Model card: leaderboard screenshots (JevBench v1.4.2.2 #1 of 91; DecisionBench #3 of 56)

.gitattributes

@@ -38,3 +38,5 @@ assets/request-listing.png filter=lfs diff=lfs merge=lfs -text
 assets/highlights-dark.png filter=lfs diff=lfs merge=lfs -text
 assets/request-listing-dark.png filter=lfs diff=lfs merge=lfs -text
 assets/request-ticket-dark.png filter=lfs diff=lfs merge=lfs -text
+assets/proof-decisionbench.png filter=lfs diff=lfs merge=lfs -text
+assets/proof-jevbench-score.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -39,6 +39,11 @@ with a probability on each and an explicit <i>can't tell</i>. Your system acts w
 
 <p align="center">The family's <a href="https://huggingface.co/mohit67890/imajev-4b">imajev-4b</a> is <b>#1 of 91 on <a href="https://benchmarkheaven.com/jev-models">JevBench v1.4.2.2</a></b>, the official board for Jev-class decision models (scored 27 Sep 2026). This 9B has no official text-board entry.</p>
 
+<p align="center"><b>Independent results</b> (screenshots of the official leaderboards, 28 Sep 2026)</p>
+<p align="center"><a href="https://benchmarkheaven.com/jev-models"><img src="assets/proof-jevbench-score.png" width="760" alt="JevBench v1.4.2.2 Composite Score: 1 Imajev-4B 67.4, 2 Plumb-4B 65.8, 3 decider-4b v2 64.1, 4 Jev 1.13.0 63.3"></a></p>
+<p align="center"><a href="https://huggingface.co/spaces/Hanno-Labs/decision-bench-leaderboard"><img src="assets/proof-decisionbench.png" width="760" alt="DecisionBench (eng, v1): 1 bosun-v3.1-1.7b 87.29, 2 bosun-v3.1-0.6b 83.20, 3 imajev-4b 79.65, ahead of glm-5.3-flash, jev-1.13, deepseek-v4.1-flash, gpt-5.6-luna"></a></p>
+<p align="center"><sub>The family's imajev-4b: JevBench v1.4.2.2, scored 27 Sep 2026 (<a href="https://benchmarkheaven.com/jev-models">board</a>). DecisionBench (eng, v1): #3 of 56 models, 79.65, ahead of GLM-5.3 Flash (320B), Jev 1.13, DeepSeek V4.1 Flash (552B) and GPT-5.6 Luna; the two above are the benchmark team's own models (<a href="https://huggingface.co/spaces/Hanno-Labs/decision-bench-leaderboard">leaderboard</a>).</sub></p>
+
 <p align="center"><img src="assets/request-listing.png" class="block dark:hidden" alt="imajev-4b checks a listing against its photo: listing.color says red, the photo shows beige shoes; the model names listing.color at 0.999 and the app holds the listing"><img src="assets/request-listing-dark.png" class="hidden dark:block" alt="imajev-4b checks a listing against its photo: listing.color says red, the photo shows beige shoes; the model names listing.color at 0.999 and the app holds the listing"></p>
 
 imajev-9b is the largest size: it automates the most decisions and is strongest on knowledge-heavy text (MMLU 79.2%, measured on the previous imajev-9b).