment voice: second person, grounded in the
seeded numbers, autonomy-supporting, never diagnostic. English only, the
demo locale. Adds insight_narratives to the cleanup block so the seed
stays idempotent.
2026-06-19 23:49:29 +02:00
1287 changed files with 140489 additions and 20419 deletions
- The settings navigation returns to the familiar flat list down the left, reversing the grouped layout introduced in 1.25.1. Two-factor and passkeys stay under Account; active sessions, trusted devices, and login activity return to Data & Privacy; account deletion returns to Advanced. Every page keeps its own address. The recent quality-of-life touches stay: active sessions and security activity remain collapsed by default, the personal-context page links to the medical-history form, and your chronic conditions show there.
A patch release that turns the document inbox into a real library, reorganises the settings, closes a cross-account data-isolation gap on shared browsers, and carries a broad quality pass across the new surfaces. One additive migration (`0222`) applies automatically on start.
### Added
- A documents library: file any clinical document — a report, discharge letter, lab result, imaging, prescription, and more — without needing an AI provider. Give it a title, a category, and a date, then search, filter, sort, and browse it grouped by date. AI fact-extraction is now a separate, optional step you run on a document you have already filed, and you can view or download the original at any time.
- Richer reference descriptions for every biomarker — a few plain sentences on what each marker measures and what it relates to, in all six languages.
### Changed
- Settings are reorganised into nine clear groups — Account, Tracking, Display, Integrations, Notifications, AI & Coach, API & Access, Data & Privacy, and About — with in-page tabs. Far fewer entries down the left, and every page keeps its own address. Sign-in security — two-factor, passkeys, active sessions, and trusted devices — now lives together under Account.
- "Inbound documents" is now simply "Documents".
- Several repeated medical disclaimers were removed in favour of one clear, non-diagnostic statement, and the mental-wellbeing check-in now stands on its own page.
- The daily briefing can draw on grip strength, waist, and pain alongside your other signals, and the grip-strength and waist reference bands are age- and sex-aware.
### Fixed
- Signing out and back in as a different account on a shared browser no longer shows the previous account's data: the in-memory cache is now cleared at every sign-in and sign-out, alongside the offline caches.
- A once-weekly medication no longer triggers a false "low adherence" alert — the alert now uses the same cadence-aware calculation as the rest of the app.
- A lab value read from a filed document is no longer saved under the wrong unit: a unit mismatch is refused and reported rather than silently mis-recorded.
- Moving between insight pages now starts at the top; scroll resets on navigation throughout the app.
- A broad pass of accessibility, mobile-layout, date-format, and wording refinements across documents, settings, and the biomarker pages.
### Security
- Cross-account data isolation on a shared browser: no cached reading from one account can survive into another's session.
## [1.25.0] — 2026-06-28 — Clinical depth
A milestone release. It broadens tracking into clinical territory — validated mental-wellbeing self-assessments, grip strength, a pain score, waist measurements, a longevity lab panel, and respiratory rate as a first-class vital — adds structured allergy and family-history records, an optional weather-and-daylight context module, and the ability to file a doctor's letter and pull its facts into your record after you review them. It also makes the Coach's check-ins warmer, restores assistant-connector sign-in, and carries a broad correctness, performance, security, and accessibility pass. Additive migrations (`0208`–`0220`) apply automatically on start.
### Added
- Mental-wellbeing self-assessments (PHQ-9 and GAD-7) — opt-in, alongside your mood tracking and never a diagnosis. Item answers are encrypted; a non-zero self-harm response surfaces calm, locale-aware crisis-support contacts. Enable the Mental wellbeing module to use it.
- New signals you can log and chart like any other, each with its own detail page, reference range, and trend: grip strength, a 0–10 pain score, waist circumference and waist-to-height ratio, and respiratory rate.
- A longevity lab panel in the biomarker catalogue — ApoB, Lp(a), hs-CRP, HbA1c, fasting glucose and insulin, eGFR, GGT, ferritin, and the omega-3 index — with reference ranges.
- Structured medical-history records: allergies and family history, encrypted at rest, gathered under a new Medical history section and included in the health-record export. Immunizations are not part of this release.
- An optional environmental-context module: daily weather, daylight, and temperature for your location, correlated against mood, sleep, and vitals — off by default, with a home location, dated location periods, and backfill.
- Inbound documents: file a doctor's report or discharge letter and review the facts it contains before they enter your record. The app transcribes what is written and never interprets it; off by default.
- Biomarker detail pages now match the metric pages — description, summary, chart, trend, and a one-tap question to the Coach.
### Changed
- The Coach's proactive check-in is warmer and in your language: it greets you by name, keeps to one calm thought, never quotes your own words back, and never arrives two days running. A setting turns the daily suggestions off; another lets the Coach compose the check-in itself.
- Sleep debt now follows a configurable source order — a wearable's own figure or the computed one — and explains what the computed number means.
- One consistent assessment card, with a one-tap Coach question, across every insight page.
- Sortable history tables and real pagination where long lists used to truncate.
- A clear, non-diagnostic description of what HealthLog and the Coach are — and are not — and an explicit statement of where your data stays.
### Fixed
- Assistant connectors (such as Claude Code) can sign in again over the Model Context Protocol — an outbound-connection fix that also hardens every outbound request aimed at a host you configure.
- The daily briefing, health score, and weekly review keep their last good text instead of going blank when a model is slow.
- The clinician-share page renders again for newly created share links.
- Wearable daily figures, paused-medication adherence, and several long-window calculations are corrected.
### Security
- Free-text on the new records and on filed documents is encrypted at rest, and key rotation covers it.
- Mental-wellbeing answers stay out of the assistant connector, the Coach, and exports by default; their scores export only once you enable the module.
- A fresh second-factor check is required when changing the password.
## [1.24.0] — 2026-06-28 — Model Context Protocol server
A connector release. HealthLog can expose your own health record to MCP-compatible assistants (Claude, ChatGPT, and others) over a standard, OAuth-secured Model Context Protocol server — off by default, behind a module switch and a connector token you mint yourself. Two additive migrations (`0206`, `0207`) apply automatically on start.
### Added
- An MCP server — remote over `/mcp` and a local stdio command — that serves your own records to a connected assistant: metric series, glucose, sleep, workouts, medication compliance and schedule, labs (latest and history), correlations, baselines, level-shift detection, recovery and illness, cycle, integration status, and the preventive-care due-list. Every value carries its unit and reference range, with honest "no data" instead of a fabricated zero.
- A confirmed, scoped write surface: log a measurement, a blood-pressure pair, or a mood entry from an assistant — preview first, then confirm. Read-only by default; writing needs a token minted with the write scope.
- Installable prompt "skills": doctor-visit summary, weekly review, medication check, recovery / glucose / sleep review, and a lab-trend brief.
- Browseable resources and resource templates for metrics, labs, medications, and the doctor-visit report.
- An OAuth 2.1 connector flow (PKCE, audience binding, dynamic and URL-based client registration) so Claude.ai and ChatGPT can connect; mint read-only or read-and-write connector tokens under Settings → MCP. Enable and connect guides live in `docs/`.
### Security
- Off by default — the module gate plus a configured `APP_URL` are required, or the surface returns 404. A connector token is audience-bound to the MCP surface: it can never write or delete over the REST API and can never reach the admin surface. Writes are append-only, idempotent, range- and timestamp-bounded, and audited.
## [1.23.0] — 2026-06-27 — Account security and data sovereignty
A security release. It adds two-factor authentication, brings every sign-in and sensitive action under a stronger check, encrypts free-text health notes at rest, lets you download an encrypted copy of your record, and gathers export, deletion, and privacy controls into one place. Seven additive migrations (`0199`–`0205`) apply automatically on start; `0205` removes the retired bug-report tables.
