Clarify paired Real1 environment-style failure in block diagnostic cohort
Browse files
s1_blocks_seq343_real2_real1palette_headrepaired40_20261009/QUALITY_NOTICE.md
CHANGED
|
@@ -10,3 +10,9 @@ are conditioned on paired Real1 and pass the full-frame color/context audit.
|
|
| 10 |
Visual non-regression was reviewed on sampled frames, not certified on every
|
| 11 |
frame. `training_release=false` remains intentional. Do not treat this
|
| 12 |
diagnostic upload as fully human-reviewed training data.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 10 |
Visual non-regression was reviewed on sampled frames, not certified on every
|
| 11 |
frame. `training_release=false` remains intentional. Do not treat this
|
| 12 |
diagnostic upload as fully human-reviewed training data.
|
| 13 |
+
|
| 14 |
+
Subsequent direct paired-Real1 environment-style review found that at least
|
| 15 |
+
case `stack_blocks_ep000000` fails (GPT-6 Luna confidence 0.98): the Real2
|
| 16 |
+
table is olive/mottled while Real1's is pale and neutral. The earlier 40/40
|
| 17 |
+
acceptance did not include this style criterion. This prefix is **not
|
| 18 |
+
Real1-style-approved** and must not be used as final Real2 training data.
|