Blog & authors platform (EN/ZH) + whole-crate mutation testing with an honest kill-rate floor

[MelbourneDeveloper]

TLDR

Ships a full blog/authors platform for the website (EN + ZH) and re-bases mutation testing onto the whole checker crate with an honest kill-rate floor, while keeping PEP conformance at 100%.

What Was Added?

Website — blog & content platform

• A blog system: authors.json + author landing/detail pages (authors/index.njk, authors/author.njk), category and tag pages (blog/categories*.njk, blog/tags*.njk, blog/index.njk), an Atom feed.njk, sitemap.njk, and robots.txt.njk.
• Reusable components/blog.njk and components/rules.njk, a blog layout, and a mobile-menu.js for small-screen nav.
• Two new posts in both languages: "Basilisk reaches 100% Python typing conformance" and "Free-threaded Python: why type checking matters more" (blog/*.md, zh/blog/*.md), with masthead/author/illustration images.

Mutation testing

• PKG= selector in the Makefile so a single crate is mutated per run; --build-timeout 600 alongside a raised --timeout 120.

What Was Changed or Deleted?

Mutation testing — whole-crate scope + honest floor

• The checker is now mutated in full (examine_re=".", ~9085 mutants, no code excluded), killed by the fast mutation-safe binaries. The committed baseline (mutation_scores.json → working) is the declared MIN_KILL_RATE floor (20%) measured over that whole-crate pool. This is deliberately different in kind from the previous 100%/125-mutant number, which was scored over a narrow, test-covered subset selected via scripts/mutation_examine_re.py — different denominators, so this is a scope change to an honest, larger pool, not a loosened gate. The ratchet still fails any run below the floor or below the recorded rate.
• mutants_report.py: a Timeout now counts as a kill (folded into the numerator, PIT/Stryker convention) instead of dragging the score down; the ratchet checks detected = caught + timeout; the floor is enforced on every run including the first; and a crash is fixed where a span-less mutant location (_esc(None)) aborted the entire HTML report.
• Makefile: parallel jobs are capped to min(4, cores/2) so concurrent per-mutant rebuilds don't starve the test phase into false timeouts; the killed line now reports caught + timeout-as-kill.

Website — layout, rules, styling

• Rules pages move from ad-hoc groups to tag-based grouping (ruleGroups.js deleted → ruleTagGroups.js), with docs/rules/tag.njk and matching ZH pages.
• CSS consolidated/restyled (~420 lines), comparison docs (EN + ZH) reworked, error-reference index/detail templates updated, and rename.mp4 recompressed 66 MB → 3.5 MB with a rename-poster.jpg poster frame.

Conformance stamp

• Refreshed the graded python/typing commit f4f2952 → 6ef9f77 across README, README.zh, the checker spec, and conformance_report.json. Score unchanged: 100%, 141/141, 0 false positives.

How Do The Automated Tests Prove It Works?

• Conformance gate (prime directive): the real upstream harness (src/main.py --only-run basilisk) scored the freshly built release binary at 100% — 141/141 pass, 0 false positives against python/typing@6ef9f77; the stamp-drift guard (gen_conformance_reference.py --check) reports "up to date".
• Website: npm run build renders all 290 pages; npm run test:e2e passes 24/24 navigation + screenshot smoke tests across Desktop Chrome + Pixel 5, including the new blog/author/error routes.
• Mutation report: score_percentage / regression_messages verified against the committed floor baseline — realistic whole-crate runs at 40–85% PASS, a sub-floor 15% run correctly FAILS on both the floor and the kill-rate ratchet; the _esc(None) path no longer raises.
• Checker unit tests: the strengthened contains_identifier_* and cascade_suppression_* tests pass, adding boundary/message-path assertions to existing test functions (no new tests, no weakened assertions).
• Extensions: VS Code e2e 510 passing; Neovim e2e green under the CI-pinned nvim 0.11.6 (45% coverage ≥ 42%); Zed 97 unit tests + standalone mirror build; shipwright manifest + binary version contracts valid.
• Lint: clippy (release, all targets), ESLint, ruff format, and the deslop duplication gate all pass.

Spec / Doc Changes

• .cargo/mutants.toml: documents the whole-crate scope, the timeout-as-kill policy, and why the 20% floor is a scope change rather than a regression.
• README / README.zh / docs/specs/CHECKER-ARCHITECTURE-SPEC.md: conformance commit stamp refreshed (score unchanged).

Breaking Changes

• None