derive cleaner min-4px YOLO corpus

docs/CODEX_EXECUTION_LOG.md

@@ -12638,3 +12638,36 @@ Open:
   not independent object-level accuracy. A future benchmark must aggregate or
   deduplicate objects before uncertainty claims.
 - No source label, frozen dataset, model weight or live runtime was changed.
+
+## 2026-08-09 - Immutable min-4px derived corpus
+
+### Improved
+
+- Added a deterministic derived-corpus builder that preserves parent hashes,
+  copies imagery, records every removed source row and refuses to turn a
+  positive tile into an unlabeled negative. It creates no release manifest or
+  human-review claim.
+- Derived `yolo-building-aoi1024-reviewedexp6-minpx4-derived-r1` from the
+  frozen min-3px corpus. Exactly 1,812 marginal labels were removed; all 252
+  tiles and 77,380 labels remain, with zero excluded tiles.
+- Corrected the quality-audit console message from `passed` to `completed`, so
+  a machine-readable `needs_attention` status is no longer contradicted by
+  human-facing output.
+
+### Verified on Tower
+
+- Structural audit: zero missing/invalid labels and zero sub-4-pixel rows.
+- Relationship audit: zero exact/near duplicates and 41 possible nesting
+  pairs, down from 43.
+- Geometry audit: 10 plausible extreme-aspect rows, down from 26; tile-edge
+  rows fell from 5,079 to 4,840 without a blanket edge deletion.
+- Cross-tile audit: zero reconstructed train/validation crossings.
+- Inspected four full contact-sheet pages covering all 252 tiles and 77,380
+  labels plus the focused nine-tile extreme-aspect sheet. Pure backgrounds
+  remain empty and sparse contexts remain distinct.
+
+### Decision and limitation
+
+- The min-4px corpus is preferred for future experimental work, but remains
+  explicitly training/release-ineligible until real human review and governed
+  release evidence exist. The active GPU model was not changed.

docs/TODO.md

@@ -1164,9 +1164,12 @@ This file now starts with the current implementation status. Older preparation/b
   sheets and retain immutable hashes.
 - [x] Reconstruct exact cross-tile interior objects and prove that overlap
   repeats 40,031 label rows but causes zero reconstructed train/val crossings.
-- [ ] Build the next immutable experimental corpus with `min_label_px >= 4`
-  and compare the existing 0.35 edge-visibility policy against a separately
-  versioned stricter ablation before any new governed training.
+- [x] Build and fully audit an immutable experimental corpus with
+  `min_label_px=4`; retain 252 tiles and 77,380 labels with zero sub-4-pixel
+  rows, missing files, invalid labels or cross-split reconstructed objects.
+- [ ] Compare the existing 0.35 edge-visibility policy against a separately
+  versioned stricter ablation from original source geometry before any new
+  governed training; do not approximate this by dropping every edge label.
 - [ ] Convert the AI-assisted ledger into no stronger claim than experimental
   triage; a real human must independently review and sign the frozen artifacts
   before the governed training wrapper may unlock.

docs/reviews/2026-08-09-ai-assisted-checkpoint-review.md

@@ -161,3 +161,29 @@ The edge sample shows expected clipped buildings under the 0.35 minimum-visible
 policy. Raising that value may reduce partial-target pressure, but it must be a
 separately versioned ablation because removing all 5,079 rows without checking
 the original visible fraction would be unsound.
+
+## Derived min-4px experimental corpus
+
+The row-level finding was converted into a new immutable derived corpus rather
+than changing the historical dataset. The derivation is checksum-bound to the
+79,192-label source summary and removes only labels whose smallest 512-tile
+dimension is below 4 pixels. It retains all 252 tiles and 77,380 labels; no
+positive tile became empty, and no source file was overwritten.
+
+Fresh audits report zero sub-4-pixel labels, zero invalid or missing label
+files, zero exact/high-IoU duplicates and zero reconstructed cross-split
+object groups. Extreme-aspect labels fall from 26 to 10, possible nesting from
+43 to 41 and tile-edge rows from 5,079 to 4,840. The complete four-page visual
+render was inspected again. Pure-background tiles remain visibly empty, sparse
+contexts remain distinct and the ten remaining elongated labels predominantly
+match plausible long structures.
+
+The generic quality auditor still reports `needs_attention` because normalized
+median box area `0.000641` is below its conservative `0.001` warning threshold.
+That corresponds to median dimensions near 12 by 13 pixels at source tile
+resolution and is not a failed minimum-dimension gate. The warning remains
+visible; it was not suppressed or relabelled as success.
+
+This min-4px version is the preferred experimental successor to the legacy
+min-3px corpus. It explicitly remains ineligible for governed training until
+real human review and all release-contract evidence exist.