Expand description
Evolution feedback collection and evaluation system (EVO-1).
Structs§
- Core
Metrics - Decision
Input - Input for recording a decision. The agent converts its ExecutionFeedback to this.
- Judgement
Summary - Rule
History Entry - Tool
Exec Detail
Enums§
- Evolution
Judgement - Feedback
Signal - User feedback signal for the last decision.
Functions§
- build_
latest_ judgement - compute_
core_ metrics_ for_ date - compute_
effectiveness - compute_
egl - System-wide EGL based on global metrics for the current day.
- compute_
egl_ for_ rule - EGL (Evolutionary Grade Level) captures a single, combined score of “goodness” of a given rule or agent behavior. EGL = (first_success_rate * A) - (avg_replans * B) - (user_correction_rate * C) where A, B, C are configurable weights. A higher EGL indicates better performance.
- compute_
tool_ sequence_ key - Build a compact tool-sequence key from tools_detail (at most 3 tools joined by →). Used to group decisions by “what tool pattern was used” rather than raw task description. Example: [weather] → “weather”; [http-request, write_output] → “http-request→write_output”.
- count_
decisions_ with_ task_ desc - Diagnostic: count unprocessed decisions with/without task_description. Evolution requires task_description to learn from decisions.
- count_
unprocessed_ decisions - ensure_
evolution_ tables - fetch_
latest_ metrics - insert_
decision - log_
evolution_ event - open_
evolution_ db - query_
rule_ history - update_
daily_ metrics - update_
last_ decision_ feedback