Agent
Independent auditor for an empirica run. Verifies the run against the ADR-20 rubric — re-reading each approved claim's research citation to confirm the cited source actually supports the claim — and writes a pass/fail verdict artifact. Spawn this before a run may report converged. The authoring agent must never write…
Agent
Fold-1 research worker for an empirica run. Resolves a needs-data claim by fetching and citing a real source outside the model's training data, then returns a structured research record. Use for the mechanical fetch-and-cite work of the empirical loop.
Agent
Fold-2 spike worker for an empirica run. Designs a real deterministic check for a needs-experiment claim, runs it through spikeharness.py so the verdict is a genuine exit code, and reports the result. Use when a claim needs runtime evidence rather than a citation.