A Portable Evaluation Record (PER) turns one AI-agent evaluation into a versioned, portable account of what was observed, what its evidence supports, what remains unknown and why a release recommendation was made. EIO-Agents 0.8.0 produces, validates, verifies and explains PER 2.1.0 across evaluation frameworks.
PER is the JSON output format of EIO-Agents. The Evaluation Intelligence Ontology (EIO) gives claims, evidence, findings, scores and release views their shared meaning. An evaluation framework supplies an observed native bundle; EIO-Agents projects a PER from it. ProofAgent Harness is one possible producer, not a requirement. A PER cites the bundle but does not copy the whole raw transcript.
This is an illustrative display of the source-complete synthetic example in the EIO-Agents repository, not a customer run or a deployed ProofAgent dashboard. The validated sample PER 2.1.0 reports EIO 0.6.0, four claims, six evidence references and one finding.
The example finding concerns an invented authority or deadline on turn 3 and is marked UNPROVEN, severity unrated. The sample's release recommendation is REVIEW under release semantics 2.2: no release policy is declared, so readiness must reach 85 (observed 42.0), and 2 HARD_BLOCK obligations (forbidden tool, secret disclosure) were never exercised. BLOCK would require a proven failure. Of 7 release gates, 2 are met, 3 unmet and 2 not determinable. A human decides.
Inspect the synthetic source bundle. Its PER header pins the schema, ontology and source archive.
header and provenance pin versions, source identity and hashes. subject and scope identify the evaluated agent and context. evidence, claims and coverage connect decisions to observed source references and tested obligations. findings, controls and reliability are derived views. scores and release_recommendation report measured or withheld values and the separate decision. limitations and telemetry bound interpretation. The exact machine contract is the PER 2.1.0 JSON Schema.
A native producer records EIO evidence during the run; an export converter maps a finished report through an explicit crosswalk. Both build the same EIO bundle with build_bundle(), and EIO-Agents turns it into the same PER. The EIO-Agents repository includes illustrative converters with synthetic data:
None of these tools is required, and none emits a PER on its own today. A native producer can reach exact predicate fidelity; an export converter is often narrower and cannot reconstruct evidence missing from the export.
PER 2.1.0 is an open, versioned EIO-Agents specification maintained by ProofAgent / ProofAI LLC under Apache-2.0. Its schema uses JSON Schema Draft 2020-12 and EIO publishes a JSON-LD context. Other evaluation producers can adopt it. PER is not a W3C- or ISO-ratified standard, a compliance certification or a claim that all frameworks already support it.
Clone the EIO-Agents source, run the included synthetic bundle through eio-agents project, then use validate, verify --bundle and explain --list. Validation checks structure; source verification re-derives evidence, scores and the digest from the local bundle. Neither proves that the original run or human judgment was correct. Review a PER for sensitive information before sharing it.
For the meaning behind every field, read the EIO semantic layer; for versioned downloads and provisional framework mappings, use the EIO schema reference. Maintained by ProofAgent / ProofAI LLC · Apache-2.0 · support@proofagent.ai.
EIO-Agents developed by Dr. Fouad Bousetouane · © 2025–2026 ProofAI LLC · Licensed under the Apache License 2.0 · Source on GitHub · support@proofagent.ai