ratingtesting commited on
Commit
049efe6
·
1 Parent(s): ae24bc4

Add KDS leaderboard table to visual page (11 models, kimi-k3 STRONG)

Browse files
Files changed (1) hide show
  1. index.html +34 -0
index.html CHANGED
@@ -127,6 +127,17 @@
127
  .ritem .rb { float:right; font-size:18px; }
128
  .rlegend { font-size:16px; color:#64748b; margin-top:16px; line-height:1.7; }
129
  .rlegend b { color:#94a3b8; }
 
 
 
 
 
 
 
 
 
 
 
130
  </style>
131
  </head>
132
  <body>
@@ -288,6 +299,29 @@
288
  <div class="rlegend"><b>Honesty as a feature:</b> three ⚠ (style consistency, sycophancy-as-trait, confabulation) cannot be closed by a cheap machine detector — keelwright catches their measurable consequences and enforces verification discipline, not the disposition itself. Everything marked ✅ is machine-enforced or verified on disk, not "eyeball it."</div>
289
  </div>
290
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
291
  <div class="foot"><a href="https://github.com/ratingtesting/keelwright" style="color:#94a3b8;text-decoration:underline">keelwright</a> by ratingtesting · CC BY 4.0 · <a href="https://huggingface.co/spaces/ratingtesting/keelwright" style="color:#94a3b8;text-decoration:underline">Hugging Face</a> · source of truth: assets/architecture.md</div>
292
  </body>
293
  </html>
 
127
  .ritem .rb { float:right; font-size:18px; }
128
  .rlegend { font-size:16px; color:#64748b; margin-top:16px; line-height:1.7; }
129
  .rlegend b { color:#94a3b8; }
130
+ .kds { background:#0a0f1a; padding:28px; border-radius:12px; margin-top:24px; }
131
+ .kds h3 { font-size:26px; color:#fbbf24; margin:0 0 8px; text-align:center; }
132
+ .ksub { font-size:15px; color:#94a3b8; text-align:center; margin-bottom:18px; line-height:1.6; }
133
+ .ksub b { color:#e2e8f0; }
134
+ .ktable { width:100%; border-collapse:collapse; font-size:15px; }
135
+ .ktable th { background:#1e293b; color:#fbbf24; padding:8px 12px; text-align:left; border:1px solid #334155; }
136
+ .ktable td { padding:8px 12px; border:1px solid #1e293b; color:#e2e8f0; }
137
+ .ktable tr:nth-child(even) td { background:#0f1626; }
138
+ .khi { font-weight:800; color:#4ade80; font-size:17px; text-align:center; }
139
+ .kpart { font-weight:800; color:#fbbf24; text-align:center; }
140
+ .kzero { font-weight:800; color:#f87171; text-align:center; }
141
  </style>
142
  </head>
143
  <body>
 
299
  <div class="rlegend"><b>Honesty as a feature:</b> three ⚠ (style consistency, sycophancy-as-trait, confabulation) cannot be closed by a cheap machine detector — keelwright catches their measurable consequences and enforces verification discipline, not the disposition itself. Everything marked ✅ is machine-enforced or verified on disk, not "eyeball it."</div>
300
  </div>
301
 
302
+ <!-- KDS LEADERBOARD -->
303
+ <div class="kds">
304
+ <h3>📊 KEELWRIGHT SCORE (KDS) — PROVEN BY A/B TESTING</h3>
305
+ <div class="ksub">KDS = Execution Rate × Discrimination Rate / 100. 11+ models tested, all verified on disk with <code>validate_run.py</code>.</div>
306
+ <table class="ktable">
307
+ <thead><tr><th>Model</th><th>Tier</th><th>Benchmark</th><th>KDS</th></tr></thead>
308
+ <tbody>
309
+ <tr><td>poolside/laguna-s-2.1</td><td>STRONG</td><td>ML 78.5%</td><td class="khi">83</td></tr>
310
+ <tr><td>stepfun/step-3.7-flash</td><td>MEDIUM</td><td>Pro ~56%</td><td class="khi">67</td></tr>
311
+ <tr><td>nvidia/nemotron-3-ultra</td><td>STRONG</td><td>ML 67.7%</td><td class="khi">40</td></tr>
312
+ <tr><td>deepseek-v4-flash</td><td>STRONG</td><td>Verified ~79%</td><td class="khi">29</td></tr>
313
+ <tr><td>inclusionai/ling-3.0-flash</td><td>UNKNOWN</td><td>SWE-bench/GPQA N/A</td><td class="khi">22</td></tr>
314
+ <tr><td>kimi-k3</td><td>STRONG</td><td>Terminal-Bench 88.3, ProgramBench 77.8</td><td class="khi">25</td></tr>
315
+ <tr><td>mimo-v2.5</td><td>MEDIUM</td><td>Verified 78.9%, Pro 57.2%</td><td class="khi">18</td></tr>
316
+ <tr><td>nvidia/nemotron-3-super-120b-a12b</td><td>STRONG</td><td>Verified 60.47%</td><td class="kpart">PARTIAL</td></tr>
317
+ <tr><td>claude-opus-4-8</td><td>STRONG</td><td>frontier</td><td class="khi">17</td></tr>
318
+ <tr><td>tencent/hy3</td><td>STRONG</td><td>ML 75.8%, Verified 78%</td><td class="khi">7</td></tr>
319
+ <tr><td>cohere/north-mini-code</td><td>WEAK</td><td>Agentic 3.1</td><td class="kzero">0</td></tr>
320
+ </tbody>
321
+ </table>
322
+ <div class="ksub"><b>Key findings:</b> Medium models (Step 3.7, KDS 67) get MORE value than some strong models · kimi-k3 (STRONG) still benefits — autonomy dial stopped a silent business-hack commit · Ling re-run after fabricated first attempt: clean 4 DISCRIMINATES · Nemotron-3-super partial (2/18, both DISCRIMINATES) · weak models (KDS 0) fabricate, caught by gate.</div>
323
+ </div>
324
+
325
  <div class="foot"><a href="https://github.com/ratingtesting/keelwright" style="color:#94a3b8;text-decoration:underline">keelwright</a> by ratingtesting · CC BY 4.0 · <a href="https://huggingface.co/spaces/ratingtesting/keelwright" style="color:#94a3b8;text-decoration:underline">Hugging Face</a> · source of truth: assets/architecture.md</div>
326
  </body>
327
  </html>