ARETE

One calm system for the day.

A mobile-first PWA that unifies habits, goals, tasks, focus sessions, meditation, breathwork, calorie and macro logging, flexibility, cognitive practice and adaptive workouts — coordinated by a single Today screen and a single evening review.

Working title. Renaming is a one-line change in src/lib/app-config.ts.

---

Status

Stage 13 of 17 — Safety, privacy, accessibility and trust. There is a real schema: 61 tables, row-level security on every one of them, and an authorization test suite that runs against a real Postgres and actually tries to read another user's data. The app reads the signed-in user's own local day through a repository interface with two interchangeable implementations.

Onboarding asks eight skippable questions and generates a first week from the answers — capped, explained item by item, and saved only once the user has edited it and agreed.

Today and Plan are live. Today holds the morning intention and the dichotomy-of-control prompt, the day's commitments with complete, skip-with-reason, move-to-tomorrow, reorder and quick-add, a minimum-day switch, a focus timer whose state lives on the server, and the evening review. Plan holds goals, tasks across four views, and habits of four kinds.

Mind is live too: five secular meditations, five slow breathing patterns with the safety copy on the screen where the decision is made, and two cognitive drills that report practice at that drill and explicitly nothing else. An optional meditation pack exists and is off — its practices are absent until someone chooses it.

Train is live: a deterministic plan engine that fits the week to the days, equipment, time and pain flags a person gives it (tennis days included), explains every rule it applied, and progresses only when the logged sets say so. Workouts are started, logged set by set with effort and pain, swapped within a movement pattern, and finished. Nine exercises have an original, data-driven SVG mannequin animation — tested for what the movement should look like — and every one of the fifty has a full text instruction.

Flexibility is live in demo mode: a four-week hamstring programme built from the source's structure in ARETE's own words, three sessions a week on days the person chooses, done in any order, with a warm-up gate, a pain exit that is always on screen, and weekly self-reports compared in words. It is used with the source author's permission.

Food is live: meals and snacks logged with numbers or without, quick add, custom foods and recipes, favourites, copy the day before's meal, water, and daily and weekly views. Targets are ranges with their working shown, overridable, and never a budget. Safeguards for under-18s, pregnancy and postpartum, eating-disorder risk and prescribed diets remove what should not be there, and every number can be hidden. Demo foods say their numbers are synthetic. An optional weight trend and an optional ingredient/oil awareness preference are both off by default.

Programmes are live: a 60-day reset whose every number is a range fitted to the person and editable before and during, with pause, travel, sick days, minimum days, weekly check-ins and an optional extension week for the source's days 61–66. The Stoic journal runs through Today's morning and evening questions into a weekly review that sets the week's facts beside the person's reflection. Contrast therapy is an optional, screened pack with a conservative timer that is always one tap from stopping.

Progress is live: date ranges, a plain-language narrative, one chart per module with a text summary and a table, streaks beside rolling consistency, observations that are labelled as such and never causal, and an export of the range. The weekly review proposes next week from open tasks, habits, training and recovery, and adds only what the person ticks. Notifications are quiet by default: per-module switches, quiet hours, a daily cap, bundling, snooze and done, and a practice reminder that is off unless chosen and only ever in the person's own words.

Stage 13 hardened all of it. One module, domain/eligibility.ts, decides what every gated feature may offer, and the strictest answer wins; Settings → Safety shows each decision and its reason. Every screen that asks for effort shows how to stop. Settings → Privacy deletes any one part of the account, and an audit trail records sensitive actions without their content. Every route is checked with axe, by keyboard, with reduced motion and at 200% text. Release is still blocked on clinical and dietitian review (docs/SAFETY_MATRIX.md, docs/THREAT_MODEL.md).

See docs/PROJECT_LEDGER.md for where the build stands and what comes next, docs/DATA_MODEL.md for the schema and the reasoning behind it, docs/PLAN_ENGINE.md for how answers become a week, and docs/DESIGN_SYSTEM.md for the visual system.

