The RETROSPECTIVE identified a bimodal pattern: avg_lines=748 but p50=310 with one outlier at 5,468 lines. The p90/p95/stddev partially capture this, but what's missing is a direct measure of how concentrated the churn is.
Proposed metric: Top-20% concentration ratio:
concentration_ratio = (sum of lines in largest 20% of commits) / total_lines
A ratio above 0.80 means 20% of commits account for 80% of all churn — a classic heavy-tail pattern that warrants investigation even when the percentage thresholds look acceptable. This is computable from data already collected, requires no new git calls.
The RETROSPECTIVE identified a bimodal pattern: avg_lines=748 but p50=310 with one outlier at 5,468 lines. The p90/p95/stddev partially capture this, but what's missing is a direct measure of how concentrated the churn is.
Proposed metric: Top-20% concentration ratio:
concentration_ratio = (sum of lines in largest 20% of commits) / total_lines
A ratio above 0.80 means 20% of commits account for 80% of all churn — a classic heavy-tail pattern that warrants investigation even when the percentage thresholds look acceptable. This is computable from data already collected, requires no new git calls.