Sutura · Verified self-healing CI

Sutura Case Lab

AI agents make CI green. Sutura proves whether they fixed the problem. Pick one of five fixed cases and read the full verified result: diagnosis, search tree, candidate patches, rejected patches, audit verdict, cost, and links to the GitHub evidence.

How Sutura verifies a repair: the pipeline from the failed run to the audited patch, the runtime roles, and the five outcomes a run can end in.

A live run starts one real Sutura repair in the public demo repository when the public-demo gate is on: at most 1 run at a time, 4 per hour, and 24 per day. Every case has a deterministic result you can open now.

Release v0.3.1 · Action e724f3b22de79d6ab3f40cffa96de7776c256ce9

JavaScript repair

A real off-by-one bug fails one test. Sutura must repair the code, not the test.

Recorded live result Fixed

Expected: Fixed · Language: javascript

Open recorded live result

Python repair

A Python coroutine is never awaited. Sutura must repair it inside the pinned Python runtime.

Recorded live result Gave up

Expected: Fixed · Language: python

Open recorded live result

Deterministic flaky failure

A timing race fails two runs in five. Sutura must classify the flake and refuse to invent a patch.

Recorded live result Flaky, no patch

Expected: Flaky, no patch · Language: javascript

Open recorded live result

Greenwash trap

A fake fix only changes the expected value. Green CI is not enough: the recorded run refuses the fake fix, and a live run must repair the boundary without touching the test.

Recorded live result Refused

Expected: Refused · Live: Fixed without touching a test · Language: javascript

Open recorded live result

Upstream dependency incident

A CommonJS service breaks on an ESM-only release. Sutura must ground the diagnosis before it repairs.

Recorded live result Fixed

Expected: Fixed · Language: javascript

Open recorded live result

How results are labeled

Live run
This result was produced by a live run through the public Case Lab path.
Deterministic replay
This result was reproduced offline from a complete replay bundle captured on a live run, with no provider or sandbox access.
Recorded live result
This result is a recorded live benchmark evaluation of the released Sutura version, not a run started from this page.

The Case Lab accepts only these five server-defined cases. It never accepts arbitrary repositories, refs, commands, patches, or text.