Traceability

Traceability is the unbroken chain from what was asked to what shipped. For AI-generated code that chain normally breaks the moment the agent session ends — the prompt is gone, and the diff is all that’s left. Provasign keeps the chain intact.

The chain

User prompt
   │  captured verbatim
   ▼
Intent YAML  (.provasign/intents/INT-….yaml)
   │  prompt_hash · agent · model · acceptance criteria
   ▼
ChangeSet  (the diff the agent produced)
   │  Grove ICR: which symbols, which blast radius
   ▼
Certificate  (Ed25519-signed: changeset · config · toolchain · results)
   │
   ▼
Admitted commit
   trailers:  Intent-ID · Intent-Hash · Certificate-ID · ICR-Hash · Signed-By

Every link is addressable. From a commit you can reach its certificate and its intent; from an intent you can find the prompt that created it; from the certificate you can replay the gates.

What each artifact pins down

Artifact Answers
Intent YAML What was asked? verbatim prompt, who/what produced it, acceptance criteria
Intent-Hash Was the recorded prompt tampered with?
ICR (from Grove) What did it actually change? symbols + blast radius
Certificate What verified it? gates, config, toolchain, results — signed
Commit trailers Where does it all hang together? stable, parseable links on the commit

Identity granularity

The intent records the agent and model that produced the change (e.g. claude-opus-4-8). Recording precise, version-pinned model identity — distinguishing, say, two Opus versions or a different vendor’s model — is what makes the chain useful for later analysis; capture it as vendor/family-version at intent-open time.

This is the artifact the EU AI Act expects for high-risk activities: a record of what the AI was asked to do, what checked the output, and a proof you can show an auditor.


This site uses Just the Docs, a documentation theme for Jekyll.