LAB · CONTROLLED · REPRODUCED RUN

Reference Harness project trust probe · reference-project-trust-v1-run-001

This is a controlled experiment artifact for Reference Harness. It demonstrates the public runner pipeline, not any vendor agent's native behavior.

reference-project-trust-v1-run-0018 eventsclean

TRACE 0.1 · CONTROLLED

A formal experiment run, event by event

trust.decision

Actor project-trust produced sequence 0. Redaction status is clean.

{
  "workspace_fingerprint": "6be1df31dc7c708006e9221b4d76d2864ca81a12331dff39a38fabb66f5b07aa",
  "trusted": true,
  "policy_version": "trust-policy-v1",
  "reason": "Explicit grant from explicit-lesson-user matches this workspace."
}

1 events are visible; hidden chain-of-thought is not part of the trace.

Download JSONL

HARNESS MODEcontrolled

MODELscripted-model-v1

NETWORKoff

COMPARABILITYnot-comparable

WHAT THIS PROVES

A fixed fixture and prompt can produce a structured result, valid trace, identical normalized fingerprint, and downloadable replay through the public runner.

WHAT IT DOES NOT PROVE

Real model capability, vendor agent behavior, OS sandbox strength, performance cost, or cross-agent ranking.

View full experiment record →