Tidy tests

A behaviour-preserving refactoring of test code only. The rules are in docs/development/conventions.md ("Tests"). Production code is not touched — if a test can only be simplified by changing production code, note it and stop there.

1. Baseline (before any edit)

Record, for the target package(s):

go test ./<pkg>/... -count=1 -v 2>&1 | grep -E '^(=== RUN|--- (PASS|FAIL|SKIP))' > before-cases.txt
go test ./<pkg>/... -count=1 -coverprofile=before.cov && go tool cover -func=before.cov | tail -1

Frontend: npx vitest run <path> --reporter=verbose > before-cases.txt and npm run test:coverage -- <path>. Keep these files outside the repo (scratch directory).

2. Plan the restructuring

List what you will change and why, grouped:

• Split a file over ~800 lines by capability (e.g. router_test.go → router_pages_test.go, router_auth_test.go, …, shared setup in router_fixture_test.go).
• Tables: groups of tests with the same setup and assertion that differ only in input and expected output. Name every row.
• Fixture: one default fixture per package with explicit overrides; t.Cleanup for every opened store; build only what each test needs.
• Names to Test<Type>_<Scenario>_<Expectation>; dead helpers removed.

Don't table cases with different setup or different assertions. No generic assertion DSL. Show the plan to the person if it touches more than one package.

3. Restructure in small steps

One kind of change per step, each meant to become its own commit (test: split router tests by slice, test: table-drive slug validation, …), the suite green after each. Commit each step locally (Conventional Commits, suite green before each); don't push. The PR is capped at ~1,000 changed lines; a big file is split over several PRs.

4. Prove nothing was lost

go test ./<pkg>/... -count=1 -v 2>&1 | grep -E '^(=== RUN|--- (PASS|FAIL|SKIP))' > after-cases.txt
go test ./<pkg>/... -count=1 -coverprofile=after.cov && go tool cover -func=after.cov | tail -1

• Coverage after ≥ before (per package).
• Cases: map every test from before-cases.txt to a test or table row in the new code. Old test functions that became rows have a different name; the mapping goes into the PR description as a table (old name → new location). A case without a mapping is a lost case: restore it.
• No assertion weakened (e.g. exact comparison replaced by "not empty"), no t.Skip added.
• Lint and format: golangci-lint run ./<pkg>/...; frontend npm run lint, npm run format.

5. Review

Run the reviewer agent on the diff with the before/after case lists and coverage numbers. A lost case or lower coverage is blocking. Don't push or open a PR without the person's OK.