AIV Integrity GateFull stored-patch benchmark

Full stored-patch run, not a sample

AIV was executed on every stored patch file: 2,098 files, 2,080 unique pull requests. 1,219 of the files are Apache Airflow. 879 are from other projects.

2,098
stored patch files executed by AIV
113
blocked by a gate finding
1,803
passed. 182 could not be scored
1,219
Airflow patch files inside this run

What was executed

This is every prs/*.json file in the stored corpus: the Airflow year folder, the Airflow proof folder, and the other project folders. It is not the 555-case sample and not the 55-case fixture subset. 15,575 is a metadata census, not a skipped gate run. Those rows have titles/bodies/sizes only (no patch), so they were never part of this scoreboard. A block is a gate finding. It is not a label that the pull request is AI slop. Unscored is not a pass.

Of these files, 1,605 had been scored earlier. Those verdicts are the same, except two Airflow files that were previously unscored because the base file could not be fetched. Those two, pull requests 61939 and 69095, are now blocked for an AI-generated marker. The other 493 files are the Airflow year folder, scored in this run. Eighteen Airflow pull requests are stored in both Airflow folders. Each of those pairs has the same verdict.

Airflow stored patches

Airflow files include the labeled proof set and the extra year folder. Where the same pull request is stored in both folders, both files were executed, so the Airflow file count can be higher than the unique pull-request count.

LabelFilesBlockedPassedUnscoredErrors
AI disclosed3111029650
Merged control10849860
Closed, not merged1359108180
Objective marker11000
Stored patch66467507900

Airflow files: 91 blocked, 1,009 passed, 119 unscored, 0 errors.

Why a patch blocked

Assertion-free test 50 AI-generated marker in the diff 38 Placeholder text 15 Syntax error 7 Provenance trailer in the diff 1 Placeholder test 1 Merge-conflict marker 1

One primary rule per blocked file. A missing-tests finding is not counted as the reason when a harder finding is on the same gate.