Add five-model comparisons on MMDM and JevBench Public

README.md

@@ -216,65 +216,254 @@ The stages use a shuffled full-pass schedule. SFT masks final distributed paddin
 
 ## 4. Evaluation Results
 
-We evaluate the released **OpenJev-4B** on [MMDM](https://huggingface.co/datasets/shenjunhao/mmdm) and [JevBench Public](https://github.com/fstandhartinger/jevbench). Only the final release model is shown below. MMDM separates hard and soft targets; JevBench Public is organized into Original, Easy, and Hard cohorts.
+We compare **OpenJev-4B** with **Jev, Decider 4B, JevK5 4B, and Intern-Decision-4B** on [MMDM](https://huggingface.co/datasets/shenjunhao/mmdm) and [JevBench Public](https://github.com/fstandhartinger/jevbench). The OpenJev column contains only the final release model. MMDM is grouped by hard/soft target type and task category; JevBench Public is grouped into Original, Easy, and Hard cohorts. Each hard/Public cell shows **Accuracy (%) ↑ / Brier ↓**, while soft-target cells show **Brier ↓**.
 
 <style>
 .openjev-eval th,.openjev-eval td{box-sizing:border-box}
 .openjev-eval td{font-variant-numeric:tabular-nums}
+.openjev-eval sup{font-size:9px;line-height:0;vertical-align:super;margin-left:2px}
 .openjev-eval a{color:inherit;text-decoration:none}
-.openjev-eval sup{font-size:10px;margin-left:3px}
+.openjev-eval a:hover{text-decoration:underline}
+.openjev-eval .oj-muted{color:#738895;font-size:11px;font-weight:400}
+.openjev-eval .oj-separator{color:#9baeb9;padding:0 3px}
 @media(prefers-color-scheme:dark){
 .openjev-eval{color:#dce5eb!important}
 .openjev-eval th{color:#bfd1de!important;border-color:#7894a5!important}
 .openjev-eval .oj-band{background:#344650!important;color:#d9e5ed!important}
-.openjev-eval .oj-total{background:#2e3e48!important;color:#e2ebf0!important}
+.openjev-eval .oj-soft{background:#3b4756!important;color:#d8e0eb!important}
+.openjev-eval .oj-own{background:rgba(137,163,184,.13)!important}
+.openjev-eval .oj-total{background:rgba(137,163,184,.16)!important}
+.openjev-eval .oj-muted{color:#a3b4c0!important}
 }
 </style>
-<div style="overflow-x:auto;margin:18px 0">
-<table class="openjev-eval" style="display:table!important;width:100%;min-width:540px;border-collapse:collapse;table-layout:fixed;font-family:-apple-system,BlinkMacSystemFont,Segoe UI,sans-serif;font-size:13px;line-height:1.5;color:#374f60">
-<thead><tr><th style="width:46%;padding:13px 12px;text-align:left;font-weight:600;border-bottom:2px solid #7894a5;color:#486779">Evaluation slice</th><th style="width:16%;padding:13px 12px;text-align:center;font-weight:600;border-bottom:2px solid #7894a5;color:#486779">Decisions</th><th style="width:19%;padding:13px 12px;text-align:center;font-weight:600;border-bottom:2px solid #7894a5;color:#486779">Accuracy (%) ↑</th><th style="width:19%;padding:13px 12px;text-align:center;font-weight:600;border-bottom:2px solid #7894a5;color:#486779">Brier ↓</th></tr></thead>
+<div style="overflow-x:auto;margin:18px 0 10px">
+<table class="openjev-eval" style="display:table!important;width:100%;min-width:860px;table-layout:fixed;border-collapse:collapse;font-family:-apple-system,BlinkMacSystemFont,Segoe UI,sans-serif;font-size:12px;color:#374f60;line-height:1.4">
+<thead><tr>
+<th style="width:23%;padding:14px 12px;text-align:left;border-bottom:2px solid #7894a5;color:#486779;font-weight:600">Evaluation slice</th>
+<th class="oj-own" style="width:15.4%;padding:14px 8px;text-align:center;border-bottom:2px solid #7894a5;color:#486779;background:rgba(137,163,184,.11);"><a href="https://huggingface.co/shenjunhao/OpenJev-4B">OpenJev-4B</a></th>
+<th class="" style="width:15.4%;padding:14px 8px;text-align:center;border-bottom:2px solid #7894a5;color:#486779;"><a href="https://docs.typesafe.ai/introduction/quickstart">Jev</a><sup><a href="#openjev-note-1">1</a></sup></th>
+<th class="" style="width:15.4%;padding:14px 8px;text-align:center;border-bottom:2px solid #7894a5;color:#486779;"><a href="https://huggingface.co/Mapika/decider-4b">Decider 4B</a><sup><a href="#openjev-note-3">3</a></sup></th>
+<th class="" style="width:15.4%;padding:14px 8px;text-align:center;border-bottom:2px solid #7894a5;color:#486779;"><a href="https://huggingface.co/alibiserikbay/JevK5">JevK5 4B</a><sup><a href="#openjev-note-3">3</a></sup></th>
+<th class="" style="width:15.4%;padding:14px 8px;text-align:center;border-bottom:2px solid #7894a5;color:#486779;"><a href="https://huggingface.co/internlm/Intern-Decision-4B">Intern-Decision-4B</a><sup><a href="#openjev-note-2">2</a></sup></th>
+</tr></thead>
 <tbody>
-<tr><td colspan="4" class="oj-band" style="padding:10px 12px;background:#dfe8ed;color:#486779;font-weight:650;letter-spacing:.25px">MMDM · Hard<sup><a href="#eval-note-1">1</a></sup></td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Answer preference</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">185</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">90.81</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.1236</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Code / database</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">122</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">84.43</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.2128</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Commonsense context</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">355</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">93.80</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0854</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Constraint planning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">334</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">76.05</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.2944</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Deductive reasoning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">619</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">90.47</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.1389</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Emotion recognition</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">9</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">100.00</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0243</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Evidence validation</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">177</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">91.53</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.1221</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Intent