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

README.md

@@ -20,7 +20,7 @@ LoRA finetunes of **Qwen3-VL-8B-Instruct**, trained jointly on **visual question
 | Targets | all linear layers, vision + language towers |
 | Data | ~6.3k images, 1 caption + 2 QA each (~18.9k samples) |
 | Caption prompts | sampled from a 30-prompt pool, so the model learns the task not one phrasing |
-| Eval | fixed 500-image draw = 1000 graded items, same seed for every model |
+| Eval | fixed 1000-image draw = 2000 graded items, same seed for every model |
 | Scoring | LLM-as-judge, 0-100 per item |
 | References | stock Qwen3-VL-8B, plus an in-house finetuned model |
 
@@ -48,31 +48,31 @@ Constant everywhere: effective batch 32, warmup 0.03, grad clip 0.3 (1.0 for 011
 
 ## Results
 
-Best checkpoint per run. **Overall** = mean of caption and VQA. Right block = in-domain subset, closest to the target distribution.
+Best checkpoint per run, scored on 1000 images (2000 graded items). **Overall** = mean of caption and VQA. Right block = in-domain subset, closest to the target distribution.
 
 | Model | Caption | VQA | Overall | In-dom. caption | In-dom. VQA | In-dom. overall |
 |---|---|---|---|---|---|---|
-| Qwen3-VL-8B (base) | 66.29 | 56.21 | **61.25** | 69.16 | 59.29 | **64.23** |
-| in-house reference | 65.14 | 73.57 | **69.36** | 73.05 | 81.90 | **77.47** |
-| **Aspaklarya-001** | 79.68 | 77.60 | **78.64** | 81.62 | 84.32 | **82.97** |
-| Aspaklarya-011 | 80.42 | 76.75 | **78.59** | 82.58 | 84.77 | **83.68** |
-| Aspaklarya-008 | 80.61 | 75.94 | **78.27** | 81.69 | 83.12 | **82.41** |
-| Aspaklarya-005 | 80.52 | 75.96 | **78.24** | 82.68 | 83.02 | **82.85** |
-| Aspaklarya-012 | 79.85 | 76.50 | **78.17** | 82.23 | 84.05 | **83.14** |
-| Aspaklarya-007 | 80.07 | 76.25 | **78.16** | 81.96 | 81.65 | **81.81** |
-| Aspaklarya-003 | 79.84 | 76.16 | **78.00** | 81.43 | 83.42 | **82.43** |
-| Aspaklarya-013 | 79.54 | 75.94 | **77.74** | 81.35 | 82.50 | **81.92** |
-| Aspaklarya-006 | 79.50 | 75.88 | **77.69** | 81.08 | 83.05 | **82.07** |
-| Aspaklarya-002 | 80.34 | 74.91 | **77.62** | 82.17 | 81.34 | **81.76** |
-| Aspaklarya-009 | 78.46 | 76.37 | **77.42** | 80.19 | 82.43 | **81.31** |
-| Aspaklarya-004 | 79.03 | 75.58 | **77.31** | 80.88 | 81.32 | **81.10** |
-| Aspaklarya-010 | 79.32 | 74.71 | **77.02** | 80.78 | 81.58 | **81.18** |
-
-Every finetune beats both references: **+17.4 overall** over the base model, **+9.3** over the in-house reference. Captioning gains most (66 to 80); VQA gains 21 points over base.
+| Qwen3-VL-8B (base) | 65.68 | 56.01 | **60.84** | 68.37 | 58.94 | **63.66** |
+| in-house reference | 65.18 | 73.34 | **69.26** | 71.88 | 83.62 | **77.75** |
+| **Aspaklarya-001** | 79.97 | 75.94 | **77.95** | 81.84 | 83.23 | **82.54** |
+| Aspaklarya-011 | 79.92 | 75.78 | **77.85** | 81.85 | 83.51 | **82.68** |
+| Aspaklarya-006 | 79.81 | 75.79 | **77.80** | 81.40 | 82.95 | **82.18** |
+| Aspaklarya-005 | 79.55 | 75.74 | **77.64** | 81.37 | 82.37 | **81.87** |
+| Aspaklarya-008 | 79.82 | 75.40 | **77.61** | 80.62 | 81.42 | **81.02** |
+| Aspaklarya-003 | 79.55 | 75.52 | **77.53** | 81.23 | 82.18 | **81.71** |
+| Aspaklarya-012 | 79.73 | 75.02 | **77.37** | 81.68 | 82.05 | **81.86** |
+| Aspaklarya-007 | 79.33 | 75.33 | **77.33** | 81.13 | 80.82 | **80.97** |
+| Aspaklarya-013 | 79.11 | 75.35 | **77.23** | 80.58 | 82.20 | **81.39** |
+| Aspaklarya-009 | 78.53 | 75.85 | **77.19** | 80.01 | 82.34 | **81.18** |
+| Aspaklarya-004 | 78.76 | 75.33 | **77.04** | 80.31 | 81.47 | **80.89** |
+| Aspaklarya-002 | 79.36 | 74.33 | **76.85** | 80.70 | 80.63 | **80.67** |
+| Aspaklarya-010 | 78.88 | 74.28 | **76.58** | 80.50 | 80.66 | **80.58** |
+
+Every finetune beats both references: **+17.1 overall** over the base model, **+8.7** over the in-house reference. Captioning gains most (66 to 80); VQA gains 20 points over base.
 
