Pattern Miner Across Runs

Output Requirements

• Start with the top 3 to 5 cross-run patterns.
• For each pattern, include recurrence, representative evidence, and expected impact.
• Explicitly separate stable patterns from noisy or uncertain ones.
• If the run set is too small, say so.

Guardrails

• Do not force unrelated findings into the same pattern bucket.
• Distinguish task variation from model variation when possible.
• Preserve uncertainty when the sample size is weak.40:["$","$L49",null,{"content":"$4a","frontMatter":{"name":"pattern-miner-across-runs","description":"Mine repeated patterns across multiple analysis runs. Use when comparing many videos or reruns to find recurring strengths, waste patterns, risk signals, instability, variability, 横断分析, 傾向抽出, or common improvement themes across work-video artifacts.","argument-hint":"Provide the selected result directories, the work type if known, and whether you want repeated inefficiencies, good practices, risk signals, or variability across providers and reruns."}}]