flight-recorder
Record every nondeterministic input your code reads (LLM answers, HTTP, database, clock, randomness) as one JSONL tape per request, then replay the tape against the real code offline, bit for bit, with the first divergence named. Open format with Python, Node, .NET, Go, Java…
- from
- Testing
- added
- 2026-10-10
- likes
- 0
similar
-
flight-recorder github.com
-
OrcaReplay github.com
Records a coding-agent session below the harness, keeping the verbatim request and response bytes, then replays the same run offline with the network off or forks it from a checkpoint onto other models with a verify command deciding the verdict.
-
-
playback github.com
-
-
LogTape logtape.org
Testing › Software > AI & LLM Testing: “Record every nondeterministic input your code reads (LLM answers, HTTP, database, clock, randomness) as one JSONL tape per request, then replay the tape against the real code offline, bit for bit, with the first divergence named. Open format with Python, Node, .NET, Go, Java…”