ci(types): type-check test files via tsconfig.tests.json, non-blocking first (#3962)

[Yeraze]

Summary

Task 1.2 of the remediation plan (#3962, Phase 1).

~557 test files currently get zero type-checking (tsconfig.json excludes **/*.test.ts*), so mock shapes silently drift from real signatures. This adds:

• tsconfig.tests.json — extends the root config, adds src/**/*.test.ts(x) to the program, relaxes only noUnusedLocals/noUnusedParameters.
• npm run typecheck:tests script.
• Non-blocking CI steps in ci.yml (gated to the 24.x leg to avoid 4× matrix cost) and pr-tests.yml, each with an inline comment stating the flip-to-blocking condition (error count = 0).
• Epic status doc updated through Task 1.2.

Baseline: 283 errors, all in test files, zero in production sources — real drift (e.g. mock checkPermission signatures that no longer match, properties read off mocks that the real type doesn't have). Top ~8 files hold ~150 of them; burndown + flip-to-blocking is follow-up work.

Deviations from the plan (measured, documented in the epic doc)

1. Full strict kept — noImplicitAny stays ON. The plan suggested turning it off initially; measured, that adds ~50 false-positive errors in production files (disabling evolving-any inference) while hiding real test drift. Strict-on is better on every axis: 283/0 test/prod vs 345/50.
2. Top-level tests/ dir excluded — including it pulls @types/node into the whole program and poisons frontend inference (~60 spurious implicit-any errors in src/components). Documented in the config header; it can get its own config later if wanted.

Verification

• Baseline reproducible: npm run typecheck:tests → 283 errors, 0 in prod files.
• Negative test: a deliberate string-into-number mock error was caught at its exact location, then reverted (283 → 284 → 283).
• npm run typecheck (prod) unaffected: clean. Builds and vitest untouched.
• Full suite: 8068 passed, 0 failed, 0 suite failures (success: true).

Refs #3962

🤖 Generated with Claude Code

https://claude.ai/code/session_01AphLfgUJWaFNAgU5UXdJ4n

[github-actions]

Claude finished @Yeraze's task —— View job

---

PR Review: ci(types): type-check test files via tsconfig.tests.json

Tasks:

• Gather context and understand the request
• Read changed files
• Analyze for code quality, bugs, security, performance
• Post review feedback

---

Overall Assessment

This is a well-structured, pragmatic addition. The approach (non-blocking first, document the flip condition explicitly, gate to one matrix leg) is the right way to introduce a new quality gate against an existing error backlog. The deviations from the original plan are well-reasoned and documented.

---

tsconfig.tests.json

Minor issue — redundant noEmit:
noEmit: true appears both in the config file (line 13) and in the npm script (tsc -p tsconfig.tests.json --noEmit). This is harmless but slightly noisy; one or the other is sufficient. Convention in this repo is to put it in the config (npm run typecheck does tsc --noEmit while the root config also sets "noEmit": true), so the flag on the CLI is redundant. Worth cleaning up when convenient.

include has a redundant glob:

"include": ["src", "src/**/*.test.ts", "src/**/*.test.tsx"]

"src" already covers src/**/*.test.ts(x) — the additional globs are redundant (though harmless). The meaningful work here is the exclude not filtering them out (unlike tsconfig.json). This is fine, just a nit.

*.spec.ts(x) pattern not included:
tsconfig.json excludes *.spec.ts and *.spec.tsx alongside *.test.ts*. tsconfig.tests.json includes src/**/*.test.ts and src/**/*.test.tsx but doesn't add the spec variants. If any spec files exist under src/ they'd be silently skipped. Worth confirming the codebase uses only *.test.* naming — if so, a comment noting that would make the intent clearer.

---

.github/workflows/ci.yml (lines 79–85)

The gate to 24.x only is clearly justified in the PR description (avoid 4× matrix cost). The inline comment explaining the flip condition is exactly right. No issues.

---

.github/workflows/pr-tests.yml (lines 93–97)

Consistent with ci.yml. One very minor observation: pr-tests.yml runs typecheck:tests unconditionally (no if: matrix.node-version filter needed since this job already pins to 24.x). This is correct as-is.

generate-test-summary step has a stale $? check (pre-existing, not introduced here):
Line 109 if [ $? -eq 0 ] always evaluates the exit code of echo, not the test runner, because run: blocks don't chain that way. This is pre-existing and out of scope for this PR, but worth flagging for follow-up.

---

package.json

Script addition is clean and follows existing convention (typecheck → typecheck:tests). No issues.

---

Summary

Area	Status
tsconfig.tests.json correctness	✅ Sound
noEmit duplication	⚠️ Minor nit (harmless)
spec file coverage gap	⚠️ Worth confirming
CI non-blocking gate	✅ Well-structured
Flip-to-blocking documentation	✅ Clear and actionable
Production typecheck unaffected	✅ Verified per PR description

The PR achieves its stated goal cleanly. The nits above are all non-blocking — the core implementation is solid and a good step toward the remediation epic goal.

---