Palier

Palier is a free, source-available web app for preparing for the Public Service Commission's Second Language Evaluation (SLE). It covers all three tests: reading comprehension, written expression and oral proficiency. It is in French, for public servants aiming at level B or C. You get timed drills and mock exams laid out like the real tests, AI feedback on your writing, and spoken practice with an AI examiner that asks, listens and follows up.

Try it: palier.study. There is no account to make and nothing to install.

Palier is an independent, source-available study tool. It is not affiliated with, endorsed by, or connected to the Public Service Commission of Canada. It contains no real test questions and its results are not official.

What you can do

Feature	What it is	Needs a key?
Reading practice	Timed sets on workplace passages. You can answer from the keyboard alone	No
Written expression practice	Drills on the grammar and usage the test checks, with a rationale for every option	No
Mock exams	Full timed forms built to a real variant's published structure and scored against its published cut table	No
Review	Items you missed come back on a Leitner schedule	No
Library	Short articles on each written-expression sub-skill	No
Progress	An estimate of your level shown with its uncertainty and what it rests on, a trend over time, and a printable page	No
Diagnostic	A placement test that ends in a written result and a starting level	Yes
Writing workshop	Write a workplace text and get feedback on it	Yes
Generated practice sets	New written-expression items made on demand	Yes
Oral, practice mode	Answer out loud one turn at a time, then get a report	Yes
Oral, studio mode	A live spoken conversation with an AI examiner	Yes

The interface is fully bilingual. Every screen is in English and French, given equal prominence.

What it costs, and where your data goes

Practising is free. Drills, review, mock exams and progress need no key and cost nothing. They work offline once the site has loaded.

The AI features run on your own OpenAI API key, sent from your own browser. Each use costs a few cents, and OpenAI bills your account directly; Palier charges no subscription. Spoken practice is billed by the minute. Each feature shows its estimated cost before it spends anything, and a meter shows what you have spent.

• Your browser is the system of record (ADR 1). Your answers, recordings, transcripts and writing stay on your device.
• Your key is stored encrypted on your device and goes only to OpenAI (ADR 2). There is one exception. To open a realtime voice session for studio mode, the key passes through a single stateless server call that keeps nothing (ADR 3). If you want to avoid that, deploy your own one-file endpoint from selfhost/.
• Sync is on by default and needs no sign-in (ADR 4). Your progress follows a random id with no name or email attached, and you link devices with a pairing code. One switch turns sync off. Exporting, importing and deleting your data each take one action.
• Palier uses no third-party analytics or error reporting (ADR 15).

To report a vulnerability, see SECURITY.md.

Where the questions come from

Every item is original, and none is copied or adapted from a real SLE or from PSC material. The current bank (content/bank/v4) holds 868 items on 84 passages, 30 oral scenarios and 12 mock-exam forms. Claude wrote the items in the register of the federal workplace. Separate Claude instances then reviewed each one blind: a reviewer had to find the intended answer and no second one. Each item then passed a set of deterministic checks. An item that fails is discarded, not repaired (ADR 24). Every item shows where it came from, and reporting a bad one takes one tap.

Items written by people are welcome, and they go through the same checks. See Contributing.

Status

The app is live and in use. Phases 0 to 7 of the implementation plan are built: practice, mock exams, item statistics, the key-backed features, and both oral modes. The steps left before 1.0 are manual checks and decisions for the maintainer. The English mirror (Phase 8), which practises English as a second language, has not started. docs/progress.md is the source of truth for what is done, what is in flight and what is next.

Running it locally

You need Node 22 and pnpm 10. Corepack picks up the version from package.json.

pnpm install
pnpm dev          # builds the packages, then runs next dev

Open http://localhost:3000.

No environment variables are needed. Without a database, practice, mock exams, progress and export all work, because they live in the browser, and only the sync routes answer 503. To try the AI features, paste your own OpenAI key in Settings. To put Palier on the internet, with Postgres for sync, see docs/deploy.md.

