Hand your auditor evidence, not a log dump.
Regulators and auditors are starting to ask how AI-driven actions are recorded and overseen. Decision Horizon produces a per-action record they can read in plain language and verify themselves.
And where the record answers them
| Question | Where the record answers it |
|---|---|
| Can you show what the system knew when it acted? | The horizon block, with content hashes and what was accessible but not read. |
| Was a human in the loop? | policy.hitl: whether approval was required, who answered, and the hash of what they were shown. |
| Is the record complete and unaltered? | Hash chain, closing record and offline verification under a key you pinned. |
| Can you preserve it for a dispute? | Legal hold in the hosted store, with a hash-chained audit log. |
| What can't the record tell us? | coverage.uncovered: the stated blind spots of each insertion point. |
What you get
Readable replays
dh replay turns a chain into sentences or an HTML report an examiner can follow.
Frameworks you're mapped to
Supports evidence for record-keeping and oversight expectations such as the EU AI Act's logging requirements for high-risk systems, the NIST AI Risk Management Framework and the OWASP Top 10 for Agentic Applications.
Honest about limits
Every record says what it can't establish. Unknown is recorded as unknown.
Designed through ten adversarial review rounds; each of 124 findings was fixed or published as a stated limit.
Decision Horizon helps you produce evidence. It doesn't by itself make you compliant with any regulation.
See it on one of your own workflows.
We'll put Decision Horizon in front of a real agent workflow of yours and replay the record with you.