 ## Caveat
 
-Paired bootstrap over images puts the between-run standard error at **~0.6 overall**, so the top five runs are statistically tied. A different 500-image draw moved the *same* model by 1.8 points — compare scores only within one fixed subset.
+Paired bootstrap over images puts the between-run standard error at **~0.45 overall**, so the top eight runs are statistically tied — every confidence interval against the leader straddles zero. Doubling the eval set from 500 to 1000 images did not separate them, and it reshuffled the order: one run climbed six places purely on noise. Only the gaps to the two reference models are real.
 
 ## Findings
 

index.html

@@ -49,7 +49,7 @@ a { color: #0969da; }
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Targets</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">all linear layers, vision + language towers</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Data</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">~6.3k images, 1 caption + 2 QA each (~18.9k samples)</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Caption prompts</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">sampled from a 30-prompt pool, so the model learns the task not one phrasing</td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Eval</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">fixed 500-image draw = 1000 graded items, same seed for every model</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Eval</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">fixed 1000-image draw = 2000 graded items, same seed for every model</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Scoring</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">LLM-as-judge, 0-100 per item</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">References</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">stock Qwen3-VL-8B, plus an in-house finetuned model</td></tr>
 </tbody></table>
@@ -82,7 +82,7 @@ a { color: #0969da; }
 </tbody></table>
 <p>Constant everywhere: effective batch 32, warmup 0.03, grad clip 0.3 (1.0 for 011/012), seed 3407.</p>
 <h2>Results</h2>
-<p>Best checkpoint per run. <strong>Overall</strong> = mean of caption and VQA. Right block = in-domain subset, closest to the target distribution.</p>
+<p>Best checkpoint per run, scored on 1000 images (2000 graded items). <strong>Overall</strong> = mean of caption and VQA. Right block = in-domain subset, closest to the target distribution.</p>
 <table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
 <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">Model</th>
 <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">Caption</th>
@@ -92,25 +92,25 @@ a { color: #0969da; }
 <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">In-dom. VQA</th>
 <th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">In-dom. overall</th>
 </tr></thead><tbody>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Qwen3-VL-8B (base)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">66.29</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">56.21</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>61.25</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">59.29</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>64.23</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in-house reference</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.14</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.57</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>69.36</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.90</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.47</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-001</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">77.60</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.64</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">84.32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.97</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-011</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.42</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">76.75</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.59</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">84.77</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>83.68</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-008</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.61</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.94</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.27</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.69</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.12</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.41</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-005</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.52</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.24</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.02</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.85</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-012</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.85</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">76.50</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.17</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.23</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">84.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>83.14</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-007</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.07</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">76.25</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.16</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.96</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.65</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.81</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-003</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.84</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">76.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>78.00</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.43</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.42</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.43</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-013</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.54</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.94</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.74</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.35</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.50</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.92</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-006</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.50</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.69</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.08</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.07</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-002</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.62</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.17</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.76</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-009</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.46</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">76.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.42</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.19</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.43</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.31</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-004</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.31</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.10</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-010</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.71</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.02</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.78</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.18</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Qwen3-VL-8B (base)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">56.01</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>60.84</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">58.94</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>63.66</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in-house reference</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>69.26</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.75</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-001</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.97</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.94</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.95</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.84</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.23</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.54</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-011</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.92</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.78</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.85</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.85</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">83.51</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.68</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-006</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.81</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.79</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.80</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.40</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.95</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.18</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-005</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.55</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.74</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.64</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.87</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-008</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.40</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.61</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.42</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.02</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-003</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.55</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.52</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.53</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.23</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.71</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-012</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.02</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.37</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.86</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-007</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.33</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>80.97</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-013</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.11</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.35</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.23</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.20</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.39</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-009</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.85</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.19</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.01</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.18</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-004</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.76</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.04</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.47</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>80.89</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-002</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.36</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.85</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.70</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.63</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>80.67</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-010</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.88</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>76.58</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.50</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.66</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>80.58</strong></td></tr>
 </tbody></table>
-<p>Every finetune beats both references: <strong>+17.4 overall</strong> over the base model, <strong>+9.3</strong> over the in-house reference. Captioning gains most (66 to 80); VQA gains 21 points over base.</p>
+<p>Every finetune beats both references: <strong>+17.1 overall</strong> over the base model, <strong>+8.7</strong> over the in-house reference. Captioning gains most (66 to 80); VQA gains 20 points over base.</p>
 <h2>Caveat</h2>
-<p>Paired bootstrap over images puts the between-run standard error at <strong>~0.6 overall</strong>, so the top five runs are statistically tied. A different 500-image draw moved the <em>same</em> model by 1.8 points — compare scores only within one fixed subset.</p>
+<p>Paired bootstrap over images puts the between-run standard error at <strong>~0.45 overall</strong>, so the top eight runs are statistically tied — every confidence interval against the leader straddles zero. Doubling the eval set from 500 to 1000 images did not separate them, and it reshuffled the order: one run climbed six places purely on noise. Only the gaps to the two reference models are real.</p>
 <h2>Findings</h2>
 <ul>
 <li><strong>Optimization is saturated.</strong> LR, rank, alpha, dropout, weight decay, clipping, batch geometry and schedule were all swept; none beat the noise floor.</li>