| name | priority-rebalancer |
| description | Adjust exploration priorities based on accumulated evidence from rollouts. Updates genome weights, toolchain routing preferences, and bead priorities based on observed success rates and pattern analysis. Use when working on sleeptime tasks. |
| metadata | [object Object] |
Priority Rebalancer
Adjust exploration priorities based on accumulated evidence from rollouts.
Inputs
| Parameter | Type | Required | Default | Description |
|---|---|---|---|---|
patterns |
object | Yes | - | Output from pattern-distiller |
current_config |
object | No | - | Current routing/priority config |
adjustment_strength |
number | No | 0.3 | How aggressively to adjust (0-1, higher = more change) |
min_evidence |
integer | No | 5 | Minimum pattern occurrences before adjusting |
Outputs
| Field | Type | Description |
|---|---|---|
recommended_adjustments |
array | Suggested config changes: - target (routing/genome/priority) - current_value - recommended_value - evidence (pattern_ids supporting change) - confidence |
exploration_suggestions |
array | Areas that need more exploration (low sample count) |
exploitation_candidates |
array | High-confidence patterns to exploit more |
Usage
python scripts/priority-rebalancer.py [arguments]
Examples
Rebalance based on pattern analysis
Inputs:
adjustment_strength: 0.3
patterns:
anti_patterns:
- '...'
patterns:
- '...'
Outputs:
recommended_adjustments:
- confidence: 0.78
current_value: codex
evidence:
- pattern_tc_003
- pattern_tc_007
rationale: Claude shows 85% success on high-risk tasks vs Codex 62%
recommended_value: claude
target: routing.risk.high.toolchain
- confidence: 0.65
current_value: 0.7
evidence:
- anti_pattern_001
rationale: Lower temperature reduces oscillation in repair loops
recommended_value: 0.5
target: genome.repair_prompt.temperature
Generated from `skills/sleeptime/priority-rebalancer.yaml`