AI Operator Evidence Ledger

A narrow control layer that checks whether an AI operator's "done" claim is backed by proof now, and rechecks whether it stays connected to a next action after a review window passes.

Each case is enriched with an AI step (Google Gemini API when configured, deterministic local fallback otherwise) that classifies the claim, suggests evidence types, and writes a plain-language summary. The "source" label on each card shows which path produced it.

Synthetic demo. These cases are illustrative examples of operating failure modes, not real work logs, customer data, or credentials.

claim-happy-readme-tests
claimed by OpsAgent · risk: low · window: 24h
evidence-backed

“Task complete. README updated, tests passed, deploy ready.”

artifact_exists PASS artifact present at happy-readme.md
artifact_content_match PASS artifact contains expected marker 'evidence-marker: setup-section-added'
test_pass_recorded PASS recorded test result: passed (readme-smoke)
decision_link_present PASS decision linked: decision-local-green-001
next_action_visible PASS next action visible: next-record-demo-video
recency_within_window PASS claim age 0.00h is within the 24h review window
AI enrichmentsource: Gemini API
classification: task_completion_summary
All claimed actions were successfully verified.
Suggested evidence to strengthen this claim:
  • artifact_exists
  • artifact_content_match
  • test_pass_recorded
  • decision_link_present
  • next_action_visible
  • recency_within_window
decision: decision-local-green-001 · next action: next-record-demo-video · last checked: 2026-07-05T05:25:31.276Z
claim-missing-artifact
claimed by BuildAgent · risk: high · window: 24h
rejected (missing proof)

“Done. Build artifact created and deployment is ready.”

artifact_exists FAIL artifact not found at missing-build.tar
artifact_content_match FAIL cannot match content: artifact missing
test_pass_recorded FAIL no recorded test result alongside the artifact
decision_link_present PASS decision linked: decision-review-required-014
Repair action: Re-open work and supply valid proof for: artifact_exists, artifact_content_match, test_pass_recorded. Done state is withheld until proof is attached.
AI enrichmentsource: local fallback
classification: build_artifact
Claim is missing required evidence: the referenced artifact is not present, the artifact does not contain the expected proof marker, no recorded passing test result.
Suggested evidence to strengthen this claim:
  • next_action_visible
  • recency_within_window
decision: decision-review-required-014 · next action: next-repair-missing-build-proof · last checked: 2026-07-05T05:25:31.278Z
claim-stale-next-action
claimed by RunAgent · risk: medium · window: 6h
evidence-backed

“Done. Review complete and next action queued.”

artifact_exists PASS artifact present at stale-followthrough.md
artifact_content_match PASS no expected content marker specified; treated as satisfied
decision_link_present PASS decision linked: decision-time-axis-watch-006
next_action_visible PASS next action visible: next-follow-through-current
recency_within_window PASS claim age 0.00h is within the 6h review window
AI enrichmentsource: Gemini API
classification: task_completion_and_handoff
No failure occurred as the claim was fully backed by evidence.
Suggested evidence to strengthen this claim:
  • artifact_exists
  • artifact_content_match
  • decision_link_present
  • next_action_visible
  • recency_within_window
decision: decision-time-axis-watch-006 · next action: next-follow-through-current · last checked: 2026-07-05T05:25:31.279Z