Jens/geointel
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

audit YOLO geometry and overlapping validation rows
GeoIntel release gates / Compile, test, contracts and builds (push) Canceled after 0s
Details
GeoIntel release gates / Python and npm vulnerability policy (push) Canceled after 0s
Details
GeoIntel release gates / GIS image, SBOM and container scan (push) Canceled after 0s
Details

Browse Source
This commit is contained in:
Jens
2026-08-09 19:36:15 +02:00
parent fc18e72c7f
commit 736e773fb8
13 changed files with 789 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files
artifacts/evidence/accuracy/model-training/20260809-reviewedexp6-ai-assisted-review-ledger.json
+32 -1
View File
@@ -1,7 +1,7 @@
{
"schema_version": 1,
"ledger_id": "reviewedexp6-ai-assisted-review-r2-full-corpus",
"reviewed_at": "2026-08-09T19:10:57+02:00",
"reviewed_at": "2026-08-09T19:34:31+02:00",
"reviewer": {
"id": "openai-codex",
"type": "ai_assistant",
@@ -96,5 +96,36 @@
"share_of_rendered_labels": 0.000543,
"ai_visual_followup": "all 86 implicated label rows were index-highlighted; no cyan exact duplicates or red near duplicates exist, while the 43 magenta nesting pairs are dispersed across dense reference AOIs and predominantly show larger GRB envelopes around smaller building components rather than one systematic duplicate-label pattern",
"action": "retain for experimental analysis, require human adjudication before release, and perform no automatic rewrite or exclusion"
},
"geometric_outlier_audit": {
"audit_path": "/app/storage/operator-data/model-review/reviewedexp6-corpus/full-ai-review-r1/label_outlier_audit.json",
"audit_sha256": "cac2c0cefb49aa3785ee9fb173a0187b3dd8abfc9c25e8e91df6c4af91ff5a40",
"reviewed_label_count": 79192,
"unique_flagged_label_count": 6657,
"small_dimension_below_4_px": 1812,
"extreme_aspect_ratio_at_least_8": 26,
"tile_edge_rows": 5079,
"extreme_rows_also_below_4_px": 16,
"extreme_contact_sheet_sha256": "67ad11e7d80678c78157a01a5e7ceb7cff9ac24410b127bdfaafd4350170cdb6",
"small_dimension_contact_sheet_sha256s": [
"5b3ca4d5bfd57b466854a00b4791ac82cf4a826fad8a4694d0227c931b00dfad",
"0e5d2058154f8298fc6d2c064150bcf896e4e51b9bcce5fccbfc049e1ebb9515",
"03060f65843f3c2a291dd009124484f4cf798fe653e658bab007375613fa3010",
"2fe7c748a68b21a673f3d950909b07e3ddba81b736977fbee8d64ea62b422ca2"
],
"tile_edge_sample_contact_sheet_sha256": "5d96c5a0a732077ed16b22d33bed998b37bd33b5536b1491e778999a2a6638dd",
"ai_visual_followup": "extreme rows are mostly plausible elongated structures, while sub-4-pixel rows are often visually marginal; tile-edge rows are expected clipped objects under the 0.35 visibility policy",
"next_corpus_decision": "restore the exporter minimum dimension to at least 4 px for the next immutable experimental corpus; retain elongated structures above that floor and measure any higher edge-visibility threshold in a separately versioned ablation"
},
"cross_tile_repetition_audit": {
"audit_path": "/app/storage/operator-data/model-review/reviewedexp6-corpus/full-ai-review-r1/cross_tile_repetition_audit.json",
"audit_sha256": "3cdf5174fec61e0aa426b4a3f44ed943a738bf038f3b58028ec856ef1a9d000c",
"interior_label_rows": 74113,
"unique_interior_objects": 34082,
"extra_repeated_rows": 40031,
"repeated_object_groups": 24701,
"max_repetition": 4,
"cross_split_repetition_groups": 0,
"interpretation": "the 512 px tiles use stride 256, so tile-level validation rows are not statistically independent even though no reconstructed object crosses train and validation splits"
}
}