Filter problems and summarize recent practice

[Danielncku]

What this does

The lobby already lets candidates choose a difficulty, but it offers no way to narrow the problem bank by topic. This change adds a local topic selector that combines with the existing difficulty checkboxes and limits both the visible cards and the recommendation pool. The selection is remembered in local storage.

Problem topics are generated from the existing problem bank and stored only as card metadata. They are not rendered on the cards, so the scenario does not reveal the underlying problem category.

The lobby also shows a compact snapshot of the latest eight assessed interviews. It reports passes, misses, pass rate, the current pass or miss streak, and topics with more misses than passes. Only dated reports with an explicit HIRE or NO_HIRE decision are counted. The snapshot stays hidden when there is no assessed history or report loading fails.

Not included

This version does not track interview starts, incomplete sessions, account-specific local state, or a Failed Before filter. Those earlier changes were removed to keep the pull request focused on topic selection and report-derived insights.

Validation

Focused unit tests cover topic extraction, combined difficulty and topic filtering, recent-result ordering, pass-rate calculation, unscored reports, and weak-topic detection.

Playwright tests cover filtering and reset behavior, rendering the recent snapshot from assessed reports, and hiding it when report history cannot be loaded.

ESLint, Prettier, and the generated problem-card consistency check pass locally. The lobby and /interview routes both return 200 from the official Windows Rust server with this checkout's web/ override.

The full browser suite is not green on this Windows checkout: 16 failures remain in avatar, vendor, and source-policy checks outside the changed paths, so this description does not claim a fully green local gate.

[jserv]

Check https://cbea.ms/git-commit/ carefully and enforce the rules.

[ColtenOuO]

Could you also include a commit body? It will make it much easier to understand the changes when tracking them down in the future.

[Danielncku]

Addressed the requested changes in 4d9caf7:

• Squashed the branch into one commit following the referenced commit-message rules, including an explanatory body wrapped at 72 columns.
• Scoped browser-local practice attempts to the current normalized GitHub login, with a separate anonymous scope.
• Added unit and Playwright coverage for account isolation.
• Changed small topic metadata from --muted to --sub for sufficient contrast.
• Preserved the existing behavior that prioritizes due reviews outside the currently suggested difficulty.

Verification:

• 77/77 focused Node tests passed.
• 57/57 real-browser lobby tests passed.
• ESLint, JavaScript syntax checks, generated problem-card check, and git diff --check passed.

The complete Windows run executed 627 tests: 608 passed, 5 skipped, and 14 unrelated pre-existing platform/vendor failures remained (Windows ESM paths and CRLF-sensitive checks, plus missing MediaPipe vendor assets).

[jserv]

erasable just below is entries.length > 0 || readDeviceHistory().length > 0, so a candidate whose only local practice data is recorded starts gets #history-header hidden and no Delete control, while this call still prints 1 recent interview start saved locally. That makes the new store the one thing the lobby writes to the device that the candidate cannot erase. Count practiceAttempts.length in erasable.

[Danielncku]

practiceAttempts.length is now part of the erasable condition. A start-only browser test confirms that Delete is visible, removes the active and anonymous start data, and preserves another signed-in account scope.

[Danielncku]

Hi @jserv, thank you very much for taking the time to provide such a detailed review. I have replied in each thread with my current understanding and proposed direction. As I am still learning the project’s design conventions, I would like to make sure my interpretation is correct before changing the implementation.

Two details—the deletion boundary across local scopes and the count-based incomplete model—seem especially worth confirming first. I will therefore hold off on pushing another revision for now. Once the intended behavior is clear, I will rebase onto the latest main, resolve the conflicts carefully, implement the agreed changes together, run the full relevant test suite, and then update this pull request. Please feel free to correct any part of my understanding.

[jserv]

Once those interpretations are confirmed, I will rebase onto the latest main, resolve the conflicts, implement the agreed changes together, run the full relevant test suite, and then update this pull request.

Always rebase onto the latest main and synchronize with the latest changes before making further updates, to minimize potential code duplication and merge conflicts.

Don't have to say greeting. Focus on efficient discussions.

[Danielncku]

Rebased onto upstream/main at 42c3b08 before applying the review changes, resolved the app.js conflicts, and updated this PR with commit 62b5bd7. The ten inline issues have been addressed and replied to individually. Validation completed: 86 focused Node tests passed; 8 targeted Playwright regression scenarios passed; ESLint, JavaScript syntax checks, generated-card consistency, and git diff checks passed. I will use the same sync-first workflow for any further update.

[jserv]

Rebased onto upstream/main at 42c3b08 before applying the review changes, resolved the app.js conflicts, and updated this PR with commit 62b5bd7.

There is NO need to make statements like this. GitHub already tracks every commit. Speak like a human instead of producing AI slop.

[ColtenOuO]

Please rebase onto the latest main and resolve any conflicts.

[jserv]

Rebase the current branch onto the upstream default branch and rework the series into functionally minimal commits, folding similar ones and enforcing the project's commit message rules.

[Danielncku]

I narrowed this pull request to topic filtering and a recent practice snapshot. The topic selection now combines with the existing difficulty filter and stays local to the browser; the snapshot counts only dated HIRE or NO_HIRE reports and stays hidden when history cannot be loaded.

The earlier incomplete-start and account-scoping changes have been removed. Focused unit and browser tests cover filtering, snapshot calculation, and report-load failure handling.

[ColtenOuO]

Check the CI error logs and fix the issues.

[jserv]

Passing the topic-narrowed pool into pickProblem changes two things the description does not mention. Due reviews outside the selected topic are never offered, even though the difficulty filter deliberately lets a review through ("A review can fall outside the levels now selected"), and the topic persists across visits. And once every problem in the topic is passed, choice.repeat produces "You have passed every problem at this level." while unpassed problems at that level remain in other topics. Keep due drawn from all cards, as the difficulty path does, and qualify the repeat message with the topic when one is selected.

[Danielncku]

sorry i didnt find the due review. i already rebased the newest.
1.Due reviews are now selected from the complete problem set, independently of the topic-filtered visible pool, so an overdue review keeps its existing highest priority!
2.When every problem matching the selected topic and level has been passed, the message now says: “You have passed every problem matching this topic and level.” Both behaviors are covered by browser tests.

[Danielncku]

I will fix this CI later