### Added
- Two-factor authentication: an authenticator app (TOTP) or a hardware security key as a second factor, with one-time recovery codes — set up under Settings → Security.
- Sign-in asks for the second factor when it is enabled; you can trust a device for 30 days to skip it, while the password is always still required.
- Passkeys now show when they were last used and can be renamed, with a gentle prompt to add one.
- An active-session list with sign-out-everywhere, and an alert when a new device signs in.
- A security activity log of recent sign-ins, second-factor changes, and exports.
- A passphrase-encrypted export, so a downloaded copy of your record is never plain text — keep the passphrase safe, there is no recovery.
- A Data & Privacy page that gathers export, deletion, retention, encryption status, sessions, and activity in one place.
- A check that warns when a chosen password has appeared in a known breach.
- Admin controls to require a second factor and to review encryption coverage and key-rotation progress.
### Changed
- Free-text notes on measurements and mood entries are now encrypted at rest.
- Sensitive actions — account deletion, data reset, encrypted export, disabling the second factor, and key rotation — require a fresh second-factor check.
### Removed
- The bug-report integration has been removed.
### Security
- A second-factor session is bound to the credential it was issued for; an API token can never satisfy a fresh-factor check or reach the admin surface.
## [1.22.1] — 2026-06-27 — Mobile and Coach polish
A reliability and quality-of-life patch focused on the mobile interface and the Coach.
### Added
- Each lab biomarker shows a short description under its heading, the way Insights describes its metrics.
- Hover a Coach message for an action row: copy it, read it aloud, rate the reply, regenerate it, and see the time it was sent.
### Changed
- Read-aloud now selects a natural device voice instead of the default robotic one.
- The Coach prompt stays on one line and the conversation search field spans the full width.
- The settings and admin section navigation stays fixed while only the content scrolls, on both desktop and mobile.
- The admin "Module availability" section is now simply "Modules".
### Fixed
- The settings and admin side navigation no longer drifts while scrolling.
- Selecting a settings section on mobile no longer animates a scroll from the top of the page.
- The profile menu rows are evenly spaced, and the "More" menu labels are no longer clipped on narrow screens.
- Charts no longer crush the layout in landscape on small screens.
- Sleep insights no longer repeats the nightly average above its tiles.
## [1.22.0] — 2026-06-27 — A Coach that speaks, remembers, and charts; richer narratives across the app
A feature release. The Coach reads its replies aloud, can draw the metric it is discussing, remembers what you ask it to and resurfaces those reminders, and proposes checkups or measurement reminders you confirm inline. Insights and Coach assessments are rewritten for connected, motivating narratives instead of restated figures, and the daily read now folds in labs, preventive care, workouts, and glucose. Four additive migrations (`0195` hidden biomarkers, `0196` coach reminders, `0197` coach plan outcomes, `0198` AI settings); applied automatically on start — no operator action required.
### Added
- The Coach can read its replies aloud, and a reply can include a chart of the metric under discussion so the number has its trend alongside it.
- The Coach remembers things you ask it to keep and resurfaces a reminder when it comes due; a reminders list in settings shows what it is holding and lets you clear any of it.
- The Coach can propose a checkup or a measurement reminder, which you confirm inline before anything is saved.
- Hover a message to see the date and time it was sent.
- A configurable AI response timeout, and a separate optional provider dedicated to scanning lab documents.
- Steps now appear in Insights, and selecting a measurement type opens its insight.
### Changed
- Insight and Coach assessments are rewritten to connect signals into a motivating narrative rather than restate figures, with substantially expanded score summaries; generated text now renders in paragraphs and carries clearer "updated" timestamps — today with the time, yesterday, and a date for anything older.
- The daily read folds in labs, preventive care, workouts, and glucose, and correlations surface emerging signals over a recent window alongside more cross-signal links.
- The preventive-care area is now "Checkups", reached at `/checkups`; the German interface keeps "Vorsorge".
- Settings pages share one consistent layout, and the settings and admin sidebar stays pinned while the content scrolls.
- Lab biomarkers are a compact, editable list with hide and restore.
- Local AI servers stream their replies and honour the configurable timeout, so slower self-hosted models finish cleanly.
### Fixed
- The daily briefing and the insight trend no longer go blank when generation runs long.
- The blood-pressure "usual range" no longer shows a nonsensical band.
- A Coach notification now opens straight into the conversation it began.
- Insight headings drop the trailing info icon in favour of a short description.
A patch release. No migrations, no breaking changes.
### Changed
- The Coach settings link now lives in the composer's "+" menu, next to New chat and Conversations, and the redundant gear in the page toolbar is gone; the drawer keeps its own gear. The composer is the single control hub.
- Conversations open as a dedicated page with the search field at the top and the list grouped by recency — today, yesterday, this week, and earlier — instead of a slide-in panel. Both the "+" menu and the drawer's conversations control route there.
- The seeded "worth a look" opener spans the full width of the composer and carries a dismiss control that hides it for the rest of the day.
## [1.21.3] — 2026-06-27 — Coach memory and a standing quality harness
A patch release. One additive migration (`0194` coach plans); no breaking changes.
### Added
- The Coach can remember a goal or an if-then plan you confirm — "if my resting pulse is up in the morning, I'll keep the evening easy" — and carry it across conversations. It proposes the plan; it is saved only when you accept, and you stay in control of what it keeps.
- A standing quality harness runs in CI for the Coach: a graded set of reference cases plus an adversarial battery checks that every cited number traces to your own data, that the warmth never tips into over-validation, and that a red-flag symptom always escalates. An optional nightly model-judged pass is wired in and stays off until a key is configured.
### Fixed
- The dashboard and insights snapshot is cached per language, so switching language no longer briefly shows the previous language's briefing text.
- The Coach and the daily briefing failed for accounts signed in through the ChatGPT provider: the provider rejected the multi-step tool-call request, and the failure surfaced as a server error instead of a graceful message. A provider failure now always degrades to a clear in-chat notice — with the correct reconnect or rate-limit prompt where those apply — rather than a 500, and the provider request follows the documented tool-call wire format so the Coach's multi-round reasoning and the briefing generate again. The provider's own error detail is now captured for diagnosis.
### Changed
- While the Coach is thinking, the bubble shows only the three animated dots — the "Thinking…" word is gone (it remains for screen readers).
## [1.21.2] — 2026-06-27 — Ambient Coach presence
A patch release. The Coach surfaces what it already computes — at the metric, on the score, in the briefing — instead of waiting to be asked. No migrations, no breaking changes.
### Added
- A "Coach read" on every metric page: one line placing today's reading against your own usual range, and — when the data supports it — one line naming the strongest connected signal in plain association language. While your range is still forming it says so rather than guessing.
- Opening the Coach from a metric or card now shows that it is already on that metric, with a tappable opening question; the unscoped Coach opens on the day's most notable signal instead of a blank box.
- The daily briefing recalls the prior period and points ahead.
- The health score surfaces an honest "internal read" when its contributors disagree — good sleep but a rising resting pulse — and notes when a metric has come back inside your usual range after a dip.
### Changed
- The Coach's motivational-interviewing repertoire gained develop-discrepancy and roll-with-resistance moves and a clearer anti-persuasion boundary.
- The prose number-check now runs on the local-model path too, so grounding holds regardless of provider.
A patch release. No migrations, no breaking changes.
### Fixed
- Pop-up forms now scroll only their body on desktop, keeping the action row in view. A tall form — or one made tall by browser zoom, display scaling, or a longer-language label — could push the primary button below the fold and force a scroll to reach it; the centred dialog now matches the bottom sheet, which already pinned its footer. Every form the app mounts through the shared sheet surface inherits the fix.