routing</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">253</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">94.07</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.1045</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Knowledge QA</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">572</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">95.63</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0693</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Multi-hop reasoning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">211</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">92.42</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0992</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Option logic</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">287</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">51.92</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.5694</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Policy exceptions</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">576</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">95.31</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0688</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Quantitative reasoning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">330</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">69.09</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.3752</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Reading comprehension</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">48</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">95.83</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0523</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Score rubrics</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">135</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">78.52</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.2939</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Temporal / numeric</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">138</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">55.07</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.4695</td></tr>
-<tr class="oj-total" style="background:#eaf0f3;"><td style="padding:8px 12px;text-align:left;border-bottom:2px solid #afc0cb;font-weight:650">Total</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">4,351</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">85.57</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">0.1864</td></tr>
-<tr><td colspan="4" class="oj-band" style="padding:10px 12px;background:#dfe8ed;color:#486779;font-weight:650;letter-spacing:.25px">MMDM · Soft<sup><a href="#eval-note-1">1</a></sup></td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Constraint planning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">1</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0070</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Evidence validation</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">22</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.1066</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Intent routing</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">22</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0681</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Policy exceptions</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">18</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0423</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Probabilistic reasoning</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">505</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0278</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Score rubrics</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">30</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0864</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Temporal / numeric</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">12</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">—</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.2168</td></tr>
-<tr class="oj-total" style="background:#eaf0f3;"><td style="padding:8px 12px;text-align:left;border-bottom:2px solid #afc0cb;font-weight:650">Total</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">610</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">—</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">0.0391</td></tr>
-<tr><td colspan="4" class="oj-band" style="padding:10px 12px;background:#dfe8ed;color:#486779;font-weight:650;letter-spacing:.25px">JevBench · Public<sup><a href="#eval-note-3">3</a></sup></td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Original</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">72</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">100.00</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.0312</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Easy</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">48</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">100.00</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.000011</td></tr>
-<tr class="" style=""><td style="padding:8px 12px;text-align:left;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">Hard</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">111</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">75.68</td><td style="padding:8px 12px;text-align:center;border-bottom:1px solid rgba(118,147,164,.20);font-weight:400">0.3469</td></tr>
-<tr class="oj-total" style="background:#eaf0f3;"><td style="padding:8px 12px;text-align:left;border-bottom:2px solid #afc0cb;font-weight:650">Total</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">231</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">88.31</td><td style="padding:8px 12px;text-align:center;border-bottom:2px solid #afc0cb;font-weight:650">0.1764</td></tr>
