mirror of
https://github.com/MBombeck/HealthLog.git
synced 2026-11-08 20:42:43 +00:00
Marc-André Bombeck
10cfdf8b66
A scan sent the whole report to one call under an 8000-token ceiling and validated the reply against a 100-row envelope. A long multi-panel report lost everything either way: the reply stopped mid-row and failed to parse, which the person saw as "try a clearer photo", or it held more than 100 rows and the schema refused all of them. One malformed row refused the lot as well. Reasoning models made it worse, because the call sent no reasoning setting and OpenAI, Codex and Claude's 5 generation reason by default against the same output allowance. A PDF of more than one page is now rendered and read a page per call on every provider (up to 20 pages), as many reads as the run's deadline allows, and the pages are merged in order with a reading printed twice listed once. A photo, a one-page PDF on a provider that reads PDFs, and OCR'd text up to 4000 characters stay one read; longer text is split at line boundaries. Each read carries an explicit reasoning off, so a model that can stop reasoning does and one that cannot gets its minimum budget on top of the answer's. A reply cut at the output limit is detected from the finish reason (Codex's incomplete stream included), keeps the rows that closed before the cut and is named in the result instead of retried; when no read produced a row the run fails as labs.ocr.outputLimit. A provider failure after the first page keeps the pages already read. Rows are validated one by one, the row cap is 400 for the scan and the commit, and a report past it keeps its first 400 and says how many it had. The run settles what the provider reported it spent rather than the fixed reservation, so a page-by-page read still counts against the day's budget in full.
src/lib/
The server and runtime toolkit. Everything that is not a page, component, or route handler lives here: the API kit, auth, crypto, the rollup tier, AI providers, notifications, background jobs, and the integration clients. Route handlers and RSC pages compose these modules; the modules never reach back into app/.
Core request kit
api-handler.ts— theapiHandlerwrapper every route uses. Auth resolver (cookie OR Bearer),requireAuth/requireAdmin(admin is cookie-only by construction), idempotency plumbing, Wide-Event builder, GlitchTip forwarder.api-response.ts— the{ data, error, meta? }envelope:apiSuccess/apiError,safeJson, trusted-proxy IP resolver,returnAllZodIssues+sanitiseZodIssues.rate-limit.ts— Postgres-backed sliding-window limiter (single atomic upsert); multi-instance correct.idempotency.ts— user-scoped idempotency keys that refuse to cache secret-shaped responses.query-keys/— the only legal source of TanStack QueryqueryKey/mutationKeyarrays (enforced by an ESLint rule); per-feature files behind theindex.tsbarrel.safe-fetch.ts— the one documented outbound-egress entry (redirect: "manual"+ timeout; optional connect-time DNS-rebinding pin). Rawfetch(underlib/+app/is banned by ESLint.db.ts— the Prisma client singleton.crypto.ts— AES-256-GCM at rest with versioned key ids, fail-closed loader, rotation primitives.
Subsystems
auth/— Postgres sessions, secure-cookie flag policy, passkey (WebAuthn) + Argon2id password, per-device refresh-token rotation, HMAC Bearer tokens.ai/— multi-provider insight + Coach engine:provider-chain.ts, hand-rolled{openai,anthropic,local,codex,mock}-client.ts,coach/,prompts/, citation + confidence helpers.rollups/— DAY/WEEK/MONTH/YEAR pre-aggregations for measurements, mood, and medication compliance. Read-swap pattern: probe the rollup, fall back to live SQL on a coverage miss.notifications/— the dispatcher cascade (APNs → Telegram → ntfy → Web Push), per-channelsenders/, retry policy, VAPID config.jobs/— pg-boss workers: reminders, insight pre-generation, backups, retention sweeps, integration backfills. Recurring work belongs here, not inscripts/.integrations/,withings/,whoop/,fitbit/— OAuth clients and sync services for the device providers;sync/holds the shared delta-feed plumbing.analytics/— trend calculations, cadence-aware medicationcompliance.ts, correlations.insights/,fhir/,doctor-report-*.ts— insight pipeline, FHIR R4 export, client-side doctor-report PDF.openapi/— Zod registry that is the source of truth fordocs/api/openapi.yaml.logging/,observability/,monitoring/— Wide-Event transports and the central redaction denylists (redact at the egress boundary, not the call site).validations/— shared Zod schemas, including the SSRFisPublicUrlfloor.i18n/,tz/+timezone.ts,format*.ts— translations, timezone helpers (Berlin for display, UTC in DB), locale formatting.
Conventions
Kebab-case filenames. See ../../CLAUDE.md "Code conventions" and "Security-relevant patterns"; the critical-files map there cites the exact entry points. Most modules carry tests in a sibling __tests__/.