test(kattis): snapshot the real problem-page HTML parser

[cameroncuster]

Problem — parseKattisProblemPage scrapes open.kattis.com markup for a problem's name and difficulty, but it's the one Kattis function with no direct coverage: every existing test (unit and e2e) mocks parsePage out. A selector change or upstream markup drift would slip through silently.

Solution — add a Playwright spec that exercises the production function (serialized via .toString(), not a re-implementation) inside a real browser DOM through page.evaluate, asserting name + rating against captured fixtures:

• full page (<h1> + .difficulty_number)
• legacy .difficulty class fallback
• missing difficulty → default rating 5
• missing <h1> → falls back to the problem id

It runs in the browser because the parser uses the browser-only DOMParser; this deliberately avoids adding a jsdom/happy-dom dev dependency (per repo guidelines on new deps). The spec runs in the existing mocked desktop/mobile projects and uses about:blank, so it's decoupled from app/network state.

Testing — I could not run the suite in my environment (only Node 22 available; repo requires Node ≥24 with engine-strict). Opened as draft so CI on Node 24 validates lint/lint:es/check and the Playwright run. I'll address any CI feedback.

_{Opened by an Augment PR Author agent on behalf of @cameroncuster.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
gitgud	Ready	Preview, Comment	Jul 24, 2026 11:31pm

[cameroncuster]

^{PR Author Agent⚡ on behalf of @cameroncuster}

👋 This PR is monitored by an Augment Agent.

I'll answer questions, implement suggestions, fix CI failures, and resolve merge conflicts. I slow down if the PR goes quiet, but any new comment wakes me back up.

[cameroncuster]

^{PR Risk Analyzer Agent🛡️}

👀 Human Input Needed
A pair-review briefing is ready for you.

→ Pair Review Briefing

[cameroncuster]

^{PR Author Agent⚡ on behalf of @cameroncuster}

Addressing the Risk Analyzer's two focus topics:

Fixtures pin real behavior and catch drift. The test runs the production parseKattisProblemPage (not a copy), so it asserts the actual selectors: h1 for the name and .difficulty_number (with .difficulty fallback) for the rating. The four fixtures cover each branch — primary selector, legacy-class fallback, missing-difficulty default (rating 5), and missing-h1 (falls back to the problem id). If Kattis renames/moves those elements or someone changes a selector, the name/rating assertions fail. It won't catch drift in markup the parser doesn't look at (by design — the parser only reads those two selectors).

Why .toString() + new Function. The parser depends on the browser-only DOMParser, so it can't run in the Node unit process. Serializing the real function and reconstructing it inside page.evaluate runs the genuine production code against a real DOM, which is the whole point — the alternative (re-implementing selectors in the test) would test a copy, not the code that ships. It's safe here: the function is self-contained (only DOMParser/document, no closure or module refs), and the fixture HTML is static test data, not untrusted input. The trade-off vs. adding jsdom/happy-dom was deliberate — this avoids a new dependency, per the repo's guidance.

Happy to switch to a DOM shim + Node unit test instead if a maintainer prefers that shape.

[cameroncuster]

^{Deep Code Review Agent🐛}

Review completed with 0 suggestions.

[cameroncuster]

^{Verifier 🛰️}

🛰️ Verifier Report

TL;DR: Ran the new Kattis-parser spec in a real browser across desktop + mobile, then broke the production selector to prove the spec actually catches a regression rather than passing blindly.

🔍 Walkthrough

• Ran the new spec end-to-end in a real browser → all 8 checks pass (4 cases × mocked-desktop + mocked-mobile), driving the browser-only DOMParser path CI's Node unit tests can't reach.

  the four cases map 1:1 onto the parser's branches (name via h1, rating via .difficulty_number → .difficulty → default 5)

  ✓ reads the h1 name and .difficulty_number rating        → { name: 'Two Stones', rating: 2.1 }
  ✓ falls back to the .difficulty class                    → { name: 'Legacy Problem', rating: 7 }
  ✓ defaults the rating to 5 when no difficulty is present → { name: 'Nameless Difficulty', rating: 5 }
  ✓ falls back to the problem id when the h1 is missing     → { name: 'missingtitle', rating: 5 }
  

  Each expected value matches the production function in src/lib/providers/kattis/ingestion.ts exactly.

• Confirmed the spec binds to the production parser (via .toString() + page.evaluate), not a re-implementation → breaking the real selector fails exactly the two selector-dependent cases while the other two still pass.

  mutation: .difficulty_number, .difficulty → a wrong class in ingestion.ts, then re-ran

  ✘ reads the h1 name and .difficulty_number rating   Expected rating 2.1, Received 5
  ✘ falls back to the .difficulty class               Expected rating 7,   Received 5
  ✓ defaults the rating to 5 when no difficulty present
  ✓ falls back to the problem id when the h1 is missing
  

  On mainline the two difficulty cases assert 2.1 / 7; with the selector broken the parser returns its default 5 and they fail — so a markup/selector drift fails loudly, as the PR intends. Selector restored afterward; working tree left clean.

📋 Scope

Under test: a new Playwright spec (e2e/kattis-parser.spec.ts, +109 lines, the only changed file) that pins parseKattisProblemPage's name/difficulty scraping against captured HTML fixtures by running the serialized production function inside a real browser DOM.

Not tested:

• CI's static gates (lint, lint:es, check, build) — CI owns these; not re-run here.
• The rest of the e2e suite and the live-smoke path (SUPABASE_SMOKE=1) — unrelated to this change; the spec uses about:blank and is decoupled from app/network/DB state, so no Supabase surface was touched.

📦 Artifacts · clean run · mutation proof

---

Don't like what you see? Add a verification skill so the next run tests this kind of change.

Want another run? Comment cosmos verify on this PR.

👍 / 👎 · View session