+<tr><td colspan="6" class="oj-band " style="padding:10px 12px;background:#dfe8ed;color:#486779;font-weight:600;letter-spacing:.3px">MMDM · Hard · Accuracy (%) ↑ / Brier ↓<sup><a href="#openjev-note-4">4</a></sup></td></tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Answer preference<br><span class="oj-muted">n = 185</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>90.81</strong><span class="oj-separator">/</span><strong>0.1236</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">88.11<span class="oj-separator">/</span>0.1794</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">85.95<span class="oj-separator">/</span>0.2263</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.84<span class="oj-separator">/</span>0.3276</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.84<span class="oj-separator">/</span>0.3142</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Code / database<br><span class="oj-muted">n = 122</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>84.43</strong><span class="oj-separator">/</span><strong>0.2128</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">83.61<span class="oj-separator">/</span>0.2189</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.05<span class="oj-separator">/</span>0.3293</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">72.95<span class="oj-separator">/</span>0.3546</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">68.03<span class="oj-separator">/</span>0.3364</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Commonsense context<br><span class="oj-muted">n = 355</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">93.80<span class="oj-separator">/</span>0.0854</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>96.90</strong><span class="oj-separator">/</span><strong>0.0409</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">95.21<span class="oj-separator">/</span>0.0687</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">93.80<span class="oj-separator">/</span>0.1073</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">87.04<span class="oj-separator">/</span>0.2096</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Constraint planning<br><span class="oj-muted">n = 334</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>76.05</strong><span class="oj-separator">/</span>0.2944</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">75.75<span class="oj-separator">/</span><strong>0.2859</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">71.26<span class="oj-separator">/</span>0.3685</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">64.37<span class="oj-separator">/</span>0.4816</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">58.38<span class="oj-separator">/</span>0.5222</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Deductive reasoning<br><span class="oj-muted">n = 619</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">90.47<span class="oj-separator">/</span>0.1389</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>94.51</strong><span class="oj-separator">/</span><strong>0.0781</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">86.27<span class="oj-separator">/</span>0.1861</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">79.97<span class="oj-separator">/</span>0.2809</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">68.66<span class="oj-separator">/</span>0.3848</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Emotion recognition<br><span class="oj-muted">n = 9</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>100.00</strong><span class="oj-separator">/</span><strong>0.0243</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.78<span class="oj-separator">/</span>0.2353</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span>0.0684</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.78<span class="oj-separator">/</span>0.3633</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">77.78<span class="oj-separator">/</span>0.3690</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Evidence validation<br><span class="oj-muted">n = 177</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>91.53</strong><span class="oj-separator">/</span>0.1221</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">90.96<span class="oj-separator">/</span><strong>0.1211</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">85.31<span class="oj-separator">/</span>0.2103</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">88.14<span class="oj-separator">/</span>0.1960</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">84.18<span class="oj-separator">/</span>0.2301</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Intent routing<br><span class="oj-muted">n = 253</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>94.07</strong><span class="oj-separator">/</span><strong>0.1045</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">93.28<span class="oj-separator">/</span>0.1047</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">92.49<span class="oj-separator">/</span>0.1223</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">87.75<span class="oj-separator">/</span>0.1688</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">64.82<span class="oj-separator">/</span>0.2352</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Knowledge QA<br><span class="oj-muted">n = 572</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">95.63<span class="oj-separator">/</span>0.0693</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>98.95</strong><span class="oj-separator">/</span><strong>0.0202</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">96.85<span class="oj-separator">/</span>0.0527</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">94.76<span class="oj-separator">/</span>0.0847</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">88.99<span class="oj-separator">/</span>0.1722</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Multi-hop reasoning<br><span class="oj-muted">n = 211</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>92.42</strong><span class="oj-separator">/</span><strong>0.0992</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">89.57<span class="oj-separator">/</span>0.1274</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">73.93<span class="oj-separator">/</span>0.3978</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">74.41<span class="oj-separator">/</span>0.3046</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">64.45<span class="oj-separator">/</span>0.4399</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Option logic<br><span class="oj-muted">n = 287</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>51.92</strong><span class="oj-separator">/</span><strong>0.5694</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">41.11<span class="oj-separator">/</span>0.6551</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">34.84<span class="oj-separator">/</span>0.7715</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">35.19<span class="oj-separator">/</span>0.7176</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">28.22<span class="oj-separator">/</span>0.7979</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Policy