A competent operator escalates only when continuation would exceed authority or materially increase risk.
Persist / Evidence dossier
Blocker detection
The organization can distinguish a real authority, information, safety, or budget blocker from ordinary uncertainty.
Current answer
Implicit Contract Bench specifies escalation outcomes but has not produced results.
Why it matters
Over-escalation recreates a human manager; under-escalation causes hidden stalls or unsafe action.
Matched-human contract
Recall at least 95% of hard blockers while keeping false escalations below 10% of decisions.
Known failure
Agents often ask for permission on reversible actions or proceed past a true external constraint.
Claims
provisionalclaim-human-prompt-bottleneck
Routine human continuation prompts are a material constraint on current long-horizon Codex operation.
Evidence ledger
qualitative-observationuncontrolled
Operator-reported continuation bottleneck
The operator repeatedly observes that capable Codex threads stop for routine continuation and that he becomes the effective heartbeat. This motivates an evaluator; it is not a performance result.
Evaluator
protocol-readyimplicit-contract-bench
Implicit Contract Bench
An external deterministic ledger scores decisions, resource use, reports, hidden state, consequences, and authority violations. The evaluated agent cannot edit the ledger.
Inspect protocol