Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Synthesize multi-source evidence into structured argumentation. Weaves findings from literature, web, and analysis into coherent evidence maps with explicit strength ratings.
.claude/skills/yogsoth-ai-evidence-synthesis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-22 | ✓→✗ | ▼ Worse | -67% | 0% |
Weave multi-source evidence into structured argumentation.
Subagent — spawned via subagent-spawning/spawn-agent.
Synthesis requires processing large accumulated evidence corpus in dedicated context.
Quantity target is set by the calling strategy's budget table. This SOP executes one unit = one evidence synthesis pass producing a structured evidence map.
<!-- 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→fail | 10,187 | 28,307 | +178% | 1 | 1 | 0% | 1,589 | 3,857 | +143% | 0 | 0 | — |
case-02 | fail→fail | 17,282 | 13,417 | -22% | 1 | 1 | 0% | 2,576 | 2,196 | -15% | 0 | 0 | — |
case-03 | fail→fail | 6,232 | 9,731 | +56% | 1 | 1 | 0% | 952 | 1,724 | +81% | 0 | 0 | — |
case-04 | pass→pass | 11,226 | 3,450 | -69% | 1 | 1 | 0% | 1,627 | 713 | -56% | 0 | 0 | — |
case-05 | fail→fail | 8,072 | 1,825 | -77% | 1 | 1 | 0% | 1,144 | 441 | -61% | 0 | 0 | — |
case-06 | fail→pass | 8,669 | 1,711 | -80% | 1 | 1 | 0% | 1,278 | 418 | -67% | 0 | 0 | — |
case-07 | fail→fail | 10,108 | 1,846 | -82% | 1 | 1 | 0% | 1,581 | 391 | -75% | 0 | 0 | — |
case-08 | fail→pass | 14,672 | 5,155 | -65% | 1 | 1 | 0% | 2,218 | 939 | -58% | 0 | 0 | — |
case-09 | fail→pass | 16,969 | 7,912 | -53% | 1 | 1 | 0% | 2,451 | 1,334 | -46% | 0 | 0 | — |
case-10 | fail→fail | 17,572 | 23,814 | +36% | 1 | 1 | 0% | 2,388 | 3,831 | +60% | 0 | 0 | — |
case-11 | fail→fail | 7,929 | 4,477 | -44% | 1 | 1 | 0% | 1,150 | 624 | -46% | 0 | 0 | — |
case-12 | fail→pass | 2,092 | 1,185 | -43% | 1 | 1 | 0% | 216 | 323 | +50% | 0 | 0 | — |
case-13 | fail→fail | 18,975 | 15,482 | -18% | 1 | 1 | 0% | 2,940 | 2,472 | -16% | 0 | 0 | — |
case-14 | fail→fail | 14,453 | 15,587 | +8% | 1 | 1 | 0% | 2,299 | 2,385 | +4% | 0 | 0 | — |
case-15 | fail→fail | 18,067 | 28,355 | +57% | 1 | 1 | 0% | 2,622 | 4,432 | +69% | 0 | 0 | — |
case-16 | pass→pass | 13,710 | 6,088 | -56% | 1 | 1 | 0% | 2,103 | 1,053 | -50% | 0 | 0 | — |
case-17 | fail→fail | 19,871 | 14,031 | -29% | 1 | 1 | 0% | 2,872 | 2,172 | -24% | 0 | 0 | — |
case-18 | fail→fail | 12,792 | 24,186 | +89% | 1 | 1 | 0% | 1,761 | 3,624 | +106% | 0 | 0 | — |
case-19 | fail→fail | 19,649 | 19,465 | -1% | 1 | 1 | 0% | 2,634 | 2,693 | +2% | 0 | 0 | — |
case-20 | fail→fail | 1,620 | 1,952 | +20% | 1 | 1 | 0% | 239 | 416 | +74% | 0 | 0 | — |
case-21 | pass→pass | 16,018 | 10,077 | -37% | 1 | 1 | 0% | 2,043 | 1,642 | -20% | 0 | 0 | — |
case-22 | pass→fail | 8,977 | 8,075 | -10% | 1 | 1 | 0% | 1,670 | 546 | -67% | 0 | 0 | — |
case-23 | pass→pass | 3,968 | 3,897 | -2% | 1 | 1 | 0% | 701 | 910 | +30% | 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 +13 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.