fix(conformance): CI-robust honest scorer (40.4%) + corrected site

.github/workflows/ci.yml

@@ -96,7 +96,12 @@ jobs:
     runs-on: ubuntu-24.04
     timeout-minutes: 10
     steps:
+      # fetch-depth: 0 — mirrors deploy-pages.yml so the conformance over-time
+      # chart (built from the git history of conformance_status.csv) is exercised
+      # by the pre-merge build check, not just the post-merge deploy.
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          fetch-depth: 0
 
       - uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
         with:
@@ -110,6 +115,11 @@ jobs:
 
       - name: Build site
         working-directory: website
+        # GITHUB_TOKEN raises the GitHub API rate limit for _data/releases.js
+        # (the releases page is generated from the live Releases API at build
+        # time); the build still degrades gracefully if the call fails.
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
         run: npm run build
 
   # ── Lint (runs in parallel with all test jobs) ─────────────────────────────
@@ -225,16 +235,16 @@ jobs:
       - name: Install cargo-llvm-cov
         uses: taiki-e/install-action@81ee9698f20724138a785d788c7567d40f14cd2d # cargo-llvm-cov
 
-      # The PEP conformance suite is fetched fresh from the upstream typing repo
-      # on every run (network-bound, one HTTP request per test file). Cache it
-      # keyed on conformance.sh, which holds the pinned TYPING_REF — bumping the
-      # ref edits that file and busts the cache; conformance.sh itself re-fetches
-      # whenever the stamped ref differs, so a stale restore self-heals.
-      - name: Cache conformance suite
+      # Only the conformance FIXTURES are fetched (the calculator
+      # conformance/upstream_main.py is committed). Cache the downloaded fixtures
+      # keyed on score.py, which holds the pinned ref (PINNED_TYPING_REF) — bumping
+      # the ref edits that file and busts the cache; score.py re-fetches whenever
+      # the stamped ref differs, so a stale prefix restore self-heals.
+      - name: Cache conformance fixtures
         uses: actions/cache@27d5ce7f107fe9357f9df03efb73ab90386fccae # v5.0.5
         with:
-          path: crates/basilisk-cli/tests/conformance
-          key: conformance-suite-${{ hashFiles('scripts/conformance.sh') }}
+          path: conformance/tests
+          key: conformance-suite-${{ hashFiles('conformance/score.py') }}
           restore-keys: conformance-suite-
 
       - name: Run Rust tests with coverage

.github/workflows/deploy-pages.yml

@@ -37,7 +37,12 @@ jobs:
       name: github-pages
       url: ${{ steps.deploy.outputs.page_url }}
     steps:
+      # fetch-depth: 0 — the conformance over-time chart (_data/conformance.js)
+      # reads the full git history of conformance/conformance_status.csv. A
+      # shallow clone would collapse the chart to a single point.
       - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0 # v7.0.0
+        with:
+          fetch-depth: 0
 
       - uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e # v6.4.0
         with:
@@ -51,6 +56,11 @@ jobs:
 
       - name: Build site
         working-directory: website
+        # GITHUB_TOKEN raises the GitHub API rate limit for _data/releases.js
+        # (the releases page is generated from the live Releases API at build
+        # time); the build still degrades gracefully if the call fails.
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
         run: npm run build
 
       - uses: actions/configure-pages@45bfe0192ca1faeb007ade9deae92b16b8254a0d # v6.0.0

.gitignore

@@ -106,9 +106,11 @@ deslop-*.log
 benchmarks/results/
 
 # =============================================================================
-# Conformance test suite (fetched via make conformance / scripts/conformance.sh)
+# Conformance test FIXTURES: downloaded on demand into conformance/tests/ (via
+# make conformance / score.py --fetch) — never committed. NOTE: the official
+# calculator conformance/upstream_main.py IS committed and is NOT ignored.
 # =============================================================================
-crates/basilisk-cli/tests/conformance/
+conformance/tests/
 
 # =============================================================================
 # Tool / agent state
@@ -149,3 +151,8 @@ URGENT_READ_ME_NOW.md
 
 .deslop-cache/
 .ghissues/
+
+website/_verify/
+
+
+/*.png

Makefile

@@ -167,7 +167,8 @@ mutation-test:
 ## conformance/conformance_status.csv. Fetches the upstream suite if missing;
 ## use FETCH=1 to force a re-download.
 conformance:
-	@bash scripts/conformance.sh $(if $(filter 1,$(FETCH)),--fetch,)
+	@cargo build -p basilisk-cli --bin basilisk
+	@python3 conformance/score.py --bin target/debug/basilisk $(if $(filter 1,$(FETCH)),--fetch,)
 
 ## bench: Benchmark Basilisk vs pyright/mypy/ty/pyrefly on the fixture suite.
 ## Requires hyperfine; competitor tools are skipped if not installed.