Can an investigation be checked independently?
CipherLoop → TraceForge
Experimental agent workflow + independent evidence reader · Python
My implementation. I built an investigation workflow that records tool observations and failed execution, plus a separate reader that checks the saved record without importing the producer’s runtime.
Recorded result. The pinned Linux offline replay passes integrity checks for valid captures, reports unavailable scanner output as an error, and rejects altered ledger bytes.
Limits. Scripted tool responses; no live audit or detection-accuracy result. Integrity checks do not prove task success or target safety.