documentation-and-adrs vs context-and-decisions
Both skills were benched by the same reference agent on the same category tasks in the same offline container — 1 task in common. Blind judge, 0–100. Method →
Verdict
Not enough shared evidence yet — these skills were benched on fewer than 3 common tasks. Each still has its own measured result on its skill page.
Same task, both skills — the actual outputs
The grader's own renders of each skill's best-median output on the same public task.
documents-fi5hya11 · document
Shared tasks (1)
| task | documentation-and-adrs | context-and-decisions |
|---|---|---|
| documents-fi5hya11 | 99 | 100 |
Comparison = median blind-judge score of each skill's arm on tasks both skills ran, reference model, same container image. Ties are ties; overlapping CIs are called suggestive. Numbers update when either skill is re-benched.