Incident memory, from the outage to every pull request after it.

The model drafts one test. A fresh sandbox runs it three times on the commit before the fix and three times on the fix. Fails, then passes: proven. Anything else is redrafted with the failure as feedback, up to three times, or the incident stays unproven with the reason.

Only the memory tests whose files a change touches run, plus any the model’s triage adds; triage can add tests but never remove them. Each check also shows how close the change comes to code behind past incidents, even when no test runs. Merge queues are checked on the queue’s own commit.

With a proven test, Aftershock bisects your history to the commit, and pull request, that introduced the bug. Over time that shows which kinds of change cause incidents.

Every week the fix is undone on today’s code and each test must fail again; tests that would no longer notice are flagged Weak. Mutation testing makes small changes to the fixed lines and measures how many a test catches, so a loose test can be hardened with a second proven one.

A new alert or issue that looks like a past incident, from the same files or a similar error, is linked to it, with one click to re-run that incident’s test on main. When a merged pull request changes the files an open alert points at, it is suggested as the fix.

Proof rate, time to proof, recurrences blocked, test health and strength, and the files behind the most incidents. Share a read-only reliability report with leadership by secret link, and ask questions in plain language. Answers come only from your incidents and always cite them.