Add eval artifacts: summary, per-case scores, significance tests, rubric stats, HTML report
Browse files- eval/rubric_stats.json +30 -0
eval/rubric_stats.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"n_prompts": 200,
|
| 3 |
+
"checks_total": 4983,
|
| 4 |
+
"checks_per_prompt": {
|
| 5 |
+
"min": 23,
|
| 6 |
+
"max": 25,
|
| 7 |
+
"mean": 24.92
|
| 8 |
+
},
|
| 9 |
+
"distribution": {
|
| 10 |
+
"23": 1,
|
| 11 |
+
"24": 15,
|
| 12 |
+
"25": 184
|
| 13 |
+
},
|
| 14 |
+
"by_dimension": {
|
| 15 |
+
"Alignment": 616,
|
| 16 |
+
"Layout": 842,
|
| 17 |
+
"Typography": 642,
|
| 18 |
+
"Components": 1517,
|
| 19 |
+
"Assets": 584,
|
| 20 |
+
"Aesthetics": 782
|
| 21 |
+
},
|
| 22 |
+
"by_track_surface": {
|
| 23 |
+
"A/dashboard": 1000,
|
| 24 |
+
"A/landing": 2486,
|
| 25 |
+
"B/dashboard": 748,
|
| 26 |
+
"B/landing": 749
|
| 27 |
+
},
|
| 28 |
+
"all_check_with_screenshot": true,
|
| 29 |
+
"note": "Additionally every prompt carries 5 judge-side Prompt-Fit rubrics scored 0-2 (surface-specific), not part of the 23-25 frozen screenshot checks."
|
| 30 |
+
}
|