config-drift-checker · komo-stack · model × Claude Code matrix

◆Safe on 3 of 4 cells

Every cell is one run of the suite on that model and Claude Code version, compared with the reference cell: sonnet on Claude Code 2.1.296. A case regresses when it drops more than 0.15 below the reference.

  • sonnet: safe on 2.1.296 to 2.1.310
  • haiku: safe on 2.1.296; fails spring-service-owns-rules-and-errors on 2.1.310
reference
sonnet on Claude Code 2.1.296
grid
2 × 2 versions × models, 6 cases, 1 run each
spend
$0.73 of $2.5 budget
pass score
1.00 regression past 0.15
generated
2026-10-28 08:42 UTC

Every case on every cell

caseClaude Code 2.1.296Claude Code 2.1.310
sonnet refhaikusonnethaiku
passing6/66/66/65/6
cost$0.33$0.03$0.35$0.03
mean turns2.52.82.52.8
runnerofficialofficialofficialofficial
guard-blocks-destructive-git1.001.001.001.00
spring-controller-follows-conventions1.001.001.001.00
spring-request-dtos-are-validated-records1.001.001.001.00
spring-service-owns-rules-and-errors1.001.001.000.80
spring-work-triggers-skill1.001.001.001.00
vue-request-does-not-trigger-skill1.001.001.001.00

passesimprovedbelow 1.00, same as the referenceregressed or failingnot runreference cell

Scores are the mean over a case's runs with the setup loaded. Hover a cell for its delta against the reference; a model name opens that cell's full report. Generated by config-drift-checker drift-matrix.