Every staged fact carried a sourceText the model echoed back, stored verbatim
as 'the exact verbatim span of the document the fact came from'. Nothing
checked that the span existed. A paraphrase, a normalised rendering, or an
outright invention was persisted and shown to the reviewer as a transcription
of their own clinical record — an anchor pointing at nothing, or at a span the
document never contained.
Treat the echo as a lookup key instead. anchorSourceText resolves it against
the extracted text (exact, then whitespace-collapsed and case-folded to absorb
OCR reflow) and stores the span the document actually carries, with its offset.
A quote that matches nothing is dropped and the fact is flagged unanchored;
vision mode has no extracted text at all, so its facts are unanchored by
construction. Rows written before this resolve to unanchored on read — their
quote is preserved but never presented as checked.
The full-backup export (plaintext + passphrase-encrypted, and the weekly
auto-backup) only ever covered measurements, medications, mood, and cycle
data. Labs, illness episodes, allergies, family history, and workouts were
silently missing, and the export UI implied "everything" was included.
Adds a shared records-backup section (src/lib/export/records-backup.ts) that
both the on-demand export routes and the weekly data-backup worker read from:
lab results + the biomarker catalog, illness episodes (including flares via
parentConditionId) with their day-logs, allergies, family history, and
workout summaries. Every read is owner-scoped and soft-delete-filtered;
encrypted free-text columns decrypt fail-soft so one bad row never aborts
the whole backup.
Document original files and workout GPS/per-sample time series stay
deliberately out of the JSON payload (size). The payload now carries an
explicit manifest disclosing exactly that, and the export UI states it in
plain text instead of implying completeness. These new domains are
export-only for now — admin restore still only recreates the original
measurements/medications/mood/cycle domains; the payload validates and
reports counts for the rest, but restoring a snapshot does not yet recreate
them.
Also fixes the admin backup-upload total to count every domain, not just
the original four.
Two directions, folding the vault and labs pipelines into one linking path
without weakening either manual flow.
Vault document → lab values (auto-stage). After the per-document auto-index
job transcribes a stored document, a lab-looking document is run through the
same inbound extraction the manual "Extract" button uses, but the facts land
PENDING for the existing human review — nothing is committed. Gated on both
the documents and labs modules plus the auto-AI-read consent gate the
auto-index external path already uses; a text-structuring pass over the
transcription the index just produced, so no second vision call is spent. The
staging is a guarded STORED→EXTRACTED transition, and a keyword heuristic (or
an explicit LAB_RESULT kind) keeps an arbitrary photo from burning a call.
Everyone without the opt-in keeps the manual extract button.
Lab scan → vault. A lab report scanned in the labs area is now also filed
into the Documents vault (kind LAB_RESULT) through the existing upload —
encrypted at rest, EXIF-stripped, thumbnailed, sha256-deduped — and the
committed labs are cross-linked back to it with the same APPROVED-observation
provenance the manual vault → labs path produces. Text-mode scans keep the
image on-device and file nothing. Idempotent: a re-commit inserts no rows and
links nothing new, and the vault dedupe means no duplicate document.
Owner-scoped and module-gated on both sides; the cross-link is best-effort so
it never fails the authoritative write.
Introduce an operator-level ChatGPT-subscription (Codex) credential that the
operator connects once and shares server-wide, plus a per-user switch to route
AI egress through it.
- Schema (migration 0235): three encrypted columns on AppSettings for the
operator's central Codex access/refresh tokens and the ChatGPT-Account-ID
claim, plus expiry/connected-at/status; a per-user useCentralCodex flag
(default false). Register the new encrypted columns in the rotation registry
and script.
- Resolution: add the admin-codex chain type; append it to a user's resolved
chain only when they opted in AND the operator connected it. Tokens refresh
server-side against the singleton row (CAS-style), same primitives as the
per-user Codex path.
- Budget: bill an admin-codex primary against the operator cost cap, not the
user plan — it drains the operator's account.
- Consent: admin-codex is server-managed external egress, gated by the existing
consent machinery (self-snapshot and document surfaces alike). No existing
consent path is weakened.
- Document order: rank admin-codex with the subscription tier (last), so the
document class stays local-first.
- UI: admin device-code connect/disconnect card (cookie-only admin routes) with
honest copy on account pooling, shared rate limits, and training-by-default;
a per-user opt-in switch shown only once the operator has connected it, off by
default, behind a one-time honesty confirm.
- i18n across all six locales; unit tests for the chain type, budget cap,
consent gate, vision capability, and document order; a mocked e2e for both
the admin connect surface and the per-user switch.
The central credential's tokens are encrypted at rest and never returned to a
non-admin API; the opt-in only takes effect once the operator has connected the
shared account.
Lab-report OCR now rasterizes a PDF to page images for any non-Anthropic
vision provider, exactly as the document vault already does — Anthropic keeps
its native PDF block, every other vision provider (codex/OAuth, OpenAI, local)
reads the rendered pages. The capability probe reports `pdfSupported` whenever
the picked provider reads PDFs natively or the server-side rasterizer is
available, so the "scan a PDF" affordance shows for a codex provider too.