exceptions<br><span class="oj-muted">n = 576</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>95.31</strong><span class="oj-separator">/</span><strong>0.0688</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">90.97<span class="oj-separator">/</span>0.1301</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">82.81<span class="oj-separator">/</span>0.2620</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">72.74<span class="oj-separator">/</span>0.3858</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">63.89<span class="oj-separator">/</span>0.4630</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Quantitative reasoning<br><span class="oj-muted">n = 330</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">69.09<span class="oj-separator">/</span>0.3752</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>77.88</strong><span class="oj-separator">/</span><strong>0.2841</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">51.21<span class="oj-separator">/</span>0.6098</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">46.36<span class="oj-separator">/</span>0.6636</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">37.58<span class="oj-separator">/</span>0.7533</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Reading comprehension<br><span class="oj-muted">n = 48</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">95.83<span class="oj-separator">/</span>0.0523</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span><strong>0.0091</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">95.83<span class="oj-separator">/</span>0.0640</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">91.67<span class="oj-separator">/</span>0.1074</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">89.58<span class="oj-separator">/</span>0.1868</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Score rubrics<br><span class="oj-muted">n = 135</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">78.52<span class="oj-separator">/</span>0.2939</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>85.93</strong><span class="oj-separator">/</span><strong>0.2055</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">72.59<span class="oj-separator">/</span>0.3487</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">73.33<span class="oj-separator">/</span>0.3190</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">69.63<span class="oj-separator">/</span>0.3827</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Temporal / numeric<br><span class="oj-muted">n = 138</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">55.07<span class="oj-separator">/</span>0.4695</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>64.49</strong><span class="oj-separator">/</span><strong>0.4235</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">54.35<span class="oj-separator">/</span>0.5197</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">53.62<span class="oj-separator">/</span>0.5379</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">59.42<span class="oj-separator">/</span>0.5422</td>
+</tr>
+<tr class="oj-total" style="background:#eaf0f3;border-top:2px solid #afc0cb"><td style="padding:9px 12px;border-bottom:2px solid #afc0cb;font-weight:650">Total<br><span class="oj-muted">n = 4,351</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;background:rgba(137,163,184,.11);">85.57<span class="oj-separator">/</span>0.1864</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;"><strong>86.37</strong><span class="oj-separator">/</span><strong>0.1733</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">78.88<span class="oj-separator">/</span>0.2832</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">74.70<span class="oj-separator">/</span>0.3330</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">66.95<span class="oj-separator">/</span>0.4069</td>
+</tr>
+<tr><td colspan="6" class="oj-band oj-soft" style="padding:10px 12px;background:#e4e8f0;color:#486779;font-weight:600;letter-spacing:.3px">MMDM · Soft · Brier ↓<sup><a href="#openjev-note-4">4</a></sup></td></tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Constraint planning<br><span class="oj-muted">n = 1</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>0.0070</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1412</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1338</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0251</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0258</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Evidence validation<br><span class="oj-muted">n = 22</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">0.1066</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>0.0493</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1517</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0773</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0914</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Intent routing<br><span class="oj-muted">n = 22</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>0.0681</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1570</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1366</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1304</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0838</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Policy exceptions<br><span class="oj-muted">n = 18</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>0.0423</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0606</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1128</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0916</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0674</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Probabilistic reasoning<br><span class="oj-muted">n = 505</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>0.0278</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1164</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1387</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1092</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1242</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Score rubrics<br><span class="oj-muted">n = 30</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>0.0864</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.0895</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1653</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1242</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1746</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Temporal / numeric<br><span class="oj-muted">n = 12</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">0.2168</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>0.0822</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1100</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1283</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">0.1747</td>
