Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Gather independent ranking ballots from multiple judges or perspectives for a given candidate set.
.claude/skills/yogsoth-ai-ballot-collection/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -63% | 0% |
Collects independent ranking ballots from multiple judges or evaluation perspectives. Each judge produces a complete or partial ranking of the candidates without seeing other judges' rankings.
Runs as a subagent. Receives candidates and perspective definitions, returns structured ballots.
Each ballot must be generated independently to prevent anchoring. The subagent evaluates from a single perspective without access to other judges' outputs, ensuring genuine independence.
Output MUST contain one ballot per perspective. Each ballot MUST rank all candidates (complete ranking) or explicitly mark unranked candidates. No two ballots may be identical unless perspectives are genuinely indistinguishable.
<!-- 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-01 | fail→pass | 20,914 | 12,404 | -41% | 1 | 1 | 0% | 3,118 | 2,218 | -29% | 0 | 0 | — |
case-02 | pass→pass | 13,445 | 4,561 | -66% | 1 | 1 | 0% | 2,155 | 992 | -54% | 0 | 0 | — |
case-03 | pass→fail | 9,768 | 3,129 | -68% | 1 | 1 | 0% | 1,444 | 732 | -49% | 0 | 0 | — |
case-04 | fail→pass | 13,089 | 2,789 | -79% | 1 | 1 | 0% | 2,083 | 639 | -69% | 0 | 0 | — |
case-05 | pass→pass | 9,087 | 2,710 | -70% | 1 | 1 | 0% | 1,302 | 734 | -44% | 0 | 0 | — |
case-06 | pass→pass | 8,361 | 4,092 | -51% | 1 | 1 | 0% | 1,275 | 962 | -25% | 0 | 0 | — |
case-07 | pass→pass | 12,775 | 7,061 | -45% | 1 | 1 | 0% | 1,833 | 1,283 | -30% | 0 | 0 | — |
case-08 | pass→fail | 5,976 | 1,805 | -70% | 1 | 1 | 0% | 898 | 489 | -46% | 0 | 0 | — |
case-09 | fail→fail | 9,314 | 2,455 | -74% | 1 | 1 | 0% | 1,389 | 631 | -55% | 0 | 0 | — |
case-10 | pass→pass | 8,381 | 1,621 | -81% | 1 | 1 | 0% | 1,210 | 477 | -61% | 0 | 0 | — |
case-11 | fail→pass | 13,024 | 2,896 | -78% | 1 | 1 | 0% | 1,986 | 653 | -67% | 0 | 0 | — |
case-12 | fail→pass | 7,602 | 1,977 | -74% | 1 | 1 | 0% | 1,025 | 502 | -51% | 0 | 0 | — |
case-13 | fail→fail | 14,129 | 11,895 | -16% | 1 | 1 | 0% | 2,069 | 1,764 | -15% | 0 | 0 | — |
case-14 | pass→fail | 10,155 | 2,409 | -76% | 1 | 1 | 0% | 1,543 | 573 | -63% | 0 | 0 | — |
case-15 | fail→pass | 11,608 | 2,384 | -79% | 1 | 1 | 0% | 1,688 | 627 | -63% | 0 | 0 | — |
case-16 | pass→pass | 9,209 | 3,271 | -64% | 1 | 1 | 0% | 1,624 | 753 | -54% | 0 | 0 | — |
case-17 | fail→pass | 11,048 | 2,268 | -79% | 1 | 1 | 0% | 1,614 | 582 | -64% | 0 | 0 | — |
case-18 | pass→pass | 6,489 | 2,830 | -56% | 1 | 1 | 0% | 1,005 | 691 | -31% | 0 | 0 | — |
case-19 | fail→pass | 10,403 | 8,116 | -22% | 1 | 1 | 0% | 1,572 | 1,454 | -8% | 0 | 0 | — |
case-20 | pass→pass | 12,112 | 6,810 | -44% | 1 | 1 | 0% | 1,780 | 1,180 | -34% | 0 | 0 | — |
case-21 | pass→pass | 4,841 | 2,977 | -39% | 1 | 1 | 0% | 759 | 670 | -12% | 0 | 0 | — |
case-22 | pass→pass | 10,384 | 3,563 | -66% | 1 | 1 | 0% | 1,668 | 744 | -55% | 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. 22 cases were attempted. The headline lift of +18 percentage points is the difference between those two pass rates over the 22 comparable cases. 3 cases got worse with the skill loaded, and they are 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.