up

2025-12-01 21:16:22 +02:00
parent c11d87d252
commit 909d9b6220
208 changed files with 860954 additions and 832 deletions
--- a/bench/reachability-benchmark/website/README.md
+++ b/bench/reachability-benchmark/website/README.md
@@ -0,0 +1,20 @@
+# Reachability Benchmark Website
+
+Static, offline-ready page for the public benchmark (task BENCH-WEBSITE-513-015).
+
+## Files
+- `index.html` — single-page site (no external assets) with:
+  - Quick start steps
+  - Download pointers (cases, schemas, truth, baselines)
+  - Determinism checklist
+  - Leaderboard panel that reads `leaderboard.json` if present
+
+## Usage
+1) Generate leaderboard locally:
+```bash
+ci/run-ci.sh    # or run rb_compare manually
+cp out/leaderboard.json website/
+```
+2) Serve the `website/` folder with any static file server (or open `index.html` directly).
+
+No external fonts or network calls are used; works fully offline.
--- a/bench/reachability-benchmark/website/index.html
+++ b/bench/reachability-benchmark/website/index.html
@@ -0,0 +1,147 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+  <meta charset="UTF-8">
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  <title>Stella Ops · Reachability Benchmark</title>
+  <style>
+    :root {
+      --bg: #0f172a;
+      --panel: #111827;
+      --accent: #22d3ee;
+      --muted: #9ca3af;
+      --text: #e5e7eb;
+      --mono: "SFMono-Regular", Consolas, "Liberation Mono", Menlo, monospace;
+      --sans: "Inter", "Segoe UI", system-ui, -apple-system, sans-serif;
+    }
+    * { box-sizing: border-box; }
+    body {
+      margin: 0;
+      background: var(--bg);
+      color: var(--text);
+      font-family: var(--sans);
+      line-height: 1.5;
+      padding: 24px;
+    }
+    header { margin-bottom: 24px; }
+    h1 { margin: 0 0 8px; font-size: 28px; }
+    h2 { margin-top: 32px; margin-bottom: 12px; font-size: 20px; }
+    p { margin: 6px 0; color: var(--muted); }
+    code, pre { font-family: var(--mono); }
+    .panel {
+      background: var(--panel);
+      border: 1px solid #1f2937;
+      border-radius: 10px;
+      padding: 16px;
+      margin-bottom: 16px;
+    }
+    .grid {
+      display: grid;
+      gap: 12px;
+    }
+    @media (min-width: 720px) {
+      .grid { grid-template-columns: repeat(2, minmax(0, 1fr)); }
+    }
+    .leaderboard table {
+      width: 100%;
+      border-collapse: collapse;
+    }
+    .leaderboard th, .leaderboard td {
+      padding: 8px;
+      border-bottom: 1px solid #1f2937;
+      text-align: left;
+      font-size: 14px;
+    }
+    .leaderboard th { color: var(--muted); font-weight: 600; }
+    .pill {
+      display: inline-block;
+      padding: 2px 8px;
+      border-radius: 999px;
+      background: rgba(34, 211, 238, 0.15);
+      color: var(--accent);
+      font-size: 12px;
+      font-weight: 600;
+    }
+    .badge-warning { background: rgba(234,179,8,0.18); color: #facc15; }
+    .list { padding-left: 18px; color: var(--muted); }
+  </style>
+</head>
+<body>
+  <header>
+    <div class="pill">Offline ready</div>
+    <h1>Stella Ops · Reachability Benchmark</h1>
+    <p>Deterministic, reproducible cases and scoring harness for reachability analysis tools.</p>
+  </header>
+
+  <section class="panel">
+    <h2>Quick Start</h2>
+    <ol class="list">
+      <li>Build cases deterministically: <code>python tools/build/build_all.py --cases cases</code></li>
+      <li>Run your analyzer and emit <code>submission.json</code> in <code>schemas/submission.schema.json</code> format.</li>
+      <li>Score: <code>tools/scorer/rb_score.py --truth benchmark/truth/&lt;aggregate&gt;.json --submission submission.json</code></li>
+      <li>Compare: <code>tools/scorer/rb_compare.py --truth ... --submissions submission.json baselines/*/submission.json --output leaderboard.json</code></li>
+    </ol>
+    <p>All tooling is offline-friendly; no network calls or external fonts.</p>
+  </section>
+
+  <section class="grid">
+    <div class="panel">
+      <h2>Downloads</h2>
+      <ul class="list">
+        <li>Cases: <code>cases/</code></li>
+        <li>Schemas: <code>schemas/</code></li>
+        <li>Truth: <code>benchmark/truth/</code></li>
+        <li>Baselines: <code>baselines/</code> (Semgrep, Stella, CodeQL)</li>
+        <li>CI script: <code>ci/run-ci.sh</code></li>
+      </ul>
+    </div>
+    <div class="panel">
+      <h2>Determinism Checklist</h2>
+      <ul class="list">
+        <li>Set <code>SOURCE_DATE_EPOCH</code> in builds.</li>
+        <li>Disable tool telemetry/version checks.</li>
+        <li>Sort cases and sinks before emission.</li>
+        <li>Keep outputs local; no registry or network pulls.</li>
+      </ul>
+    </div>
+  </section>
+
+  <section class="panel leaderboard">
+    <h2>Leaderboard</h2>
+    <p id="lb-note" class="muted">Looking for <code>leaderboard.json</code> in this directory…</p>
+    <div id="lb-table"></div>
+  </section>
+
+  <script>
+    const note = document.getElementById('lb-note');
+    const tableHost = document.getElementById('lb-table');
+
+    fetch('leaderboard.json')
+      .then(r => r.ok ? r.json() : Promise.reject(r.status))
+      .then(data => {
+        note.textContent = `Truth version: ${data.truth_version || 'n/a'} · Entries: ${data.entries.length}`;
+        const rows = data.entries.map((e, i) => `
+          <tr>
+            <td>${i + 1}</td>
+            <td>${e.name}</td>
+            <td>${e.tool_name} ${e.tool_version}</td>
+            <td>${e.f1.toFixed(4)}</td>
+            <td>${e.precision.toFixed(4)}</td>
+            <td>${e.recall.toFixed(4)}</td>
+            <td>${e.determinism_rate.toFixed(4)}</td>
+            <td>${e.explainability_avg.toFixed(4)}</td>
+          </tr>`).join('');
+        tableHost.innerHTML = `
+          <table>
+            <thead>
+              <tr><th>#</th><th>Name</th><th>Tool</th><th>F1</th><th>P</th><th>R</th><th>Det</th><th>Explain</th></tr>
+            </thead>
+            <tbody>${rows}</tbody>
+          </table>`;
+      })
+      .catch(() => {
+        note.innerHTML = 'No <code>leaderboard.json</code> found yet. Run <code>ci/run-ci.sh</code> to generate.';
+      });
+  </script>
+</body>
+</html>