Skip to content

PatchProofProve the test. Not just the patch.

Deterministic counterfactual testing for pull requests—run the changed regression test against the real code before and after the fix.

The missing CI check

A passing regression test is not enough.

Ordinary CI tells you the branch passes. PatchProof answers the harder review question: would this exact test have caught the bug before the fix existed?

1Find the changed test

Diff-aware adapters identify added and materially changed pytest, Jest, and Vitest cases.

2Run the counterfactual

The test-side patch is transplanted onto an isolated worktree of the real base revision—never the production fix.

3Issue a conservative verdict

Base assertion failure plus head pass can become proven; uncertainty stays inconclusive.

Evidence you can inspect

One command. A review artifact with receipts.

$ patchproof check --base origin/main --head HEAD
PatchProof 0.1.0-alpha.3
Base: 41e8c91b0f21
Head: a02f7d43c8aa
PROVEN tests/orders.test.ts::rejects negative quantity
base: assertion_failure
head: pass
Aggregate: PROVEN

Designed for trust

Useful in a portfolio. Serious enough for a pipeline.

3 frameworks

pytest, Jest, and Vitest with case-aware selection and conservative fallback.

3 operating systems

Continuously exercised on Linux, macOS, and Windows with Node.js 22 and 24.

0 hosted dependency

No telemetry, API key, required AI model, or external proof service.

Built for honest automation

Proof stays narrow—and that is the point.

PatchProof does not claim the whole patch is correct. It verifies one valuable fact with reproducible evidence: the submitted test distinguishes the fixed implementation from its base. Opt-in reason checks can also require the base failure to match the intended assertion exactly.

Explore the public historical benchmarks, read the security model, or see how PatchProof compares with mutation testing and coverage.

Make “the test passes” mean more.

Inspect first, approve the commands, then produce deterministic evidence locally or in GitHub Actions.

Run your first proof →

Deterministic regression-test evidence. No telemetry or required AI.