+</tr>
+<tr class="oj-total" style="background:#eaf0f3;border-top:2px solid #afc0cb"><td style="padding:9px 12px;border-bottom:2px solid #afc0cb;font-weight:650">Total<br><span class="oj-muted">n = 610</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;background:rgba(137,163,184,.11);"><strong>0.0391</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">0.1119</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">0.1391</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">0.1092</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">0.1232</td>
+</tr>
+<tr><td colspan="6" class="oj-band " style="padding:10px 12px;background:#dfe8ed;color:#486779;font-weight:600;letter-spacing:.3px">JevBench · Public · Accuracy (%) ↑ / Brier ↓<sup><a href="#openjev-note-5">5</a></sup></td></tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Original<br><span class="oj-muted">n = 72</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>100.00</strong><span class="oj-separator">/</span>0.0312</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">98.61<span class="oj-separator">/</span><strong>0.0283</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">98.61<span class="oj-separator">/</span>0.0574</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">94.44<span class="oj-separator">/</span>0.1075</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">98.61<span class="oj-separator">/</span>0.0340</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Easy<br><span class="oj-muted">n = 48</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);"><strong>100.00</strong><span class="oj-separator">/</span><strong>0.000011</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span>0.000196</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span>0.000045</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span>0.002509</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>100.00</strong><span class="oj-separator">/</span>0.001119</td>
+</tr>
+<tr class="" style="background:transparent;border-top:none"><td style="padding:9px 12px;border-bottom:1px solid rgba(118,147,164,.20);font-weight:500">Hard<br><span class="oj-muted">n = 111</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);background:rgba(137,163,184,.11);">75.68<span class="oj-separator">/</span>0.3469</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">72.07<span class="oj-separator">/</span>0.3594</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">64.86<span class="oj-separator">/</span>0.4686</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);"><strong>77.48</strong><span class="oj-separator">/</span><strong>0.3118</strong></td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:1px solid rgba(118,147,164,.20);">73.87<span class="oj-separator">/</span>0.3439</td>
+</tr>
+<tr class="oj-total" style="background:#eaf0f3;border-top:2px solid #afc0cb"><td style="padding:9px 12px;border-bottom:2px solid #afc0cb;font-weight:650">Total<br><span class="oj-muted">n = 231</span></td>
+<td class="oj-own" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;background:rgba(137,163,184,.11);"><strong>88.31</strong><span class="oj-separator">/</span>0.1764</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">86.15<span class="oj-separator">/</span>0.1816</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">82.68<span class="oj-separator">/</span>0.2431</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">87.45<span class="oj-separator">/</span>0.1839</td>
+<td class="" style="padding:9px 5px;text-align:center;white-space:nowrap;border-bottom:2px solid #afc0cb;">87.01<span class="oj-separator">/</span><strong>0.1761</strong></td>
+</tr>
 </tbody>
 </table>
 </div>
 
 <ol style="font-size:12px;line-height:1.65;color:#627f92">
-<li id="eval-note-1"><strong>MMDM:</strong> the published test split contains 4,961 decisions: 4,351 hard and 610 soft. Accuracy uses all hard examples. To match the MMDM dataset card, hard Brier uses its 4,294-example common-valid comparison cohort; intent routing uses 196 of 253 examples for Brier. OpenJev itself produces valid predictions for every MMDM example. Soft Brier uses all 610 examples.</li>
-<li id="eval-note-2"><strong>Metrics:</strong> Accuracy is the percentage of top-option predictions matching the gold outcome. Brier is the mean sum of squared probability errors across options, without division by option count. Soft-target accuracy is left blank because a single selected label does not measure agreement with a full target distribution. Totals are weighted by their respective scoring counts.</li>
-<li id="eval-note-3"><strong>JevBench Public:</strong> these are local evaluations of the v1.3.0 public suite: Original (72), Easy (48), and Hard (111). Accuracy and Brier follow its gold-label scoring protocol; all 231 predictions are valid. “Hard” here names a difficulty cohort, whereas MMDM Hard names a target type. Public results are separate from the official private leaderboard.</li>
-<li id="eval-note-4"><strong>Evaluation context:</strong> the published MMDM split combines the previously separate development and test sets, so these numbers are not a new blind evaluation. Training includes newly constructed scenarios informed by public decision-task patterns. The tables report performance on these fixed evaluation collections.</li>
+<li id="openjev-note-1"><strong>Jev:</strong> official API, evaluated as Jev 1.13.0.</li>
+<li id="openjev-note-2"><strong>Intern-Decision-4B:</strong> official inference code and temperature 1.99241824, BF16, SDPA, no input truncation, in the existing Torch 2.11.0 / Transformers 5.17.0 environment. Its option-count limit prevents scoring 57 MMDM hard examples; these count as incorrect in accuracy and are excluded from the common probability-metric cohort.</li>
+<li id="openjev-note-3"><strong>External model versions:</strong> Decider 4B v2.1 and JevK5 4B v0.3. Their returned decision distributions are used for probability metrics. OpenJev-4B is our final release model; earlier OpenJev checkpoints are omitted.</li>