Drop the dedicated document-scan (Lab-OCR) provider: labs OCR always uses the
main configured provider chain. Removes the settings card, the `aiOcr*`
resolution branch, the GET/PATCH handler fields, the OpenAPI schema fields, the
i18n keys across all locales, and the five backing `User` columns (migration
0234). The on-device tesseract fallback is untouched.
Dashboard hero: stack the overdue-dose action directly beneath the verdict
text (tight gap, left-aligned) on every viewport instead of pushing it to the
far container edge on desktop.
The cycle/illness/records note DTOs, the biomarker context, and the custom
mood/cycle labels each degraded an undecryptable value to a blank/generic
fallback with no signal — a key-rotation gap read as 'left blank' and the
catch was silent. Keep the fail-soft read posture (one bad row never 500s
the list) but emit a wide-event on each swallowed decrypt so a systemic
key gap surfaces. The visible per-field 'unreadable' marker in the UI is
the deferred remainder.
Add a document-class provider resolution that demotes external/train-by-default
providers below local (local -> BYOK no-train API -> operator key -> codex last)
and a pick-based document-egress consent gate that requires an active receipt
for any external provider. Coach/insights keep the app-wide cost-first order.
Co-Authored-By: Marc-André Bombeck
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.
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).
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.
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.
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.
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.
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.
One-shot prettier pass so the new format:check gate starts from a clean
baseline. Formatting only — no behaviour change. Generator-owned and
vendored artifacts stay untouched via .prettierignore.
Add a 'showing latest N of M' hint to the lab list and biomarker detail
when the 500-row cap truncates. Drop the inaccurate amend comment from the
edit history, justify the inline-SVG vs Recharts sparkline split, remove
the unused BiomarkerPanelKey export, and document the deliberate UX-only
(non-server) module gate on the labs + biomarkers routes. New labs i18n
keys added across all six locales.
The catalog GET eagerly threw on the first un-decryptable context row,
500ing the whole list. The list serializer now decrypts fail-soft (null
on a bad-key/malformed row) via decryptContextSoft, bounding the
fail-closed blast radius to one marker; the single-resource GET stays the
authoritative throwing read for the plaintext context.
A PUT that moves a single reference bound past the row's stored other
bound slipped through the schema refine (which only fires when both bounds
arrive together). Both the lab-result and biomarker routes now merge the
effective bounds and 422 on low > high. Editing catalog-resolved fields
(analyte/unit/panel/range) on a linked reading is also rejected rather than
silently no-oped. Shared isInvertedRange/effectiveBound helpers with tests.
The edit sheet loaded the decrypted note from the single-resource GET and
sent note:null on a load miss, silently wiping a note that simply could
not be read. The submit path now omits the note key entirely in that case
(server preserves the stored ciphertext), disables the textarea, and shows
a hint. Decision encoded in a pure resolveNoteForUpdate helper with tests.
A free-text lab write resolves or mints a user-scoped biomarker by
(userId, lower(analyte)) so no LabResult row ever persists unlinked. An
existing marker is reused as-is (its unit/range stay authoritative); a
miss mints one from the reading's own spelling. Every reading is now
editable and detail-navigable, and the boot-time backfill becomes a pure
pre-upgrade migration. Closes the unlinked-row dead-end and the contract
hole in one path.
Add a user-scoped Biomarker catalog so a marker is defined once — name,
unit, lower/upper bound, optional encrypted context, panel — and every
later reading just links it. New /api/biomarkers CRUD plus a structured
/api/labs create path: a biomarkerId resolves the unit + reference range
server-side, so the list and detail DTOs carry the resolved values and
the client never recomputes the range. Lab reads include the linked
marker and fall back to the legacy free-text fields for unlinked rows.
Add a lab-result restore endpoint for the delete-Undo affordance, a
shared serialiser, a static common-panel seed list, and tests for
catalog validation, structured-entry resolution, and the context codec.
Reference-range classification + low/high range rendering was implemented three
times (validations, PDF core, lab list). Extract a single classifier and a
structure-only range formatter that takes a per-surface number formatter, so
each surface keeps its own number formatting while sharing the canonical
classification. Behaviour is unchanged per surface and now covered by one test.
The annual-blood-panel auto-resolve that findLatestLabResultForAnalytes was
meant to feed never got wired into the Vorsorge reminder, so the export sits
with zero callers and the labAnalyteTrend query-key has no consumer (the
sparkline is prop-fed and there is no labs trend endpoint). Remove both so the
knip gate is clean; the Vorsorge lab read-back can come back with the wiring
that needs it.
Record discrete biomarker readings (HbA1c, LDL, ferritin, TSH, …) with
their value, unit, reference range and sample date. The store pairs with
the annual blood-panel reminder, which until now had nowhere to record
the result it reminds the user to obtain.
The list/create/update/delete endpoints narrow userId from the session,
build the write data field-by-field, soft-delete via deletedAt, and
encrypt the optional note at rest. A neutral three-state reference-range
classifier (in-range / below / above / unknown) is computed
server-side so every surface renders the same calm verdict.
findLatestLabResultForAnalytes exposes the read point the blood-panel
reminder consults to mark itself satisfied, without coupling the two.
Marc-André Bombeck