org card: per-experiment hyperparameters and VQA/caption results

README.md

@@ -117,3 +117,11 @@ The top eight runs are statistically tied: every confidence interval against the
 - **Extra QA per image does not help.** 4-QA and 6-QA mixtures tie with 2-QA.
 - **Larger batch trades generality for domain fit.** Effective batch 64 gives the best in-domain captioning in the table and the worst RSICD VQA.
 - **Data is the remaining lever**, not hyperparameters.
+
+## Status
+
+The hyperparameter search is **closed**. Every axis available in this setup has been swept — rank, alpha, dropout, learning rate, schedule, weight decay, gradient clipping, vision-tower freezing, QA-per-image, caption prompt pool, batch geometry, effective batch size, and learning rate at large batch. Seventeen recipes were trained and evaluated under identical conditions, and none displaced the reference recipe by more than the paired error bar.
+
+The final three runs illustrate the ceiling. Effective batch 16 came in at **-0.59** overall, batch 64 at 2e-4 at **-0.56**, and batch 64 at 3e-4 at **-0.34** overall and **-0.87** in-domain; batch 64 at 4e-4 was stopped at epoch 4 while declining. In each case captioning held steady or improved while VQA fell, which is the same trade seen across the whole sweep.
+
+Any further gain has to come from the training corpus: more in-domain imagery, denser or higher-quality question-answer pairs, or a different caption source.

index.html

@@ -163,6 +163,10 @@ a { color: #0969da; }
 <li><strong>Larger batch trades generality for domain fit.</strong> Effective batch 64 gives the best in-domain captioning in the table and the worst RSICD VQA.</li>
 <li><strong>Data is the remaining lever</strong>, not hyperparameters.</li>
 </ul>
+<h2>Status</h2>
+<p>The hyperparameter search is <strong>closed</strong>. Every axis available in this setup has been swept — rank, alpha, dropout, learning rate, schedule, weight decay, gradient clipping, vision-tower freezing, QA-per-image, caption prompt pool, batch geometry, effective batch size, and learning rate at large batch. Seventeen recipes were trained and evaluated under identical conditions, and none displaced the reference recipe by more than the paired error bar.</p>
+<p>The final three runs illustrate the ceiling. Effective batch 16 came in at <strong>-0.59</strong> overall, batch 64 at 2e-4 at <strong>-0.56</strong>, and batch 64 at 3e-4 at <strong>-0.34</strong> overall and <strong>-0.87</strong> in-domain; batch 64 at 4e-4 was stopped at epoch 4 while declining. In each case captioning held steady or improved while VQA fell, which is the same trade seen across the whole sweep.</p>
+<p>Any further gain has to come from the training corpus: more in-domain imagery, denser or higher-quality question-answer pairs, or a different caption source.</p>
 </main>
 </body>
 </html>