Requirements

• Node.js ≥ 20.9
• npm 10 (the repo pins packageManager; there was no lockfile before this stage)

Getting started

npm install
cp .env.example .env.local   # optional: the defaults already work
npm run dev                  # http://localhost:3000 -> /today

No credentials are needed. The app runs in demo mode by default: a fixed, deterministic account served through the same repository interface real data uses, with writes that work and last until the server restarts. The UI says so; it is never presented as durable.

For live mode you need a Supabase project and three environment variables — see docs/PROJECT_LEDGER.md §8 for the twenty-minute setup. ARETE refuses to start in live mode half-configured, and names what is missing, rather than quietly serving demo data as if it were the account.

Commands

Command	What it does
npm run dev	Development server
npm run build	Production build
npm run start	Serve the production build
npm run lint	ESLint (flat config, eslint-config-next)
npm run format / format:check	Prettier
npm run typecheck	tsc --noEmit, strict
npm test	Vitest unit, component and database tests
npm run test:coverage	Vitest with V8 coverage
npm run e2e	Playwright smoke tests at phone and desktop widths
npm run e2e:install	Install the Chromium browser Playwright needs
npm run db:migrate	Apply every migration, in order, from zero
npm run db:seed	Content; plus the demo account outside production
npm run db:reset	Both
npm run verify	Everything CI runs, in order

The database commands read SUPABASE_DB_URL, the service-role connection string. They are deployment work and never run on the request path.

Architecture

src/
  app/            Next.js App Router — routes, layouts, loading/error/not-found
  components/     UI primitives, icons and navigation. No data fetching, no vendor SDKs
  design/         Design tokens and WCAG contrast maths. Pure data; the tests assert it
  domain/         Types, local days, recurrence, units, idempotency. No React, no database
  auth/           Who is this? Demo identity, or a Supabase session
  persistence/    Repository interface, demo adapter, SQL adapter. The only layer SQL touches
supabase/
  migrations/     The schema: 37 tables, row-level security on all of them
  test/           An auth-schema stand-in, for the tests only. Never applied to a real database
  content/        Content packs as validated data, each carrying provenance
  analytics/      Typed events and a no-op sink. No free text ever leaves the device
  integrations/   External services, each behind an interface. Currently empty by design
  lib/            Environment validation and app config
e2e/              Playwright smoke tests
docs/             Source map, product spec, ledger, acceptance checklist

The dependency direction is one-way: app → components → domain, with persistence and content reached only through their own interfaces. A component never imports a database client, and the domain layer never imports React.

Routes

Every destination is live. The full table, with what each route does, is in docs/PROJECT_LEDGER.md §4: Today, Plan, Train (with exercises, flexibility and food), Mind, Progress, Reflect, Programmes (the reset and contrast therapy), onboarding, and Settings (notifications, your data, privacy, safety and modules). /api/export downloads the signed-in person's data; /api/health reports build, version and mode, with no secrets.

Principles this codebase is held to

• No fake anything. No mock buttons, no dead links, no writes that vanish, no silent failures. A feature that is not built says so. A list row is pressable only when it has somewhere to go.
• Accessibility claims are computed, not asserted. Every colour pair the product renders is checked against WCAG AA in the test suite, and the stylesheet is parsed and compared against the token source so the two cannot drift.
• Security claims are tried, not stated. The authorization tests run against a real Postgres and attempt to read, forge, update and delete another user's rows.
• Rules that can be constraints are constraints. No analytics event can hold free text, no quote can carry an unverified author, no skip can lack a reason — enforced in the database, where application code cannot forget.
• Calm and non-shaming. A missed day is quiet, not red. The word "streak" does not appear in copy about a missed day — there is a test for that.
• Health features are self-management tools, not diagnosis or treatment. Cold exposure and contrast therapy are opt-in and safety-gated.
• Religious content is never assumed from identity, name, language or location.
• Cognitive drills report practice at those drills and never claim IQ growth or transfer.
• Source PDFs are reference material. Every content item carries provenance, and docs/SOURCE_MAP.md records what may and may not be shipped.

Privacy

No third-party analytics, no ad SDKs, no session replay. Journal text, reviews, food, body measurements, injuries, health answers and the practice reminder never appear in an analytics payload — not even by name — enforced by schema tests, not by a convention. Any part of the account can be deleted on its own, or all of it at once. See docs/PRIVACY.md.