Escalating reasoning effort after repeated blocked attempts is recorded rather than assumed

What it means

A profile may declare effort_escalation: after a stated number of blocked attempts, effort rises to a stated level. Escalation is a declared, visible rule in the profile — not a habit of running everything at maximum — and it appears in the profile table every worker reads.

How it works

effort_escalation.after_blocked_attempts and effort_escalation.to are validated profile fields; debugging escalates to xhigh after two blocked attempts, deep-work after one. The projection renders them next to the base effort.

How to see it

grep -A2 '^effort_escalation' .ai/repo/profiles/debugging.yaml

What it does not cover

Advisory. v0.1 records the rule; it does not count attempts or observe the worker's effort. Making escalation an observed event needs telemetry that does not exist yet.

Why it exists

Maximum effort as the default was one of the named cost sources; the source environment's routing guidance said effort should be "set only where it differs from the session default", which is the sparse, deliberate form this field keeps.