Evidence levels and honesty rules

Every event, receipt, score and page in Proofroom declares its evidence level and its limitations. This page defines the vocabulary.

Evidence levels

self_reported

The agent's operating stack submitted the event. No independent system has confirmed the underlying action took place. This is the default level and it is honest about being the weakest.

system_confirmed

An external system reference was verified to exist (for example, a GitHub pull request referenced as github:pr:owner/repo#42 was confirmed via the GitHub API). Confirmation covers the reference's existence and metadata, not the quality or correctness of the action.

operator_confirmed

A named human resolved an approval for this action through the approval inbox or Telegram. It attests that the action was authorised and occurred, not that it was correct.

What a chain_valid badge means

Every event hash is a SHA-256 digest over the event's canonical fields including the previous event's hash. Verification recomputes every link from the stored rows. chain_valid means the stored history has not been altered since it was written. It does not mean the history is complete or that the events are true: completeness depends on what was submitted.

What we never claim

Proofroom never states that an agent is safe, accurate, unbiased, compliant or certified. Framework references (ISO 42001, SIG, CAIQ, NIST) indicate topical mapping only. Our own marketing is run through the same claims linter that governs our internal agents.