Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Multi-axis systematic perturbation — define variation axes, perturb along each, measure degradation, construct validity envelope.
.claude/skills/yogsoth-ai-systematic-perturbation/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 130% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 171% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 252% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 77% | 0% |
Map validity by perturbing along defined axes.
Subagent: variation-axis-definition, controlled-perturbation, validity-envelope-construction Import: paper-search
Define orthogonal axes of variation (at least 3), perturb one axis at a time while holding others constant, record performance at each point, construct envelope from degradation thresholds.
<HARD-GATE>
- variation axes defined: >= 3
- perturbation curves: >= 3
- validity envelope: >= 1 constructed
</HARD-GATE><!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | controlled-perturbation | Systematically vary parameters along defined axes, recording performance at each point to identify degradation thresholds. | | 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-validity-envelope-construction | Combine multi-axis perturbation data into a multi-dimensional validity description with boundary conditions and interaction effects. | | variation-axis-definition | Identify orthogonal axes along which a method's validity might vary. Ensures axes are independent, measurable, and span the relevant parameter space. |
<!-- 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 | pass→pass | 23,224 | 29,684 | +28% | 1 | 1 | 0% | 2,791 | 4,930 | +77% | 0 | 0 | — |
case-02 | fail→pass | 20,736 | 32,317 | +56% | 1 | 1 | 0% | 2,402 | 5,526 | +130% | 0 | 0 | — |
case-03 | pass→pass | 16,071 | 29,063 | +81% | 1 | 1 | 0% | 2,549 | 5,367 | +111% | 0 | 0 | — |
case-04 | pass→pass | 11,270 | 18,194 | +61% | 1 | 1 | 0% | 1,727 | 2,609 | +51% | 0 | 0 | — |
case-05 | fail→pass | 12,291 | 28,935 | +135% | 1 | 1 | 0% | 1,851 | 5,009 | +171% | 0 | 0 | — |
case-06 | pass→pass | 20,918 | 18,477 | -12% | 1 | 1 | 0% | 2,459 | 2,834 | +15% | 0 | 0 | — |
case-07 | pass→pass | 20,565 | 29,743 | +45% | 1 | 1 | 0% | 2,336 | 4,606 | +97% | 0 | 0 | — |
case-08 | fail→fail | 19,848 | 16,388 | -17% | 1 | 1 | 0% | 2,126 | 2,297 | +8% | 0 | 0 | — |
case-09 | pass→pass | 21,673 | 51,200 | +136% | 1 | 1 | 0% | 2,554 | 8,580 | +236% | 0 | 0 | — |
case-10 | fail→pass | 16,503 | 39,692 | +141% | 1 | 1 | 0% | 1,446 | 5,083 | +252% | 0 | 0 | — |
case-11 | pass→pass | 21,415 | 39,955 | +87% | 1 | 1 | 0% | 2,753 | 5,183 | +88% | 0 | 0 | — |
case-20 | pass→pass | 18,733 | 39,945 | +113% | 1 | 1 | 0% | 2,008 | 6,856 | +241% | 0 | 0 | — |
case-12 | fail→fail | 29,604 | 7,874 | -73% | 1 | 1 | 0% | 1,484 | 891 | -40% | 0 | 0 | — |
case-13 | pass→pass | 17,786 | 8,930 | -50% | 1 | 1 | 0% | 2,146 | 1,183 | -45% | 0 | 0 | — |
case-14 | pass→pass | 9,938 | 9,290 | -7% | 1 | 1 | 0% | 1,444 | 1,069 | -26% | 0 | 0 | — |
case-15 | pass→pass | 19,291 | 9,606 | -50% | 1 | 1 | 0% | 2,246 | 1,100 | -51% | 0 | 0 | — |
case-16 | pass→pass | 18,804 | 27,194 | +45% | 1 | 1 | 0% | 2,055 | 4,279 | +108% | 0 | 0 | — |
case-17 | fail→pass | 32,005 | 8,220 | -74% | 1 | 1 | 0% | 978 | 976 | -0% | 0 | 0 | — |
case-18 | pass→pass | 19,859 | 34,709 | +75% | 1 | 1 | 0% | 2,388 | 5,652 | +137% | 0 | 0 | — |
case-19 | pass→pass | 27,481 | 34,065 | +24% | 1 | 1 | 0% | 2,708 | 5,568 | +106% | 0 | 0 | — |
case-21 | pass→pass | 13,130 | 37,598 | +186% | 1 | 1 | 0% | 2,997 | 7,297 | +143% | 0 | 0 | — |
case-22 | pass→pass | 11,772 | 36,537 | +210% | 1 | 1 | 0% | 1,371 | 7,168 | +423% | 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 +18 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.