## [1.21.0] — 2026-06-26 — Date-format preference, a far more connected Coach, and broad correctness work
A feature release. The Coach reaches every data domain on demand, surfaces the cross-metric patterns the analytics tier already discovers, opens in context from any screen, and speaks with a warmer, forward-looking voice on one shared set of safety thresholds. Dates render in your chosen format everywhere. Two additive migrations (`0193` rollup x-rescale, `0192` date format); no breaking changes.
### Added
- A date-format preference in the profile — automatic (follows the language), day-month-year, month-day-year, or ISO — honoured across the app, including every date and date-time field, which now render in your chosen order regardless of the browser's locale.
- The Coach answers from the full data picture: every metric domain on demand plus cycle and workouts, and it surfaces the discovered cross-metric correlations — medication adherence against symptoms, short sleep against the next morning's vitals, and so on — citing only what the analytics tier actually found.
- Open the Coach in context from any metric page or insight card and it arrives already scoped to what you were looking at, with a relevant opening question.
- "Learn more" pointers on the vitals tiles, the glucose panel, the resilience tile, and the lab biomarker detail link out to the matching guide; the Coach references the same guides and can no longer offer a link that doesn't exist.
- Medication compliance and symptom severity are now first-class signals in the correlation engine, so an adherence dip that tracks a symptom flare can finally be surfaced.
### Changed
- The Coach connects signals into one story instead of listing metrics, looks ahead with gentle, ranged outlooks, ends an action turn by checking your confidence and offering a single doable step, and keeps affirmation earned. A closed acute-symptom clause points to prompt medical attention for crisis signs.
- Critical-threshold numbers — blood pressure, fever, glucose — now come from one source, so the dashboard banner, the Coach, the status cards, and the notifications always state the same thresholds.
- The trend regression is composed on an origin-rescaled, mean-centred basis, removing a floating-point cancellation on long windows; the rollup now matches the live computation to the last digit.
- The daily AI usage budget is provider-aware: usage on your own OpenAI key, ChatGPT plan, or local model is no longer limited by the server-cost ceiling that applies to an operator-provided key.
- The Coach builds its data snapshot once per turn and may take an extra reasoning round for a deep cross-metric question.
- Discovered correlations are held to an effect-size floor and a confidence tier, and sparse personal signals are shrunk toward the baseline, so a real-but-trivial association is no longer stated as a confident driver.
### Fixed
- ChatGPT/OpenAI sign-in users no longer hit a spurious "daily limit reached" after only a couple of messages.
- The symptom journal's recovery-return and the SpO2 red-flag count consecutive calendar days and are order-independent; the "two or more vitals out of band today" flag is keyed to your calendar day rather than UTC.
- Withings activity sync and the medication-intake dedup issue far fewer queries on a long backfill.
- The trend read returns the same boundary day whether served from the rollup or from a live query.
- The native date primitives that ignored the locale are gone — every field routes through the format-aware inputs.
## [1.20.2] — 2026-06-26 — Hardening across insights, safety flags, and integrations
A patch release. A broad pass over the server tier and the AI surfaces: a few correctness and safety fixes, a numerical hardening of the trend tier, and several smaller robustness and cost improvements. No schema changes.
### Added
- The lab-report review screen marks low-confidence rows so an uncertain reading is easy to spot before it is saved.
### Changed
- The trend regression (slope, r² and variability) is composed from mean-centered sums, which removes a floating-point cancellation on long, near-flat windows and keeps the rollup result in line with the live computation. Each accumulator now uses the exact stored sum rather than reconstructing it.
### Fixed
- The symptom journal's "sustained fever" flag — which can raise an urgent prompt to seek care — now counts only consecutive calendar days. A sparse series of isolated febrile entries no longer trips it, so the alert reflects a genuine multi-day run.
- Insight generation no longer reads a full measurement history per metric to compute its 30-day comparisons; it reads only the window those averages need. Generating insights on a long-history account is markedly faster, on both the page request and the overnight pre-generation.
- Photo and PDF scanning is no longer offered for text-only reasoning models that cannot read an image, so a scan attempt fails clearly up front instead of erroring at the provider.
- A notification channel whose stored configuration cannot be read now disables itself immediately with a precise reason, instead of retrying a permanent error several times and ending on a misleading message. This surfaces during an encryption-key rotation gap.
- The Coach cancels its in-flight model call when the browser tab disconnects mid-answer, rather than paying the full cost into a closed connection.
- Local-OCR text structuring reserves a budget proportionate to its real cost instead of the larger vision-scan ceiling, and refunds cleanly on a failed read.
- The admin diagnostic endpoints return a 422 for a malformed query string instead of a 500.
- A user reported the dashboard crashing on returning to a backgrounded tab. The on-focus snapshot refetch re-rendered the page, and each chart's "data ready" notify — keyed on a handler that the page recreated every render — re-fired on every commit, driving the chart row into an unbounded update loop until the browser tripped its render-depth guard and the page fell back to its error card. The notify now fires once, when a chart's initial query settles, regardless of how often the page re-renders.
## [1.20.0] — 2026-06-22 — Coach on-demand retrieval, deeper recovery insight, Fitbit, and a leaner data tier
A feature release. The Coach fetches what it needs on demand instead of carrying a full snapshot; the recovery-gap learns from the symptom journal and sleep; Fitbit connects over its own Web API; the trend tier serves slope and variability without a live scan; and several built surfaces become discoverable. Two additive migrations (`0190`, `0191`); no breaking changes.
### Added
- The Coach answers from on-demand retrieval: a small base context plus a data inventory, with tools it calls to pull the exact series, lab, sleep, medication or recovery figures a question needs. It cites only what it fetched and says so plainly when there is no data. First-turn cost drops substantially while the grounding and cross-metric reasoning are unchanged.
- The recovery-gap reads the illness journal's own symptom-severity curve and adds a sleep-context line, and it is now relapse-aware — it reports the final sustained return, not the first.
- Fitbit connects experimentally over the Fitbit Web API with PKCE, so a self-hoster can link it with an instant developer app.
- The sleep, steps and glucose dashboard tiles and the preventive-care row are surfaced by default (each self-hides without data), so built features are easier to find.
### Changed
- The AI client layer moves to a structured message format with per-provider prompt-caching, which lowers cost on repeated context and lays the groundwork for retrieval.
- Trend slope, r² and standard deviation are served from the rollup tier via stored regression accumulators instead of a live scan on every request; the result matches the live computation.
### Fixed
- The fever red-flag now reads days in chronological order, so an episode logged from both a thermometer and the journal can no longer raise or hide a sustained-fever flag incorrectly.
- A vital that settles and then shows a single stray out-of-range reading keeps its recovery day instead of reporting none.
- Fitbit sleep segments are placed on the correct day for users outside UTC.
- Concurrent syncs for Withings/WHOOP/Oura/Fitbit no longer park a connection at a spurious reconnect when a one-time-use refresh token is rotated.
- The Coach reply is no longer corrupted into raw JSON on some providers, and a tool request is no longer dropped on an unexpected stop reason.
- The trend tier records when slope is briefly unavailable pending backfill rather than reporting it silently.
### Refactor
- The Coach's per-request reads are shared across the retrieval fan-out instead of repeated per tool.
### Security
- Dependency bumps: next, pg-boss, radix-ui, pg, @testcontainers/postgresql, and the CI actions.
A follow-up that folds the illness journal's own symptom curve into the recovery-gap, surfaces Oura resilience, fixes multi-year chart ranges, and widens the heart-rate and Telegram data paths. Two additive migrations (`0188`, `0189`); no breaking changes.
