Predictive Quality Gate Reference
Purpose: Predict likely Judge and Zen findings before review so Guardian can warn early without over-blocking.
Contents
- Judge prediction
- Zen prediction
- historical learning
- predictive report
- review-flow integration
- accuracy tracking
Judge Prediction
Typical Judge issue families:
- null handling
- race conditions
- logic bugs
- missing guards
- unsafe assumptions
Confidence bands:
- high:
>= 80% - medium:
60-79% - low:
< 60%
Warning rule:
- high confidence -> always warn
- medium confidence -> warn only if
risk_score > 50
Zen Prediction
Typical Zen issue families:
- generic names
- mixed naming conventions
- large functions
- magic numbers
- noisy formatting mixed with logic
Auto-fix rule:
- only for code-quality issues
- only when confidence
> 90%
Historical Learning
Adjust predictions with:
- accepted or rejected Judge findings
- accepted or rejected Zen suggestions
- project-specific false positive patterns
False positive handling:
- decrease pattern confidence by
5%
Predictive Report
## Predictive Quality Gate Analysis
### Predicted Judge Findings
- possible null handling issue (85%)
### Predicted Zen Suggestions
- split formatting noise from logic change (92%)Integration With Review Flow
Use predictive findings before review to:
- enrich Guardian's report
- decide whether Judge or Zen should be called early
- avoid unnecessary review churn
Pause when:
- predicted security issue is high confidence
- predicted logic bug is high confidence and high impact
- multiple predicted fixes conflict
Accuracy Tracking
Targets:
- Judge prediction accuracy
> 75% - Zen prediction accuracy
> 80% - false positive rate
< 20%
Canonical accuracy report:
## Prediction Accuracy Report (Last 30 Days)
### Judge Predictions
- accuracy: ...
### Zen Predictions
- accuracy: ...