How it is built

Palier is a pnpm and Turborepo monorepo, written in TypeScript. It is a practice engine for any banded language exam, and the SLE is its first profile. Every exam-specific rule lives in content/profiles/psc-sle.json rather than in code: item counts, time limits, cut scores, level descriptors, the sub-skill taxonomy and the review intervals (ADR 9). If you swap the profile and the bank, you have a practice app for a different exam.

The code is split into six packages behind ports and adapters, and dependencies point strictly inward (ADR 10). The core does scoring, trend, selection and scheduling, and it is pure: deterministic functions over plain data that know nothing of React, IndexedDB, OpenAI or HTTP. dependency-cruiser fails the build on any import that points the wrong way. The item bank ships as static JSON (ADR 18). The web app is Next.js on Vercel (ADR 11), and Postgres holds only the sync replica (ADR 21).

Workspace	Depends on	Contains
apps/web	everything	The Next.js app: routes, API handlers, and the one composition root
apps/factory	domain, adapters	The content-pipeline CLI: harvest, draft, adversarial review, deterministic validation, then the bank build
packages/domain	—	Types, invariants, Zod schemas for every content artefact, and the exam profile loader
packages/engine	domain	The pure core: scorer, band mapper, trend calculator, Leitner scheduler, selector, daily planner, oral session machine
packages/app	domain, engine	Port interfaces and the use cases that consume them
packages/adapters	app, domain	Concrete adapters: Dexie, bank, OpenAI, sync, telemetry, ids
packages/ui	domain	Design tokens, primitives, and item renderers registered by type
packages/testing	app, domain	In-memory ports, port contract suites, fixture builders, and a fixture bank
content	—	The exam profile, the item bank, the library and authored items. Data, not code
selfhost	—	Not a workspace: two one-file realtime secret endpoints that users deploy on their own accounts
docs	—	Requirements, architecture, plan, ADRs, progress, deploy

Commands

Command	Does
pnpm dev	Builds the packages, then runs next dev
pnpm build	Builds every package to its own dist, then runs next build
pnpm verify	Runs check-types, lint, boundaries and test, in that order. Nothing merges that does not pass it
pnpm check-types	Type-checks every workspace
pnpm lint	Runs ESLint across the repo with zero warnings allowed
pnpm boundaries	Runs dependency-cruiser on the import graph
pnpm test	The fast lane: one Vitest process across every project, with coverage thresholds
pnpm test:integration	The integration projects
pnpm test:e2e	The Playwright end-to-end suite for apps/web
pnpm verify:medium	Integration, then end-to-end
pnpm mutation	Mutation testing with Stryker
pnpm clean	Removes dist, .next and the build info

A few conventions apply everywhere. Use named exports only; the root eslint.config.mjs lists the framework files exempt from that rule. Keep strict, noUncheckedIndexedAccess and exactOptionalPropertyTypes on, set repo-wide from tsconfig.base.json. Packages are ESM with NodeNext.

Documentation

Read	For
docs/product-requirements.md	What the product must do: the 14 requirements, the exam model, screens and content standards
docs/architecture.md	How it works: data model, ports, sync, the security model, engine algorithms
docs/implementation-plan.md	Module boundaries, testing strategy, phases, definition of done
docs/content-factory.md	How the item bank is made
docs/adr/	Why each significant choice was made, and what would reverse it
docs/progress.md	Where the build is: status, deviations, the session log, the next slice
docs/deploy.md	Deploying, smoke checks and rollback

docs/README.md explains how the documents fit together and which one wins when two disagree. An agent working in this repository should start at CLAUDE.md. AGENTS.md points there too.

Contributing

Items, passages and code are all welcome. Start with CONTRIBUTING.md. It opens with the rule that matters most: every item is original, with no real test material. It then shows how to write an item in content/authored/, the quality bar the item must clear, and what happens to it after you open a pull request.

Licence

Palier is source-available but not open source in the OSI sense. You may read, copy, change and run it for any non-commercial purpose. Commercial use is not permitted.

• The application code is under the PolyForm Noncommercial License 1.0.0 (LICENSE). It expressly permits personal study and use by educational institutions, charities and government institutions.
• The content is under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0, LICENSE-CONTENT). The content is the item bank, passages, exam forms and other practice material.

See ADR 23.