### Added
- The recovery-gap now reads the illness journal's own daily functional-impact curve (and linked symptom severity), not just passive vitals. When symptoms lead the recovery, the card names them — "you tend to feel better before your logged symptoms ease". Sparse journals withhold rather than guess.
- An Oura resilience tile on the recovery surface, showing the latest resilience band with a quiet above/in-line/below-average cue and a calm learning state until enough days are recorded.
- Hourly heart-rate uploads can carry a per-bucket low and high alongside the average, so a high-frequency client no longer loses the within-hour spread.
- A numeric reply to a Telegram measurement reminder is captured as a reading (for example a weight or a glucose value), bound to the linked chat and self-cleaning like the mood path. Blood pressure stays a simple "done" for now, since it needs two values.
### Changed
- A multi-year "All" chart range now renders the whole history, stepped up to a weekly or monthly tier, instead of being silently truncated to the most recent year.
### Fixed
- `telegram_chat_id` is now unique, with a one-time clean-up of any stale duplicate bindings (no account is removed).
- Migration `0187` is rerun-safe.
- The sleep peer-card comment no longer overstates window parity; each card discloses its own night count. Dead streaming state left after the reasoning-disclosure removal is gone.
A follow-up to 1.19.0 that refines the Coach conversation flow and cost, adds a sleep average card, completes the mood colour pass, tidies navigation, and sharpens the illness recovery-gap. No schema changes; no breaking changes.
### Added
- A sleep average-per-night card sits beside sleep debt and chronotype as a third peer, computed over the same scorable-night set so the three cannot disagree. It stays calm until enough nights are scored.
- The recovery-gap names the vital someone tends to recover ahead of — for example "you tend to feel better before your resting heart rate settles" — rather than a single bare line.
### Changed
- The Coach opens on a fresh chat by default. It resumes a past conversation only when the Coach has written something unread; an explicit `?c=<id>` deep-link still opens that conversation. This refines the 1.19.0 behaviour, which always reopened the most recent thread.
- The chat view carries an always-visible toolbar with a clear way into past conversations and a new-chat action, and reaching conversations from the dashboard tile now always lands on the conversation list instead of an empty pane.
- The Coach sends the full health snapshot once per conversation (and again only if the conversation grows long enough to scroll the original context out of view); short follow-ups ride the existing transcript. A one-word reply no longer re-pays the whole snapshot, cutting per-message cost on follow-ups by roughly an order of magnitude while keeping the same grounding and cross-metric correlation.
- The streaming reply shows a typing indicator that gives way to the streamed text, replacing the reasoning-status disclosure.
- The settings navigation entry that opens the dashboard layout is labelled "Dashboard" to match where it leads.
- The main sidebar order groups the care surfaces together: medications, preventive care, lab values, illness, then insights, the Coach, and achievements.
### Fixed
- The mood time-of-day chart and the mood calendar now use full colour saturation in line with the other mood surfaces; only no-data cells stay muted.
- The cross-episode recovery-gap baseline counts only days that deviated in the illness-adverse direction, so a neutral reading (such as a weight drift) can no longer pad the sample or be named the recovery driver.
- The sleep average-per-night field is part of the documented sleep-rhythm response contract.
## [1.19.0] — 2026-06-21 — Coach context, a leaner AI pipeline, new device signals, interactive reminders, and wider self-hosting
A feature release. The Coach gains the lab values and a direct way in; the daily AI work is restructured to do less and reuse more; several new device signals are captured; reminders become interactive over Telegram; charts and the cycle and sleep surfaces are corrected and polished; and the project ships ready-made templates for one-click self-hosting. Three additive migrations (`0185`–`0187`); no breaking changes.
### Added
- The Coach can answer about lab results — the most recent value per biomarker (last twelve months) is provided as grounded context, with reference range, in/out-of-range status, and date. Qualitative results are included; the Coach quotes them verbatim and never invents a value.
- The Coach entry points open directly into the most recent conversation, the same thread across devices, rather than always starting blank. A `?c=<id>` deep-link opens a specific conversation; the new-chat start remains.
- The daily briefing is surfaced on the dashboard as a short spotlight strip, lifting its key signals above the fold for a fresh briefing.
- Mood can be logged straight from a Telegram reminder — a 1–5 inline choice or a short written note — and the reminder chat self-cleans about thirty minutes later so old prompts do not linger or bias the next entry. Mood and measurement reminders also carry calm "remind me later" / "done" actions.
- Withings ECG records the device's atrial-fibrillation screening result and the full ECG waveform (stored encrypted at rest).
- Oura cardiovascular age and resilience are recorded.
- A configured multimodal model reads a scanned lab document directly; where the model is text-only, the on-device OCR path still applies.
- Ready-made self-hosting templates: an Unraid Community Applications template and a Portainer stack template, a hardened `docker-compose.yml`, and a NAS quick-start guide.
- A go-forward aggregated heart-rate upload contract (hourly buckets) so high-frequency clients stop accumulating one row per raw sample.
### Changed
- The AI pipeline does less, more coherently: the comprehensive briefing reads a bounded recent window (all-time figures stay accurate); per-domain status summaries generate once on the nightly pass instead of per metric through the day; a shared feature snapshot is computed once and reused; slow movers (weight, BMI) skip regeneration when their inputs are unchanged; illness and cycle state fold into the briefing context at no cost for users without them. Reproducibility, consent, and grounding checks are unchanged throughout.
- Medication remaining-supply is computed server-side and rendered from one canonical value, so the web and the iOS client always agree.
- High-frequency heart rate and blood oxygen are consolidated server-side into daily aggregates (mean with the daily minimum and maximum preserved; a resting-heart-rate value is derived where the source has none), matching the cumulative-metric consolidation already in place.
- Day-level bucketing for recent readings honours the user's timezone, so a late-evening reading lands on the correct local day.
- Sleep debt is a recovering rolling balance over recent nights instead of an ever-growing sum, so it reflects what is owed now and shrinks after a good night.
- The cycle view aligns its columns, uses consistent typography and equally sized toggles, and shows the prediction disclaimer once — clearly, at onboarding — rather than repeated across the feature.
- The sleep card's tiles share one consistent layout, and mood charts paint at full saturation.
- Sleep-night minute values are returned rounded; the notifications page opens on a single heading.
### Fixed
- Charts honour the 90-day and All ranges — they were stuck on roughly the last month regardless of the selected range.
- Medication remaining-supply can no longer display a nonsensical negative figure; the canonical readout is floored and a would-be underflow is recorded for diagnosis.
- Notification settings now label the measurement-reminder, medication-intake-sync, and coach-nudge events instead of showing raw keys.
### Performance
- Achievements and the dashboard summary serve from a stale-while-revalidate cache, the achievements builder precomputes day boundaries, three charts defer their library, and the analytics aggregate folds its canonical self-join into a single shared expression.
### Security
- The ECG waveform is encrypted at rest and covered by the key-rotation script, with a schema-driven test that fails the build if any encrypted column is ever left out of rotation.
- A Telegram interaction is strictly bound to the linked account before it can write, and the chat identifier is redacted from logs.
### Refactor
- Oversized modules are split into focused modules behind a barrel with no change in behaviour: the OpenAPI medication and insights registries, medication validations, the import panel, and the account settings section.
### Deferred
- Garmin: blocked by an external gate (the developer programme is invitation-only and not accepting new applicants), so there is nothing to build until access opens; only the official hosted instance could pursue it.
- Briefing map-reduce over the per-domain summaries, the cycle/sleep companion-line edge cases, multi-year "All" chart ranges, and the Oura resilience tile remain follow-ups.
