Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Multi-source cross-validation of gap authenticity — cross-database search, temporal sensitivity testing, false-gap filtering, stakeholder confirmation.
.claude/skills/yogsoth-ai-cross-validation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 7% | 0% |
Multi-source verification of gap authenticity.
Subagent: cross-database-verification, temporal-sensitivity-testing, false-gap-filtering, stakeholder-confirmation Import: web-search, paper-search
For each gap candidate: verify across 3+ databases (Semantic Scholar, Google Scholar, arXiv, domain-specific), test if gap persists across 2/5/10 year windows, apply false-gap heuristics (wrong search terms? already solved? inherently unanswerable?), confirm with stakeholder simulation.
<HARD-GATE>
- cross-database checks: >= 3 per gap
- temporal windows tested: >= 2 per gap
- false-gap filter applied: >= 1 per gap
</HARD-GATE><!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | cross-database-verification | Verify gap existence across multiple databases (Semantic Scholar, Google Scholar, arXiv, domain-specific). Distinguishes database-specific gaps from universal gaps. | | deep-insight-paper-search | AI-powered paper summary and search. Import of literature-engine/literature-search skill. AI summary level — cite as "AI-extracted" not "paper states". | | deep-insight-web-search | Quick web scanning for landscape understanding. Import of web-browsing/web-search skill. Snippets only — no conclusions from snippets alone. | | false-gap-filtering | Detect false gaps — search failures, already-solved gaps, and inherently unanswerable questions masquerading as research gaps. | | stakeholder-confirmation | Simulate stakeholder perspectives to validate gap priorities. Assesses gap value from researcher, practitioner, funder, and end-user viewpoints. | | temporal-sensitivity-testing | Test whether a gap persists across different time windows (2/5/10 years). Determines if gap is narrowing, widening, or stable over time. |
<!-- 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-14 | pass→pass | 21,878 | 12,591 | -42% | 1 | 1 | 0% | 3,731 | 2,651 | -29% | 0 | 0 | — |
case-02 | fail→fail | 8,543 | 10,634 | +24% | 1 | 1 | 0% | 1,524 | 864 | -43% | 0 | 0 | — |
case-08 | pass→pass | 10,951 | 14,711 | +34% | 1 | 1 | 0% | 1,653 | 2,896 | +75% | 0 | 0 | — |
case-09 | pass→pass | 10,594 | 13,740 | +30% | 1 | 1 | 0% | 1,623 | 2,791 | +72% | 0 | 0 | — |
case-03 | pass→pass | 3,120 | 4,545 | +46% | 1 | 1 | 0% | 506 | 1,359 | +169% | 0 | 0 | — |
case-01 | pass→pass | 11,243 | 12,673 | +13% | 1 | 1 | 0% | 2,166 | 2,914 | +35% | 0 | 0 | — |
case-04 | pass→pass | 15,025 | 13,501 | -10% | 1 | 1 | 0% | 2,598 | 2,657 | +2% | 0 | 0 | — |
case-05 | pass→pass | 18,771 | 14,818 | -21% | 1 | 1 | 0% | 2,186 | 2,709 | +24% | 0 | 0 | — |
case-06 | pass→pass | 11,854 | 9,950 | -16% | 1 | 1 | 0% | 1,748 | 2,081 | +19% | 0 | 0 | — |
case-07 | pass→pass | 7,351 | 4,725 | -36% | 1 | 1 | 0% | 1,095 | 1,372 | +25% | 0 | 0 | — |
case-10 | pass→pass | 12,364 | 16,699 | +35% | 1 | 1 | 0% | 1,777 | 2,377 | +34% | 0 | 0 | — |
case-11 | pass→pass | 10,056 | 9,627 | -4% | 1 | 1 | 0% | 1,630 | 1,945 | +19% | 0 | 0 | — |
case-12 | pass→pass | 11,753 | 9,387 | -20% | 1 | 1 | 0% | 1,693 | 1,926 | +14% | 0 | 0 | — |
case-13 | pass→pass | 7,868 | 5,919 | -25% | 1 | 1 | 0% | 1,101 | 1,379 | +25% | 0 | 0 | — |
case-15 | fail→pass | 25,579 | 2,723 | -89% | 1 | 1 | 0% | 1,216 | 949 | -22% | 0 | 0 | — |
case-16 | fail→pass | 11,748 | 15,521 | +32% | 1 | 1 | 0% | 1,697 | 2,866 | +69% | 0 | 0 | — |
case-17 | fail→pass | 16,659 | 16,228 | -3% | 1 | 1 | 0% | 2,466 | 2,956 | +20% | 0 | 0 | — |
case-18 | fail→pass | 16,156 | 18,874 | +17% | 1 | 1 | 0% | 2,307 | 3,316 | +44% | 0 | 0 | — |
case-19 | pass→pass | 10,721 | 3,809 | -64% | 1 | 1 | 0% | 1,777 | 1,092 | -39% | 0 | 0 | — |
case-20 | pass→pass | 19,267 | 17,248 | -10% | 1 | 1 | 0% | 3,031 | 3,221 | +6% | 0 | 0 | — |
case-21 | fail→pass | 19,292 | 16,602 | -14% | 1 | 1 | 0% | 2,963 | 3,184 | +7% | 0 | 0 | — |
case-22 | pass→pass | 14,212 | 9,352 | -34% | 1 | 1 | 0% | 1,896 | 1,859 | -2% | 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, and 20 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +23 percentage points is the difference between those two pass rates over the 20 comparable cases.
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.