Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Select the next comparison pairs that maximize information gain given current ratings and comparison history.
.claude/skills/yogsoth-ai-pair-selector/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 15% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 198% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -59% | 0% |
Selects the next set of comparison pairs that would most reduce ranking uncertainty. Uses information-theoretic criteria (maximum entropy reduction, uncertainty sampling, or boundary proximity) to prioritize which pairs to compare next.
Runs as a subagent. Receives current ratings and comparison history, returns an ordered list of recommended next pairs.
Pair selection requires reasoning about the full rating landscape and comparison graph structure. Isolating this as a subagent allows focused computation without polluting the orchestrator's context with matrix calculations.
Output MUST contain at least one pair. Each pair must reference exactly two distinct candidates that exist in the current_ratings input.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | spawn-agent | Spawn a customized CC subagent with full MCP tool access. Used by SOPs that declare execution: subagent. |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | fail→pass | 44,004 | 31,896 | -28% | 1 | 1 | 0% | 4,947 | 4,991 | +1% | 0 | 0 | — |
case-10 | pass→pass | 11,203 | 9,495 | -15% | 1 | 1 | 0% | 1,700 | 1,759 | +3% | 0 | 0 | — |
case-01 | fail→pass | 21,709 | 38,126 | +76% | 1 | 1 | 0% | 2,648 | 3,052 | +15% | 0 | 0 | — |
case-02 | fail→pass | 12,998 | 36,902 | +184% | 1 | 1 | 0% | 1,289 | 3,840 | +198% | 0 | 0 | — |
case-03 | pass→pass | 46,266 | 85,662 | +85% | 1 | 1 | 0% | 2,207 | 4,218 | +91% | 0 | 0 | — |
case-05 | pass→pass | 24,871 | 48,840 | +96% | 1 | 1 | 0% | 999 | 1,605 | +61% | 0 | 0 | — |
case-06 | fail→pass | 42,820 | 26,478 | -38% | 1 | 1 | 0% | 3,484 | 1,521 | -56% | 0 | 0 | — |
case-07 | fail→pass | 29,791 | 29,155 | -2% | 1 | 1 | 0% | 2,002 | 816 | -59% | 0 | 0 | — |
case-08 | fail→pass | 43,865 | 12,607 | -71% | 1 | 1 | 0% | 2,205 | 1,465 | -34% | 0 | 0 | — |
case-09 | pass→pass | 24,116 | 15,842 | -34% | 1 | 1 | 0% | 2,793 | 2,343 | -16% | 0 | 0 | — |
case-11 | pass→pass | 10,988 | 3,165 | -71% | 1 | 1 | 0% | 602 | 707 | +17% | 0 | 0 | — |
case-12 | fail→pass | 18,545 | 17,020 | -8% | 1 | 1 | 0% | 2,237 | 1,937 | -13% | 0 | 0 | — |
case-13 | fail→pass | 13,482 | 8,113 | -40% | 1 | 1 | 0% | 1,373 | 740 | -46% | 0 | 0 | — |
case-14 | fail→pass | 17,683 | 15,396 | -13% | 1 | 1 | 0% | 2,214 | 2,208 | -0% | 0 | 0 | — |
case-15 | pass→pass | 10,980 | 11,341 | +3% | 1 | 1 | 0% | 1,101 | 1,419 | +29% | 0 | 0 | — |
case-16 | pass→pass | 19,899 | 22,848 | +15% | 1 | 1 | 0% | 3,583 | 3,126 | -13% | 0 | 0 | — |
case-17 | fail→pass | 14,353 | 6,972 | -51% | 1 | 1 | 0% | 1,507 | 529 | -65% | 0 | 0 | — |
case-18 | pass→pass | 13,783 | 10,584 | -23% | 1 | 1 | 0% | 1,153 | 1,037 | -10% | 0 | 0 | — |
case-19 | pass→pass | 18,992 | 23,442 | +23% | 1 | 1 | 0% | 2,819 | 2,871 | +2% | 0 | 0 | — |
case-20 | pass→pass | 15,199 | 15,065 | -1% | 1 | 1 | 0% | 2,319 | 2,194 | -5% | 0 | 0 | — |
case-21 | pass→fail | 23,878 | 37,191 | +56% | 1 | 1 | 0% | 4,373 | 8,456 | +93% | 0 | 0 | — |
case-22 | pass→pass | 30,046 | 22,725 | -24% | 1 | 1 | 0% | 5,239 | 5,179 | -1% | 0 | 0 | — |
case-23 | pass→pass | 9,710 | 9,012 | -7% | 1 | 1 | 0% | 1,053 | 1,012 | -4% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.