## [1.18.10] — 2026-06-20 — data-loss fix, measurement consolidation, grounding, and the full-page Coach
A correctness-and-polish release. It closes a silent data-loss path and a stalled measurement-consolidation job, hardens AI output grounding, and acts on a broad round of UI feedback. Two additive migrations (a covering index and a labs setting); no breaking changes.
### Fixed
- Editing one field of a cycle or condition-journal day entry no longer wipes an encrypted note that cannot be decrypted with the current key — the stored value is preserved untouched unless that field is explicitly changed.
- The nightly consolidation of cumulative metrics (walking/running distance, flights climbed, active energy) no longer aborts on the first conflicting day, so raw per-sample history collapses to daily values as intended. A one-time compaction of the existing backlog runs after deploy.
- The daily briefing refreshes when regenerated, and the home screen reflects a saved tile layout without a manual reload.
- The Coach dictation button reports when speech input is unavailable instead of doing nothing.
### Added
- The Coach has a full-width page (no panel-within-a-panel), reachable conversation history, and a clean new-chat start.
- Scan a lab report without a vision model: optional on-device OCR (off by default) reads the image in the browser and sends only the extracted text to the configured AI provider, then the same mandatory per-row review applies. The labs "Add" action now offers "scan a document" or "add a value manually".
- Withings sleep now records heart rate, respiratory rate, blood oxygen, sleep HRV, and the sleep score (previously discarded at sync).
- Oura VO2max is recorded, and the VO2max tile is shown on the dashboard by default.
### Changed
- Preventive-care cards show the cadence on the metric row rather than beside the heading, and the heading opens the measurement history filtered to that metric.
- The danger zone and the daily-briefing card match the rest of the surface; the feedback tool is marked deprecated; the preventive-care and labs settings follow the standard section layout.
- Lab values: delete moved to the value's detail view, the overview tags are quieter, and the reference-range confidence is surfaced when a new biomarker is created.
- Location lookups default to the online resolver; the offline database is optional.
- Mobile: the full-page Coach composer stays clear of the bottom navigation, long navigation labels stay readable, and touch targets meet 44 px.
### Performance and integrity
- Blood-pressure in-target and grade read identically whether served warm or cold (one canonical source per day).
- A per-session statement timeout stops a single stuck query from holding a connection; a soft-delete-aware measurements index; the all-time fallback scan is bounded.
### Security
- Daily-briefing numbers are checked against the underlying signals; the Coach reply is screened for fabricated values or dose instructions; the period narrative is grounded before it is stored; the metric status cards state precomputed trends rather than re-deriving them.
### API
- The labs OCR capability response carries a `mode` (vision or text) and a new local-OCR preference; additional measurement types are recorded server-side. All additive and back-compatible.
A broad UI and data release. It adds two migrations (both additive and nullable — qualitative lab results and Coach token columns) and three additive, back-compatible API changes (see the note below).
### Added
- Coach opens on a calm new-chat surface: a centred greeting and a single composer, then each answer streams in word by word with a soft fade, a collapsible thinking disclosure, and a quiet per-message footer showing the tokens used and the model. The quick-access drawer is unchanged.
- Scan a paper lab report. With an AI provider connected, upload a photo or PDF and the reader proposes the measurements and biomarkers it finds. Every row goes through a mandatory review — confirm, edit, or discard each one, with duplicate and reference-range hints — and only what you confirm is saved. The uploaded file is held in memory for the read alone and is never stored.
- Record a qualitative lab result — positive/negative or another text value — where a number does not apply.
- Sort lab values alphabetically, and delete a biomarker directly from the labs list and detail views.
- VO2max shows on the dashboard by default, and Oura's VO2max reading is now ingested alongside the existing sources.
### Changed
- The labs surface defaults to the list view, and its tiles share the medication-card layout. The labs and condition-journal settings now follow the same section pattern as the rest of Settings.
- Preventive-care cards read the last-completed date as "today"/"yesterday" and the next due date as a discreet "today" (green) or "overdue" (orange), and drop the small trend strip.
- Sleep-debt and chronotype sit side by side, and the daily-briefing card aligns with the surrounding type.
### Fixed
- The dashboard reflects new readings without a hard refresh: an added or background-synced measurement now refreshes the home snapshot — including on Withings and Apple Health sync, and on returning to the tab.
- The condition journal opens quickly again. The cross-episode correlation runs only when an entry's analysis is opened, and the typical-recovery-gap insight appears only when there is enough real data to support it.
- The latest-report area shows its age and, when no AI provider is connected, points to where to connect one, instead of presenting an old report as current.
- The Coach quick-access icon uses a softer tone.
### API
- The Coach stream's `done` frame now carries an optional token `usage`, and a Coach message can carry `tokensUsed` and `model`; an optional `reasoning` frame is reserved. New `GET/POST /api/labs/ocr/{capability,extract,commit}` endpoints; rows committed from a scan carry `source: "OCR"`. All additive and back-compatible.
## [1.18.8] — 2026-06-20 — coded condition and lab terminology in the health-record export
An additive export-only release: the FHIR R4 health record now carries standard terminology codes where it can assert them honestly. No migration, no API contract change.
### Changed
- Condition resources for journal entries carry a SNOMED CT category coding per illness type (infection, allergy, injury, mental health, autoimmune, chronic, other) alongside the user's free-text label. These stay broad categories, not diagnoses — the patient-reported, unconfirmed guard rails are unchanged.
- Lab Observations now emit a LOINC code and a canonical UCUM unit for the common biomarkers (HbA1c, lipid panel, ferritin, TSH, vitamin D, creatinine, eGFR, liver enzymes, CRP, fasting glucose, hemoglobin), resolved through an analyte alias table. An unrecognised analyte, or a unit that does not match the canonical UCUM symbol, keeps the honest text-only behaviour rather than asserting a code it cannot validate.
## [1.18.7] — 2026-06-19 — dashboard, Coach and insights polish
A broad UI, insights, and AI-efficiency release. It ships database migrations (see the migration note below) and two client-facing contract changes: Coach budget responses now arrive as a stream error rather than a 429, and the measurement-reminder push carries the reminder id.
@@ -4,7 +4,7 @@ Working notes for anyone (human or otherwise) editing this repository. Skim befo
## What HealthLog is
Self-hosted personal-health-tracking PWA: weight, blood pressure, pulse, body composition, glucose, sleep, mood, medication compliance. Withings + Apple Health sync, multi-provider AI Insights (BYOK or local), doctor-report PDF, native iOS client in public beta. Single `docker compose up`, Postgres-backed, AES-256-GCM at rest. PolyForm Noncommercial 1.0.0 (AGPL-3.0 through v1.15.18). Current line: v1.18.x.
Self-hosted personal-health-tracking PWA: weight, blood pressure, pulse, body composition, glucose, sleep, mood, medication compliance. Withings + Apple Health sync, multi-provider AI Insights (BYOK or local), doctor-report PDF, native iOS client in public beta. Single `docker compose up`, Postgres-backed, AES-256-GCM at rest. PolyForm Noncommercial 1.0.0 (AGPL-3.0 through v1.15.18). Current line: v1.20.x.
HealthLog is a self-hosted health tracking app (PWA + native iOS client) that runs from a single `docker compose up` on a NAS, homelab, or small VPS. It brings weight, blood pressure, pulse, blood glucose, body composition, sleep, mood, menstrual cycle, and medications onto one timeline — synced from the devices you already own — and keeps everything encrypted at rest on hardware you control. No vendor cloud, no subscription, no telemetry.