+<li id="openjev-note-4"><strong>MMDM:</strong> the published test split contains 4,961 decisions: 4,351 hard and 610 soft. Accuracy uses all hard examples. Brier uses the same 4,294 common valid hard examples across models (196 of 253 for intent routing), and all 610 soft examples, matching the dataset card. All models' totals use the corresponding scoring counts. Soft rows show Brier only. Bold marks the best value for each metric; zero probabilities are not clipped.</li>
+<li id="openjev-note-5"><strong>JevBench Public:</strong> local evaluations of the v1.3.0 public suite: Original (72), Easy (48), Hard (111). Accuracy and Brier follow its gold-label scoring protocol; all five models return valid predictions for all 231 examples. “Hard” names a difficulty cohort here, rather than a target type. These results are separate from the official private leaderboard. Easy Brier is shown to six decimal places to retain the small differences.</li>
+<li id="openjev-note-6"><strong>Evaluation context:</strong> the published MMDM split combines the previously separate development and test sets, so these numbers are not a new blind evaluation. OpenJev training includes newly constructed scenarios informed by public decision-task patterns. The tables report performance on these fixed evaluation collections.</li>
 </ol>
 
 For hard labels, the Brier target is the one-hot gold outcome. For MMDM soft targets, it is the full supplied distribution:

evaluation.json

@@ -1,214 +1,806 @@
 {
   "model": "OpenJev-4B",
+  "comparison_models": {
+    "openjev": "OpenJev-4B",
+    "jev": "Jev",
+    "decider": "Decider 4B",
+    "jevk5": "JevK5 4B",
+    "intern": "Intern-Decision-4B"
+  },
+  "versions": {
+    "jev": "1.13.0",
+    "decider": "2.1",
+    "jevk5": "0.3"
+  },
   "snapshot_date": "2026-09-29",
   "rows": [
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Answer preference",
       "n": 185,
-      "accuracy": 90.81,
-      "brier": 0.1236
+      "models": {
+        "openjev": {
+          "accuracy_percent": 90.81,
+          "brier": 0.1236
+        },
+        "jev": {
+          "accuracy_percent": 88.11,
+          "brier": 0.1794
+        },
+        "decider": {
+          "accuracy_percent": 85.95,
+          "brier": 0.2263
+        },
+        "jevk5": {
+          "accuracy_percent": 77.84,
+          "brier": 0.3276
+        },
+        "intern": {
+          "accuracy_percent": 77.84,
+          "brier": 0.3142
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Code / database",
       "n": 122,
-      "accuracy": 84.43,
-      "brier": 0.2128
+      "models": {
+        "openjev": {
+          "accuracy_percent": 84.43,
+          "brier": 0.2128
+        },
+        "jev": {
+          "accuracy_percent": 83.61,
+          "brier": 0.2189
+        },
+        "decider": {
+          "accuracy_percent": 77.05,
+          "brier": 0.3293
+        },
+        "jevk5": {
+          "accuracy_percent": 72.95,
+          "brier": 0.3546
+        },
+        "intern": {
+          "accuracy_percent": 68.03,
+          "brier": 0.3364
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Commonsense context",
       "n": 355,
-      "accuracy": 93.8,
-      "brier": 0.0854
+      "models": {
+        "openjev": {
+          "accuracy_percent": 93.8,
+          "brier": 0.0854
+        },
+        "jev": {
+          "accuracy_percent": 96.9,
+          "brier": 0.0409
+        },
+        "decider": {
+          "accuracy_percent": 95.21,
+          "brier": 0.0687
+        },
+        "jevk5": {
+          "accuracy_percent": 93.8,
+          "brier": 0.1073
+        },
+        "intern": {
+          "accuracy_percent": 87.04,
+          "brier": 0.2096
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Constraint planning",
       "n": 334,
-      "accuracy": 76.05,
-      "brier": 0.2944
+      "models": {
+        "openjev": {
+          "accuracy_percent": 76.05,
+          "brier": 0.2944
+        },
+        "jev": {
+          "accuracy_percent": 75.75,
+          "brier": 0.2859
+        },
+        "decider": {
+          "accuracy_percent": 71.26,
+          "brier": 0.3685
+        },
+        "jevk5": {
+          "accuracy_percent": 64.37,
+          "brier": 0.4816
+        },
+        "intern": {
+          "accuracy_percent": 58.38,
+          "brier": 0.5222
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Deductive reasoning",
       "n": 619,
-      "accuracy": 90.47,
-      "brier": 0.1389
+      "models": {
+        "openjev": {
+          "accuracy_percent": 90.47,
+          "brier": 0.1389
+        },
+        "jev": {
+          "accuracy_percent": 94.51,
+          "brier": 0.0781
+        },
+        "decider": {
+          "accuracy_percent": 86.27,
+          "brier": 0.1861
+        },
+        "jevk5": {
+          "accuracy_percent": 79.97,
+          "brier": 0.2809
+        },
+        "intern": {
+          "accuracy_percent": 68.66,
+          "brier": 0.3848
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Emotion recognition",
       "n": 9,
-      "accuracy": 100.0,
-      "brier": 0.0243
+      "models": {
+        "openjev": {
+          "accuracy_percent": 100.0,
+          "brier": 0.0243
+        },
+        "jev": {
+          "accuracy_percent": 77.78,
+          "brier": 0.2353
+        },
+        "decider": {
+          "accuracy_percent": 100.0,
+          "brier": 0.0684
+        },
+        "jevk5": {
+          "accuracy_percent": 77.78,
+          "brier": 0.3633
+        },
+        "intern": {
+          "accuracy_percent": 77.78,
+          "brier": 0.369
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Evidence validation",
       "n": 177,
-      "accuracy": 91.53,
-      "brier": 0.1221
+      "models": {
+        "openjev": {
+          "accuracy_percent": 91.53,
+          "brier": 0.1221
+        },
+        "jev": {
+          "accuracy_percent": 90.96,
+          "brier": 0.1211
+        },
+        "decider": {
+          "accuracy_percent": 85.31,
+          "brier": 0.2103
+        },
+        "jevk5": {
+          "accuracy_percent": 88.14,
+          "brier": 0.196
+        },
+        "intern": {
+          "accuracy_percent": 84.18,
+          "brier": 0.2301
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Intent routing",
       "n": 253,
-      "accuracy": 94.07,
-      "brier": 0.1045
+      "models": {
+        "openjev": {
+          "accuracy_percent": 94.07,
+          "brier": 0.1045
+        },
+        "jev": {
+          "accuracy_percent": 93.28,
+          "brier": 0.1047
+        },
+        "decider": {
+          "accuracy_percent": 92.49,
+          "brier": 0.1223
+        },
+        "jevk5": {
+          "accuracy_percent": 87.75,
