eval::threshold_adapter¶

Classes¶

	Name
class	eval::threshold_adapter::ThresholdAdapter
class	eval::threshold_adapter::ThresholdRecommendation

Attributes¶

	Name
	logger
str	THRESHOLD_EVALUATION_PROMPT

Detailed Description¶

LLM-based adaptive gate threshold recommendations.

Uses the Phase 1 ``TeacherClient`` to prompt an LLM with structured evaluation
metrics and prior-run context, asking it to recommend updated gate thresholds
with justification.

Auto-apply gate (T-02-10): Recommendations are only applied automatically when
the LLM's confidence exceeds ``confidence_minimum`` AND every recommended
change is less than ``max_adjustment_percent`` of the current threshold.
Otherwise the recommendation is logged for human review only.

Attributes Documentation¶

variable logger¶

logger =  logging.getLogger(__name__);

variable THRESHOLD_EVALUATION_PROMPT¶

str THRESHOLD_EVALUATION_PROMPT;

Updated on 2026-10-09 at 03:59:43 +0000