HealthLog is a personal health record-keeping and wellness tool. It is not a medical device and does not diagnose, treat, cure, or prevent any disease — it helps you log, visualise, and export your own data, and leaves clinical judgement to a clinician.
Try the [live demo](https://demo.healthlog.dev) to see a working install — sign in with `demo` / `demo123demo123`.
## Highlights
- **Every vital on one timeline.** Weight, blood pressure, pulse, glucose, body composition, sleep, SpO₂, mood, and cycle tracking — charted with trends, moving averages, and clinical reference ranges (ESH 2023, ADA 2024), overridable with the targets your clinician set.
- **Your devices, one server.** Withings and WHOOP sync over OAuth2, Google Health/Fitbit connects experimentally, an Apple Health `export.zip` folds your full history in, and the native iOS app streams HealthKit live. A per-metric source priority decides which reading is canonical when wearables overlap.
- **Every vital on one timeline.** Weight, blood pressure, pulse, glucose, body composition, sleep, SpO₂, respiratory rate, mood, and cycle tracking — charted with trends, moving averages, and clinical reference ranges (ESH 2023, ADA 2024), overridable with the targets your clinician set.
- **Clinical signals, opt-in and additive.** Validated mental-wellbeing self-checks (PHQ-9, GAD-7) that sit beside mood tracking with calm crisis-resource handling, grip strength, a 0–10 pain score, waist and body measurements, and a longevity lab panel — each added through one signal registry, so a new signal wires into charts, correlations, the coach, the FHIR map, and the API at once.
- **Records and the documents you're handed.** Structured allergies and family history (encrypted), lab values you key in or scan, and inbound documents — discharge letters and reports — read to extract the facts they state (values, codes, dates) for you to review and confirm. HealthLog transcribes what a document says; it never interprets or diagnoses from it. (Immunization records are deliberately out of scope.)
- **Context that explains the numbers.** An opt-in environmental module folds weather, daylight, and temperature in as influence signals (winter darkness against mood, heat against blood pressure) using a home location you set, so a pattern has somewhere to point.
- **Your devices, one server.** Withings, WHOOP, Oura and Polar sync over OAuth2, Fitbit connects experimentally over the Fitbit Web API, an Apple Health `export.zip` folds your full history in, and the native iOS app streams HealthKit live. A per-metric source priority decides which reading is canonical when wearables overlap.
- **Medication tracking that tells the truth.** Flexible schedules (weekly injections, weekday-only, intervals, PRN, cyclic), a configurable intake window per dose, and a traceable dose history where every slot reads taken, late, skipped, or missed — the same ledger the compliance rate is computed from, so the percentage can never disagree with the timeline.
- **AI insights you own.** Daily briefing, health scores, correlations, and a conversational coach grounded in your own measurements, with proactive check-ins. Bring your own OpenAI or Anthropic key, or point at a local endpoint (Ollama, LM Studio, vLLM) so nothing leaves your network.
- **AI insights you own.** Daily briefing, health scores, correlations, and a conversational coach grounded in your own measurements, with calm proactive check-ins it suppresses when there's nothing genuine to say. The coach describes your data, never diagnoses. Bring your own OpenAI or Anthropic key, or point at a local endpoint (Ollama, LM Studio, vLLM) so nothing leaves your network.
- **Clinician-ready output.** A doctor-report PDF generated client-side, a read-only HL7 FHIR R4 API, and scoped, time-limited share links you revoke after the visit.
- **Private by construction.** AES-256-GCM encryption at rest with zero-downtime key rotation, passkey login, server-side sessions, strict CSP — and no third-party tracking anywhere.
- **Built to be lived in.** Installable PWA with offline support, medication reminders over APNs, Telegram, ntfy, and Web Push, a sub-second dashboard on years of imported history, English and German end to end.
The full feature tour, integration guides, and API reference live at [docs.healthlog.dev](https://docs.healthlog.dev).
# LAN / plain-HTTP host (NAS, homelab, Tailscale) — required, or login fails:
echo"SESSION_COOKIE_SECURE=false" >> .env
docker compose up -d
```
Open **http://localhost:3000** — the first registered user becomes admin. The compose file pulls a pre-built multi-arch image (`amd64` + `arm64`) from [GHCR](https://github.com/MBombeck/HealthLog/pkgs/container/healthlog); no build step required. Behind a reverse proxy, set `NEXT_PUBLIC_APP_URL` and `APP_URL` first — see the [self-hosting guide](https://docs.healthlog.dev/self-hosting/).
Open **http://localhost:3000** (or `http://<host>:3000`) — the first registered user becomes admin. The compose file pulls a pre-built multi-arch image (`amd64` + `arm64`) from [GHCR](https://github.com/MBombeck/HealthLog/pkgs/container/healthlog); no build step required.
Verify the running version:
```bash
curl -s http://localhost:3000/api/version # returns version + buildSha + builtAt
```
`SESSION_COOKIE_SECURE=false` is the one setting plain-HTTP self-hosts must set: without it the session cookie carries the `Secure` flag, the browser drops it over HTTP, and login silently fails. Leave it unset (or `true`) only when a TLS reverse proxy serves HTTPS — and then set `NEXT_PUBLIC_APP_URL` and `APP_URL` to the public URL first (see the [self-hosting guide](https://docs.healthlog.dev/self-hosting/)).
Persistent state lives in exactly one named volume (`pgdata`, the Postgres data dir); the app container is stateless, so backing up the database backs up everything.
NAS and homelab users have ready-made paths: [Unraid Community Applications](docs/self-hosting/unraid.md) and [Portainer app templates](docs/self-hosting/portainer/README.md).
| `list_metrics` | One row per domain: whether data is present, an approximate sample count, and the tool that retrieves it. **Call this first.** |
| `get_metric_series` | One metric's aggregate (count, min, max, mean, slope) plus recent-daily and weekly timelines, with units and population reference bands. |
| `get_metrics` | Several metric series in one call — a paginated fan-out over `get_metric_series`, one grounded result per metric. |
| `get_glucose_panel` | Per-context daily means plus the trailing-30-day clinical panel (time-in-range, GMI, CV%, estimated A1c). |
| `get_sleep` | Per-night asleep + stage minutes plus the sleep-rhythm summary (sleep debt + chronotype). |
| `get_workouts` | Most recent sessions (sport, duration, energy, distance, avg/max HR) plus a per-sport rollup over the window. |
| `get_medication_compliance` | Cadence-aware adherence: dose-weighted compliance rate, expected vs taken/missed, current-cycle status, any GLP-1 titration context. |
| `get_labs` | Latest reading per biomarker over the last 12 months (optionally one analyte). With `history:true` + an analyte, that analyte's paginated reading trajectory. |
| `get_illness_recovery` | Rest mode, active and recently-resolved illnesses, recovery / strain composites, and the illness retrospective (recovery-gap, nadir, red flags). |
| `get_cycle` | Menstrual-cycle context: phase + day-of-cycle, next predicted event, headline phase-correlation. Gated on cycle tracking; descriptive only. |
| `get_correlations` | FDR-controlled day-to-next-day driver pairs between behaviours and outcomes, each with direction, lag, sample size, and a descriptive (never causal) note. |
| `get_correlation` | The vetted, lag-aware association between **two named metrics**: direction, lag, sample size, Pearson r, and a descriptive note. |
| `compare_metric` | One metric vs another over the same horizon, **or** one metric across two horizons (fixed windows or `{from,to}` ranges), with per-side stats + a delta. |
| `get_metric_baseline` | Where the latest reading sits against the user's own usual range (median ± robust deviation), plus the strongest lagged driver. Needs ≥ 7 days of history. |
| `detect_changepoints` | Points where a metric's level shifted over a window or `{from,to}` range — date, direction, before/after means. High firing bar. |
| `get_medication_schedule` | When each active medication is next due and which are overdue right now — name, dose, next-due, overdue flag, as-needed flag. Reuses the recurrence engine. |
| `get_integration_status` | Sync health of connected devices/services — connected, last sync, reauth-required/failing — to answer "why is my data stale?". No secrets or tokens. |
| `get_preventive_care` | The user's own configured preventive-care (Vorsorge) reminders — upcoming/overdue checkups with next-due dates. Surfaces configured reminders, invents nothing. |
| `search` | Free-text search over the user's record (metric domains, medications, lab analytes). Returns `{ results: [{ id, title, url }], nextCursor? }`. |
| `fetch` | Hydrate one record by the id `search` returned (`metric:weight`, `med:<id>`, `lab:LDL`). Returns `{ id, title, text, url, metadata }` with a citation deep-link. |
`search` + `fetch` are the de-facto two-tool retrieval convention and the
**only** tools ChatGPT calls in its default (non-Developer) mode. Each
result carries a real, user-openable HTTPS deep link into the HealthLog
web app so the assistant can cite it.