+          "brier": 0.1688
+        },
+        "intern": {
+          "accuracy_percent": 64.82,
+          "brier": 0.2352
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Knowledge QA",
       "n": 572,
-      "accuracy": 95.63,
-      "brier": 0.0693
+      "models": {
+        "openjev": {
+          "accuracy_percent": 95.63,
+          "brier": 0.0693
+        },
+        "jev": {
+          "accuracy_percent": 98.95,
+          "brier": 0.0202
+        },
+        "decider": {
+          "accuracy_percent": 96.85,
+          "brier": 0.0527
+        },
+        "jevk5": {
+          "accuracy_percent": 94.76,
+          "brier": 0.0847
+        },
+        "intern": {
+          "accuracy_percent": 88.99,
+          "brier": 0.1722
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Multi-hop reasoning",
       "n": 211,
-      "accuracy": 92.42,
-      "brier": 0.0992
+      "models": {
+        "openjev": {
+          "accuracy_percent": 92.42,
+          "brier": 0.0992
+        },
+        "jev": {
+          "accuracy_percent": 89.57,
+          "brier": 0.1274
+        },
+        "decider": {
+          "accuracy_percent": 73.93,
+          "brier": 0.3978
+        },
+        "jevk5": {
+          "accuracy_percent": 74.41,
+          "brier": 0.3046
+        },
+        "intern": {
+          "accuracy_percent": 64.45,
+          "brier": 0.4399
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Option logic",
       "n": 287,
-      "accuracy": 51.92,
-      "brier": 0.5694
+      "models": {
+        "openjev": {
+          "accuracy_percent": 51.92,
+          "brier": 0.5694
+        },
+        "jev": {
+          "accuracy_percent": 41.11,
+          "brier": 0.6551
+        },
+        "decider": {
+          "accuracy_percent": 34.84,
+          "brier": 0.7715
+        },
+        "jevk5": {
+          "accuracy_percent": 35.19,
+          "brier": 0.7176
+        },
+        "intern": {
+          "accuracy_percent": 28.22,
+          "brier": 0.7979
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Policy exceptions",
       "n": 576,
-      "accuracy": 95.31,
-      "brier": 0.0688
+      "models": {
+        "openjev": {
+          "accuracy_percent": 95.31,
+          "brier": 0.0688
+        },
+        "jev": {
+          "accuracy_percent": 90.97,
+          "brier": 0.1301
+        },
+        "decider": {
+          "accuracy_percent": 82.81,
+          "brier": 0.262
+        },
+        "jevk5": {
+          "accuracy_percent": 72.74,
+          "brier": 0.3858
+        },
+        "intern": {
+          "accuracy_percent": 63.89,
+          "brier": 0.463
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Quantitative reasoning",
       "n": 330,
-      "accuracy": 69.09,
-      "brier": 0.3752
+      "models": {
+        "openjev": {
+          "accuracy_percent": 69.09,
+          "brier": 0.3752
+        },
+        "jev": {
+          "accuracy_percent": 77.88,
+          "brier": 0.2841
+        },
+        "decider": {
+          "accuracy_percent": 51.21,
+          "brier": 0.6098
+        },
+        "jevk5": {
+          "accuracy_percent": 46.36,
+          "brier": 0.6636
+        },
+        "intern": {
+          "accuracy_percent": 37.58,
+          "brier": 0.7533
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Reading comprehension",
       "n": 48,
-      "accuracy": 95.83,
-      "brier": 0.0523
+      "models": {
+        "openjev": {
+          "accuracy_percent": 95.83,
+          "brier": 0.0523
+        },
+        "jev": {
+          "accuracy_percent": 100.0,
+          "brier": 0.0091
+        },
+        "decider": {
+          "accuracy_percent": 95.83,
+          "brier": 0.064
+        },
+        "jevk5": {
+          "accuracy_percent": 91.67,
+          "brier": 0.1074
+        },
+        "intern": {
+          "accuracy_percent": 89.58,
+          "brier": 0.1868
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Score rubrics",
       "n": 135,
-      "accuracy": 78.52,
-      "brier": 0.2939
+      "models": {
+        "openjev": {
+          "accuracy_percent": 78.52,
+          "brier": 0.2939
+        },
+        "jev": {
+          "accuracy_percent": 85.93,
+          "brier": 0.2055
+        },
+        "decider": {
+          "accuracy_percent": 72.59,
+          "brier": 0.3487
+        },
+        "jevk5": {
+          "accuracy_percent": 73.33,
+          "brier": 0.319
+        },
+        "intern": {
+          "accuracy_percent": 69.63,
+          "brier": 0.3827
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Temporal / numeric",
       "n": 138,
-      "accuracy": 55.07,
-      "brier": 0.4695
+      "models": {
+        "openjev": {
+          "accuracy_percent": 55.07,
+          "brier": 0.4695
+        },
+        "jev": {
+          "accuracy_percent": 64.49,
+          "brier": 0.4235
+        },
+        "decider": {
+          "accuracy_percent": 54.35,
+          "brier": 0.5197
+        },
+        "jevk5": {
+          "accuracy_percent": 53.62,
+          "brier": 0.5379
+        },
+        "intern": {
+          "accuracy_percent": 59.42,
+          "brier": 0.5422
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Hard",
       "task": "Total",
       "n": 4351,
-      "accuracy": 85.57,
-      "brier": 0.1864
+      "models": {
+        "openjev": {
+          "accuracy_percent": 85.57,
+          "brier": 0.1864
+        },
+        "jev": {
+          "accuracy_percent": 86.37,
+          "brier": 0.1733
+        },
+        "decider": {
+          "accuracy_percent": 78.88,
+          "brier": 0.2832
+        },
+        "jevk5": {
+          "accuracy_percent": 74.7,
+          "brier": 0.333
+        },
+        "intern": {
+          "accuracy_percent": 66.95,
+          "brier": 0.4069
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Constraint planning",
       "n": 1,
-      "accuracy": null,
-      "brier": 0.007
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.007
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.1412
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1338
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.0251
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.0258
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Evidence validation",
       "n": 22,
-      "accuracy": null,
-      "brier": 0.1066
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.1066
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.0493
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1517
