Personal nutrition tracker — Next.js + Drizzle + d3 over PostgreSQL, sized to run (app + postgres + nginx) inside a 1 GB GCP e2-micro.

• Dashboard — three d3 charts over daily_summaries: calories vs weight (dual axis, 30-day rolling average, morning/evening weight band), macro breakdown (stacked kcal: protein / other carbs / sugar / other fat / saturated fat), and nutrients normalized to % of RDI.
• Daily Entry — searchable food + meal selector with live nutrient preview, inline-editable log entries, collapsible meal entries, calories remaining, 8 macro progress bars vs per-nutrient target/limit modes, sticky daily targets, and independent morning/evening weight tracking.
• Food Database — master-detail CRUD with per-100g / per-item unit types.
• Weekly Planner — plan a week of meals (paint one meal across many days), see weekly macros + per-day verdicts, apply today's plan into the log (today-only), and compare plan vs actual. Plan is intent, kept separate from the log.
• Auth — Google sign-in (Auth.js, JWT sessions) with an email allowlist.

• Next.js (App Router, standalone output) + TypeScript
• Drizzle ORM + postgres.js (pool max 4 — tuned for the 1 GB VM)
• d3 submodules only (d3-scale, d3-shape, d3-array)
• vitest + PGlite (data-layer tests run against the real schema)
• PostgreSQL 15 — schema in database/init.sql (unchanged from v1; updated_at is trigger-maintained, never set by the app)

# 1. local postgres (applies database/init.sql on first run)
docker compose up -d

# 2. env
cp .env.example .env   # fill in Google OAuth credentials + allowlist

# 3. run
npm install
npm run dev            # http://localhost:3000

Tests and checks:

npm test               # vitest: domain logic + PGlite data layer
npm run typecheck
npm run build          # standalone production build

These mirror the original Python app and are guarded by tests in tests/data/storage.test.ts and the parity scripts in scripts/parity/:

1. An empty day writes NULL nutrients to daily_summaries, never 0.
2. saveDailyEntry never touches the weight columns; weights are written only by updateMeasurements.
3. Saving a day is delete-all-for-date + reinsert; meal entries are food_entries rows sharing a meal_id.
4. Daily targets are sticky: existing row → most recent earlier row → defaults.

Production is a GCP e2-micro (1 GB RAM) running postgres + nginx + the app. next build never runs on the VM — CI (or scripts/deploy.sh) builds the standalone output and rsyncs it to ~/apps/nutritional/releases/<sha>, then repoints the current symlink and restarts the systemd unit (deploy/nutritional-next.service, memory-capped at 448 MB).

• Deploy & operate (routine): deploy/README.md
• First-time VM setup: deploy/SETUP.md
• DB ops (backup/restore/tuning): database/ — db.sh, README.md, postgresql.conf.template