Every read tool is annotated read-only / non-destructive / idempotent /
closed-world — the read-only guarantee is structural (only read tools are
registered for a read session), and the annotation merely advertises it.
### Clinical signals
The v1.25 clinical-signals set — **grip strength**, **pain (0–10 NRS)**,
**waist circumference**, and **waist-to-height ratio** — is reachable
through the rollup-backed reads `get_metric_baseline`, `compare_metric`,
and `detect_changepoints` (pass the signal name, e.g. `grip strength` or
`waist-to-height ratio`), and through `search` → `fetch` (`search`
surfaces a signal only when the user has recorded it; `fetch metric:<KEY>`
hydrates it). Each carries its unit and population reference band from the
single signal registry; absence is the usual `{ present: false }`. They
sit off the Coach data inventory by design, so `list_metrics` /
`get_metric_series` do not enumerate them — use the reads above. A new
signal is declared once in the in-app signal registry and shows up across
the metric tools, correlations, and the FHIR map together, so the MCP
surface never advertises a signal the record can't return.
The **longevity lab panel** (ApoB, Lp(a), hs-CRP, HbA1c, fasting glucose,
fasting insulin, eGFR, GGT, ferritin, omega-3 index) is reached like any
other biomarker: `get_labs` (latest per analyte or, with `history:true`, a
paginated trajectory), the `healthlog://lab/{analyte}` resource template,
and `search` → `fetch` (`lab:<analyte>`). Reference bounds come from the
curated biomarker catalogue.
**Off the MCP surface by construction.** The PHQ-9 / GAD-7 mental-health
screeners (item content and totals) are excluded from AI / MCP, and the
| `healthlog://profile` | Health-relevant profile only: age in whole years, gender, height, timezone, unit preferences. Direct identifiers (email, username, role) omitted. |
| `healthlog://medications` | Tracked medications with their schedules (dose, cadence, time windows). Use `get_medication_compliance` for adherence figures. |
| `healthlog://labs/catalogue` | The curated biomarker catalogue with canonical units and **suggested** reference bounds, grouped by panel. Editable defaults, not clinical limits. |
| `healthlog://measurements/inventory` | The same manifest `list_metrics` returns — what data exists and which tool retrieves each domain. Read this first. |
| `healthlog://report/doctor-visit` | A grounded clinician-oriented summary over the last 90 days — vitals (units + bands), medications & adherence, labs. Matches the exported PDF. |
### Resource templates (RFC 6570)
Per-item addresses a host can attach without a tool round-trip. Argument
completion and listing are user-scoped, so a host only ever sees the
metrics / analytes / medications the user actually has — completion
| `medication_check` | Cadence-aware adherence alongside a linked vital series (e.g. did blood pressure track with adherence). Optional `medication` / `metric`. |
| `recovery_check` | Rest mode + illness, recovery/strain composites, resting-HR and HRV baselines (latest vs usual), and the drivers behind them. |
| `glucose_review` | The glucose clinical panel (TIR, GMI, CV%, estimated A1c) plus per-context daily means. |
| `sleep_review` | Per-night asleep + stages, the sleep-rhythm summary (debt + chronotype), and the drivers behind sleep quality. |
| `lab_trend_brief` | Per-analyte lab trajectory — latest vs prior readings against the lab's own stored reference range. One analyte or the whole panel. |
## Security and write model
The MCP surface reuses the existing `hlk_` Bearer model — there is no
parallel identity store. A session is bound to `<user_id>:<token_id>`, so
audit, annotation, and rate-limit buckets are keyed to the specific
credential in use; a single leaked or shared token cannot drain the
account's other tokens.
**Read-only by default.** Only read tools are registered for a
`health:read` session. The write tools are registered **only** for a
`health:write`-scoped session (`ctx.canWrite`) — a read-only token never
sees them advertised at all. The posture is structural, not a runtime
flag a tool could flip.
**Admin is unreachable.**`requireAdmin()` is cookie-only and the MCP
wire carries no cookie, so no token — including a `*` wildcard — can
reach an admin endpoint over this surface.
**Audience-bound writes.** A `health:write` MCP token is bound to the
`/mcp` resource (RFC 8707). The REST resource-server guard refuses it on
every REST write or delete, so it can never become a general REST write
credential. It only admits the in-process write tools below.
### Write tools
Registered only for a write-scoped session. The remote `/mcp` wire is
stateless (no server→client elicitation), so each write tool uses an
| **`APP_URL` is set** | The endpoint fails closed without a pinned canonical origin. `APP_URL` (or `NEXT_PUBLIC_APP_URL`) is the origin every OAuth document and audience check is anchored to — `Host` is never trusted. Without it the surface stays invisible (`404`). |
| **The global API is on** | The operator's instance-wide API switch must be enabled (it is by default). With it off, every token path — including `/mcp` — returns `404`. |
| **The `mcp` module is on** | A per-account opt-in. It is the only module that ships dark; you turn it on in Settings. Until then `/mcp` answers `404` for that account. |
`APP_URL` is the same origin you already use for passkeys and invite
links — for example `https://health.example`. It must be reachable by the
client you are connecting (Claude.ai and ChatGPT connect from their own
servers, so the origin has to be a real public hostname for those; a
LAN-only origin works for Claude Desktop over stdio).
## 1. Enable the module
In the app: **Settings → MCP**. The card has three parts:
- **Enable** — the opt-in switch. Toggle it on; the `/mcp` endpoint goes
live for your account and the card shows your endpoint URL
(`<origin>/mcp`).
- **Connections** — cloud connectors that completed the OAuth flow appear
here; revoke any of them.
- **Tokens** — manual connector tokens (see below).
## 2. Mint a connector token
Still in **Settings → MCP → Tokens**:
1. Choose the scope with the **write** toggle:
- **off** → a read-only token (`health:read`). This is the default and
the right choice unless you actually want the assistant to log
readings.
- **on** → a read **and** write token (`health:read health:write`).
The write scope only ever admits the three confirmed, append-only
write tools (below) and is **bound to `/mcp`** — it can never write
over the regular REST API.
2. Name the token and create it.
3.**Copy the `hlk_…` value now** — it is shown exactly once and cannot
be retrieved again. Tokens default to a 90-day lifetime.
The token never carries the `*` wildcard and never the
`medication:ingest` scope the generic `/api/tokens` mints — it is the
least-privilege MCP scope and nothing more. The same list shows
last-used time and lets you revoke any token.