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.0773
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.0914
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Intent routing",
       "n": 22,
-      "accuracy": null,
-      "brier": 0.0681
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.0681
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.157
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1366
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.1304
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.0838
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Policy exceptions",
       "n": 18,
-      "accuracy": null,
-      "brier": 0.0423
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.0423
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.0606
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1128
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.0916
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.0674
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Probabilistic reasoning",
       "n": 505,
-      "accuracy": null,
-      "brier": 0.0278
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.0278
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.1164
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1387
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.1092
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.1242
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Score rubrics",
       "n": 30,
-      "accuracy": null,
-      "brier": 0.0864
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.0864
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.0895
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1653
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.1242
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.1746
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Temporal / numeric",
       "n": 12,
-      "accuracy": null,
-      "brier": 0.2168
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.2168
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.0822
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.11
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.1283
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.1747
+        }
+      }
     },
     {
       "group": "MMDM \u00b7 Soft",
       "task": "Total",
       "n": 610,
-      "accuracy": null,
-      "brier": 0.0391
+      "models": {
+        "openjev": {
+          "accuracy_percent": null,
+          "brier": 0.0391
+        },
+        "jev": {
+          "accuracy_percent": null,
+          "brier": 0.1119
+        },
+        "decider": {
+          "accuracy_percent": null,
+          "brier": 0.1391
+        },
+        "jevk5": {
+          "accuracy_percent": null,
+          "brier": 0.1092
+        },
+        "intern": {
+          "accuracy_percent": null,
+          "brier": 0.1232
+        }
+      }
     },
     {
       "group": "JevBench \u00b7 Public",
       "task": "Original",
-      "n": 72,
-      "accuracy": 100.0,
-      "brier": 0.031173049329996292
+      "models": {
+        "openjev": {
+          "accuracy_percent": 100.0,
+          "brier": 0.031173049329996292
+        },
+        "jev": {
+          "accuracy_percent": 98.61111111111111,
+          "brier": 0.028252777777777776
+        },
+        "decider": {
+          "accuracy_percent": 98.61111111111111,
+          "brier": 0.057440274164038074
+        },
+        "jevk5": {
+          "accuracy_percent": 94.44444444444444,
+          "brier": 0.10753610290390724
+        },
+        "intern": {
+          "accuracy_percent": 98.61111111111111,
+          "brier": 0.034003747946694876
+        }
+      },
+      "n": 72
     },
     {
       "group": "JevBench \u00b7 Public",
       "task": "Easy",
-      "n": 48,
-      "accuracy": 100.0,
-      "brier": 1.0634828094991897e-05
+      "models": {
+        "openjev": {
+          "accuracy_percent": 100.0,
+          "brier": 1.0634828094991897e-05
+        },
+        "jev": {
+          "accuracy_percent": 100.0,
+          "brier": 0.00019583333333333356
+        },
+        "decider": {
+          "accuracy_percent": 100.0,
+          "brier": 4.472370693105332e-05
+        },
+        "jevk5": {
+          "accuracy_percent": 100.0,
+          "brier": 0.0025092999623402795
+        },
+        "intern": {
+          "accuracy_percent": 100.0,
+          "brier": 0.0011188014255948035
+        }
+      },
+      "n": 48
     },
     {
       "group": "JevBench \u00b7 Public",
       "task": "Hard",
-      "n": 111,
-      "accuracy": 75.67567567567568,
-      "brier": 0.3469152429142525
+      "models": {
+        "openjev": {
+          "accuracy_percent": 75.67567567567568,
+          "brier": 0.3469152429142525
+        },
+        "jev": {
+          "accuracy_percent": 72.07207207207207,
+          "brier": 0.35944144144144147
+        },
+        "decider": {
+          "accuracy_percent": 64.86486486486487,
+          "brier": 0.4685710696134996
+        },
+        "jevk5": {
+          "accuracy_percent": 77.47747747747748,
+          "brier": 0.31178986168730566
+        },
+        "intern": {
+          "accuracy_percent": 73.87387387387388,
+          "brier": 0.34391459157831894
+        }
+      },
+      "n": 111
     },
     {
       "group": "JevBench \u00b7 Public",
       "task": "Total",
-      "n": 231,
-      "accuracy": 88.31168831168831,
-      "brier": 0.17641801725969836
+      "models": {
+        "openjev": {
+          "accuracy_percent": 88.31168831168831,
+          "brier": 0.17641801725969836
+        },
+        "jev": {
+          "accuracy_percent": 86.14718614718615,
+          "brier": 0.18156536796536796
+        },
+        "decider": {
+          "accuracy_percent": 82.68398268398268,
+          "brier": 0.2430702822720428
+        },
+        "jevk5": {
+          "accuracy_percent": 87.44588744588745,
+          "brier": 0.1838602617080718
+        },
+        "intern": {
+          "accuracy_percent": 87.01298701298701,
+          "brier": 0.17608870989516875
+        }
+      },
+      "n": 231
     }
   ],
   "mmdm_hard_brier_n": 4294,
   "mmdm_soft_brier_n": 610,
   "mmdm_source_revision": "4af5dba7284e7f72f7ffeb2357c391520813d987",
   "public_suite": "original72 + easy48 + hard111",
-  "notes": "MMDM category cells match the dataset card rounding; Public values retain full stored precision."
+  "notes": "MMDM cells retain dataset-card rounding; Public values retain stored precision. Accuracy is expressed as a percentage."
 }