Publish verified EvalArc 746ba6237812 (part 3)

robot/index.html

@@ -56,11 +56,12 @@ p,.label,.verdict p,.verdict .next,.tone.info{color:#222}th,a,.number,.eyebrow,.
 pre{background:#f4f4f4;max-height:none}.scroll{overflow:visible}.skip{display:none}}
 
 </style>
+<a class="skip" href="#main">Skip to content</a><main id="main">
 <div class="eyebrow">EVALARC / DEVELOPMENT AUDIT</div>
 <h1>Test the grader.<br>Then trust the signal.</h1>
-<p>Behavioral controls for task-specific agent evaluations. Correct and faulty
+<section class="verdict pass" role="status" aria-label="Result"><h2>&#10003; 6 of 6 declared faults detected</h2><p>1 fault(s) depend on a single detecting case.</p><p class="next"><strong>Next:</strong> Add cases that detect the single-case faults independently.</p></section><p>Behavioral controls for task-specific agent evaluations. Correct and faulty
 submissions are evaluated against the same externally enforced contract.</p>
-<div class="cards"><div class="card"><div class="number">6/6</div><div class="label">declared faults detected</div></div><div class="card"><div class="number">1.000</div><div class="label">reference correctness</div></div><div class="card"><div class="number">docker</div><div class="label">execution backend</div></div></div><section class="audit-coverage" aria-labelledby="coverage-title"><h2 id="coverage-title">Which faults depend on one case?</h2><p><strong>1 of 6</strong> declared faults have exactly one distinct detecting case in the assessed records. Repeating a case under more seeds does not increase this margin. A surviving fault has margin zero; an environment failure remains unassessed.</p><p>Margins below are derived from the saved failing case IDs. Rendering an older audit does not rerun its controls.</p></section><h2>Does the grader detect plausible defects?</h2><div class="scroll" tabindex="0" role="region" aria-label="Declared fault coverage"><table><thead><tr><th>Negative control</th><th>Target</th><th>Candidate score</th><th>Result</th><th title="Cases that caught this fault independently. One means the suite loses this fault if that case changes.">Margin</th><th>Evidence</th></tr></thead><tbody><tr><td><code>assume-meters</code></td><td>coordinates</td><td>0.750</td><td>Detected</td><td>2</td><td><a href="#fault-0">Inspect recorded cases</a></td></tr><tr><td><code>ignore-frame-origin</code></td><td>coordinates</td><td>0.750</td><td>Detected</td><td>2</td><td><a href="#fault-1">Inspect recorded cases</a></td></tr><tr><td><code>ignore-clock-units</code></td><td>clock</td><td>0.800</td><td>Detected</td><td>2</td><td><a href="#fault-2">Inspect recorded cases</a></td></tr><tr><td><code>assume-complete</code></td><td>completeness</td><td>0.963</td><td>Detected</td><td>1</td><td><a href="#fault-3">Inspect recorded cases</a></td></tr><tr><td><code>last-frame-is-peak</code></td><td>metrics</td><td>0.750</td><td>Detected</td><td>4</td><td><a href="#fault-4">Inspect recorded cases</a></td></tr><tr><td><code>invent-source</code></td><td>provenance</td><td>0.900</td><td>Detected</td><td>4</td><td><a href="#fault-5">Inspect recorded cases</a></td></tr></tbody></table></div><h2>Inspect the detecting cases</h2><details class="audit-control"><summary>assume-meters &#183; 2 detecting cases</summary><div id="fault-0"><p>Target: coordinates. Detecting case IDs: centimeters-offset-clock, millimeters-sensor-frame.</p><details><summary>millimeters-sensor-frame &#183; seed 17 &#183; failed</summary><pre>{