Cloud connectors that use OAuth (next section) mint the same
`health:read` scope automatically, so for Claude.ai / ChatGPT you do not
need to mint a token by hand — the manual token is for the stdio / power-
user path and for clients you wire up with a static Bearer.
## 3. Connect a client
The endpoint is `<APP_URL>/mcp` for every remote client. A non-browser
client (Claude.ai / ChatGPT server-side) sends no `Origin`, so it
connects directly; a browser `Origin` that does not match `APP_URL` is
refused (`403`) as a DNS-rebinding defence.
### Claude.ai (remote + OAuth)
1. In Claude.ai, add a custom connector and paste `<APP_URL>/mcp`.
2. Claude.ai discovers the OAuth flow automatically: the `401` from
`/mcp` points at `/.well-known/oauth-protected-resource`, which names
the authorization server, and the client walks `/authorize` →
`/token`. PKCE (S256) is mandatory; the built-in bridge accepts
client-id-metadata-document clients with dynamic registration as a
fallback.
3. Approve the consent screen. The connection then appears under
**Settings → MCP → Connections**, where you can revoke it.
No token to paste — OAuth mints and rotates a short-lived `health:read`
access token (60 minutes, with a refresh token) behind the scenes.
### ChatGPT (remote)
1. Add `<APP_URL>/mcp` as a connector. The same OAuth discovery applies.
2. In ChatGPT's **default** (non-Developer) mode, the only tools it will
call are **`search`** and **`fetch`** — these are HealthLog's two-tool
retrieval façade over the same server-authoritative reads. So in
default ChatGPT you ask in natural language ("what's my recent LDL?")
and it searches your record and fetches the matching item with a
citation deep-link back into the app. Developer mode exposes the full
tool catalogue.
### Claude Desktop (stdio)
Claude Desktop runs a local process and speaks the protocol over stdio —
no OAuth, no public hostname. Mint a token (step 2), then add an entry to
`claude_desktop_config.json`:
```json
{
"mcpServers":{
"healthlog":{
"command":"pnpm",
"args":["dlx","tsx","src/cli/mcp-stdio.ts"],
"cwd":"/path/to/HealthLog",
"env":{
"HEALTHLOG_MCP_TOKEN":"hlk_…",
"DATABASE_URL":"postgres://…",
"API_TOKEN_HMAC_KEY":"…",
"ENCRYPTION_KEYS":"…"
}
}
}
}
```
The stdio entrypoint reads the token from `HEALTHLOG_MCP_TOKEN`
(preferred — keeps the secret out of the process argument list) or as the
first CLI argument. With no token it refuses to start; running it with a
minted token **is** the opt-in. It needs database access, so it runs from
a checkout with the same `DATABASE_URL` / `API_TOKEN_HMAC_KEY` /
`ENCRYPTION_KEYS` your instance uses. Use `tsx` via `pnpm dlx` — the
production standalone image strips `tsx`, so this is a local-only tool.
stdout is the JSON-RPC channel; diagnostics go to stderr.
## Trust model — read this before enabling writes
The combination of a **remote** endpoint, **write** capability, and
**OAuth** access is exactly the high-leverage surface to be careful with.
HealthLog ships several structural limits, but the operator's choices
still matter:
- **Only enable the connector for clients you trust.** Anything that
holds a valid token or a live OAuth connection reads your health
record. Revoke connections and tokens you no longer use (Settings →
MCP).
- **Use a read-only token unless you need writes.** The default. A
read-only session never even sees the write tools advertised.
- **Writes are narrow by construction.** The only writes are
append-only, confirmed inserts of a single self-reported reading
(`log_measurement`, `log_mood`, `log_blood_pressure`). There is no
medication, lab, schedule, delete, update, or data-export tool, and a
`health:write` MCP token is refused on every REST write — it can never
become a general write credential.
- **Admin is unreachable over MCP.** `requireAdmin()` is cookie-only and
the MCP wire carries no cookie, so no token — not even a `*` wildcard —
can reach an admin endpoint over this surface.
The structural guarantees are detailed in [the capabilities
"description":"Self-hosted personal health tracker (PWA + native iOS client): weight, blood pressure, pulse, blood glucose, body composition, sleep, mood, menstrual cycle, and medications on one encrypted timeline. AES-256-GCM at rest, no vendor cloud, no telemetry. Ships app + PostgreSQL 16 as one stack.",
"note":"Generate each secret with: openssl rand -hex 32 (a 64-char hex string). For plain-HTTP / LAN access set SESSION_COOKIE_SECURE to false, or login fails. Set NEXT_PUBLIC_APP_URL and APP_URL to how you reach the app. Full guide: https://github.com/MBombeck/HealthLog/blob/main/docs/self-hosting/getting-started.md",
<ConfigName="WebUI Port"Target="3000"Default="3000"Mode="tcp"Description="HTTP port HealthLog listens on."Type="Port"Display="always"Required="true"Mask="false">3000</Config>
<ConfigName="DATABASE_URL"Target="DATABASE_URL"Default="postgresql://healthlog:CHANGE-ME@POSTGRES-HOST:5432/healthlog?schema=public"Mode=""Description="Connection string for your separately-installed PostgreSQL 16. Replace CHANGE-ME with the DB password and POSTGRES-HOST with the Postgres container IP or name."Type="Variable"Display="always"Required="true"Mask="true"/>
<ConfigName="ENCRYPTION_KEY"Target="ENCRYPTION_KEY"Default=""Mode=""Description="AES-256-GCM key for data at rest. Generate with: openssl rand -hex 32 (64-char hex). Back this up — losing it makes encrypted data unrecoverable."Type="Variable"Display="always"Required="true"Mask="true"/>
<ConfigName="API_TOKEN_HMAC_KEY"Target="API_TOKEN_HMAC_KEY"Default=""Mode=""Description="HMAC key for API token hashing. Generate with: openssl rand -hex 32 (64-char hex)."Type="Variable"Display="always"Required="true"Mask="true"/>
<ConfigName="SESSION_COOKIE_SECURE"Target="SESSION_COOKIE_SECURE"Default="false"Mode=""Description="Set to false for LAN / plain-HTTP installs (http://tower:3000) or login fails. Set to true only behind an HTTPS reverse proxy."Type="Variable"Display="always"Required="true"Mask="false">false</Config>
<ConfigName="NEXT_PUBLIC_APP_URL"Target="NEXT_PUBLIC_APP_URL"Default="http://tower:3000"Mode=""Description="Public URL users reach HealthLog at. Use the LAN URL (e.g. http://tower:3000) or your HTTPS domain behind a reverse proxy."Type="Variable"Display="always"Required="true"Mask="false"/>
<ConfigName="APP_URL"Target="APP_URL"Default="http://tower:3000"Mode=""Description="Internal app base URL. Normally identical to NEXT_PUBLIC_APP_URL."Type="Variable"Display="always"Required="true"Mask="false"/>
<ConfigName="ENCRYPTION_KEYS"Target="ENCRYPTION_KEYS"Default=""Mode=""Description="Optional. Key-rotation map, e.g. {"v1":"<hex>","v2":"<hex>"}. Leave empty unless rotating; use ENCRYPTION_KEY instead."Type="Variable"Display="advanced"Required="false"Mask="true"/>
<ConfigName="ENCRYPTION_ACTIVE_KEY_ID"Target="ENCRYPTION_ACTIVE_KEY_ID"Default=""Mode=""Description="Optional. Active key id for the ENCRYPTION_KEYS map (e.g. v2). Leave empty when using ENCRYPTION_KEY."Type="Variable"Display="advanced"Required="false"Mask="false"/>
Loaded 100 of 1287 files, more files were not shown because too many files have changed in this diff.
Show more
Reference in new issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Marc-André Bombeck