Evidence and auditability¶
A log line says that something happened. Evidence lets you reconstruct what happened, verify that it was allowed, and defend the decision later.
Why logs are not enough¶
Logs are usually optimized for operators, not for proof. They may be incomplete, mutable, missing context, or disconnected from the policy that authorized the action. When a decision is challenged, “search the logs” is not an answer.
Evidence has four properties:
- Structured — it records the action, inputs, decision, result, and identity in a known shape.
- Linked — it connects the policy, the approval, the execution, and the outcome.
- Tamper-evident — it can be checked for alteration or omission.
- Retained — it survives the process that produced it.
Evidence in the ecosystem¶
| Evidence | Produced by | Used for |
|---|---|---|
| Effect ledger | PRAXOVELA | Knowing what an agent actually did |
| Execution trace | PRAXOVELA | Replay, diagnosis, and audit |
| Process history | RHEOVELA | Recovering and auditing long-running work |
| Verification result | TEKMOVELA | Release gates and re-verification |
| Simulation evidence | PEIRAVELA | Comparing possible worlds |
| Assertion and state history | ONTOVELA | Knowing what was true, and when |
| Authorization decision | AEGIVELA | Explaining allow/deny outcomes |
| Identity and lifecycle record | EIDOVELA, NOMIVELA | Attributing actions to an agent |
Evidence is not the same as observability¶
Observability asks “is the system healthy?” Evidence asks “can I prove what happened and why?” A system can be highly observable and still have no audit trail, and vice versa. Mature systems need both.
A practical test¶
Pick a high-risk action. Try to answer, using only artifacts:
- Which agent acted?
- Under which identity and authority?
- Which policy allowed it?
- What exactly was invoked?
- What was the result?
- Who approved it, if anyone?
- Can the action be reversed or recovered?
If any answer requires a human memory or a mutable dashboard, the evidence model is incomplete.