+<div class="cards"><div class="card"><div class="number">6/6</div><div class="label">declared faults detected</div></div><div class="card"><div class="number">1.000</div><div class="label">reference correctness</div></div><div class="card"><div class="number">docker</div><div class="label">execution backend</div></div></div><section class="audit-coverage" aria-labelledby="coverage-title"><h2 id="coverage-title">Which faults depend on one case?</h2><p><strong>1 of 6</strong> declared faults have exactly one distinct detecting case in the assessed records. Repeating a case under more seeds does not increase this margin. A surviving fault has margin zero; an environment failure remains unassessed.</p><p>Margins below are derived from the saved failing case IDs. Rendering an older audit does not rerun its controls.</p></section><h2>Does the grader detect plausible defects?</h2><div class="scroll" tabindex="0" role="region" aria-label="Declared fault coverage"><table><thead><tr><th scope="col">Negative control</th><th scope="col">Target</th><th scope="col">Candidate score</th><th scope="col">Result</th><th scope="col" title="Cases that caught this fault independently. One means the suite loses this fault if that case changes.">Margin</th><th scope="col">Evidence</th></tr></thead><tbody><tr><td><code>assume-meters</code></td><td>coordinates</td><td>0.750</td><td>Detected</td><td>2</td><td><a href="#fault-0">Inspect recorded cases</a></td></tr><tr><td><code>ignore-frame-origin</code></td><td>coordinates</td><td>0.750</td><td>Detected</td><td>2</td><td><a href="#fault-1">Inspect recorded cases</a></td></tr><tr><td><code>ignore-clock-units</code></td><td>clock</td><td>0.800</td><td>Detected</td><td>2</td><td><a href="#fault-2">Inspect recorded cases</a></td></tr><tr><td><code>assume-complete</code></td><td>completeness</td><td>0.963</td><td>Detected</td><td>1</td><td><a href="#fault-3">Inspect recorded cases</a></td></tr><tr><td><code>last-frame-is-peak</code></td><td>metrics</td><td>0.750</td><td>Detected</td><td>4</td><td><a href="#fault-4">Inspect recorded cases</a></td></tr><tr><td><code>invent-source</code></td><td>provenance</td><td>0.900</td><td>Detected</td><td>4</td><td><a href="#fault-5">Inspect recorded cases</a></td></tr></tbody></table></div><h2>Inspect the detecting cases</h2><details class="audit-control"><summary>assume-meters &#183; 2 detecting cases</summary><div id="fault-0"><p>Target: coordinates. Detecting case IDs: centimeters-offset-clock, millimeters-sensor-frame.</p><details><summary>millimeters-sensor-frame &#183; seed 17 &#183; failed</summary><pre>{
   &quot;seed&quot;: 17,
   &quot;case_id&quot;: &quot;millimeters-sensor-frame&quot;,
   &quot;checks&quot;: {
@@ -57834,4 +57835,4 @@ submissions are evaluated against the same externally enforced contract.</p>
       &quot;stderr_tail_limit_bytes&quot;: 2048
     }
   ]
-}</pre></details><h2>Reproduction record</h2><p class="metadata">Task: robot-evidence-review v0.1.0<br>Seeds: [17, 41, 97]<br>Grader SHA-256: aea49d42e1acd8881163c3dcae645dd824c7185e2bac3b79c93f0246a349a425<br>Cases SHA-256: 71d979fd8ba2c414a8a3f68bc5a49ad6e2a60a82ddeb5cedbd5dcf53251a804f<br>Image ID: sha256:78387bc3881b8273120a12ebe6c1ab22b018ccc2c9adf565ae1ac9b536e184ea<br>Created: 2026-09-14T10:52:41.593549+00:00</p><p><a href="audit.json">Complete JSON evidence for all controls</a></p><footer><strong>Scope:</strong> This is a scripted grader audit, not an AI leaderboard. Public seeds are not held-out evaluation data. Detection rates apply only to the listed fault models. No human time horizon, model quality, or RL improvement is inferred.</footer></html>
+}</pre></details><h2>Reproduction record</h2><p class="metadata">Task: robot-evidence-review v0.1.0<br>Seeds: [17, 41, 97]<br>Grader SHA-256: aea49d42e1acd8881163c3dcae645dd824c7185e2bac3b79c93f0246a349a425<br>Cases SHA-256: 71d979fd8ba2c414a8a3f68bc5a49ad6e2a60a82ddeb5cedbd5dcf53251a804f<br>Image ID: sha256:78387bc3881b8273120a12ebe6c1ab22b018ccc2c9adf565ae1ac9b536e184ea<br>Created: 2026-09-14T10:52:41.593549+00:00</p><p><a href="audit.json">Complete JSON evidence for all controls</a></p><footer><strong>Scope:</strong> This is a scripted grader audit, not an AI leaderboard. Public seeds are not held-out evaluation data. Detection rates apply only to the listed fault models. No human time horizon, model quality, or RL improvement is inferred.</footer></main></html>