Review the space between pull requests.
A passing change can still violate an assumption introduced by another change.
Select and pin
Sync real open GitHub PRs, then select two to ten changes. Exact base/head revisions and repository policy become an immutable analysis snapshot. Moving revisions invalidate old results.
Rank with evidence
The analyzer combines shared files and shared changed identifiers, then compares supported Python function signatures and dictionary return keys. Coverage and source references accompany each hypothesis.
Interpret the outcome
Suspected means static evidence needs review. Reproduced requires passing baseline and individual states, then the same assertion failure twice in combination. Not reproduced means the chosen combined test passed. Missing runner, dependencies, merge conflicts or invalid tests remain inconclusive.