Expand description
Policy-oriented attempt orchestration.
The four independent axes are explicit: AttemptBody runs work,
AttemptJudge evaluates it, CarryPolicy prepares the next attempt, and
StopPolicy bounds the loop. Runtime health and quality verdict are kept
as separate fields in AttemptOutcome.
Structs§
- Attempt
Body Context - Attempt
Loop - Attempt
Outcome - Attempt
Request - Carry
Context - Continue
Session - Default carry: stable session + unchanged goal + feedback as context.
- Fresh
With Digest - Fresh
With Feedback - Explicit isolation: a new session and goal-appended feedback after attempt 1.
- Hybrid
Judge - Judge
Context - Judge
Result - LlmEval
Judge - Prepared
Attempt - Runtime
Attempt Body - Adapts
RuntimeRunnerto the body policy slot. - Stop
Policy - Verdict
FnJudge