-
Notifications
You must be signed in to change notification settings - Fork 1
Pipeline Success Pattern Analyzer — Extract and Apply Winning Patterns #646
Open
Description
Strategic Improvement
We obsessively analyze failures but ignore what makes pipelines SUCCEED. Build an analyzer that extracts patterns from successful runs (repo characteristics, template choices, stage timings, model routing decisions) and applies those patterns to improve future runs. Success rate is 77% (target 85%) — learning from the 77% is as important as fixing the 23%.
Acceptance Criteria
- Collect success metrics from pipeline events (template, repo type, stage durations, model choices, final cost)
- Extract common patterns from top-quartile successful runs (fastest, cheapest, or both)
- Generate actionable recommendations: "Repos with <10 files succeed 92% with fast template" or "Test stage completes 3x faster with haiku on projects with <50 test files"
- Store patterns in intelligence cache for pipeline composer and template recommender to consume
- Add
shipwright success-patternscommand showing learned patterns with confidence scores - Dashboard widget showing "success pattern match score" for incoming issues
Context
- Priority: P0
- Complexity: standard
- Generated by: Strategic Intelligence Agent
- Strategy alignment: P0: Reliability & Success Rate