Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Two-phase sensitivity — Morris quick screening to eliminate unimportant factors, then Sobol precise decomposition on survivors. Efficient allocation of analytical effort.
.claude/skills/yogsoth-ai-screening-then-decomposition/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 14% | 0% |
Two-phase approach: quick screening followed by precise decomposition.
morris-screening → sobol-decomposition → interaction-detection
Subagent: morris-screening, sobol-decomposition, interaction-detection Import: paper-search
Phase 1 (Morris): Screen all parameters, eliminate those with low mu and low sigma. These are unimportant and can be safely fixed at nominal values.
Phase 2 (Sobol): For surviving parameters, compute precise variance decomposition. First-order indices (Si) measure direct effects, total-order indices (STi) measure direct + all interactions.
Phase 3: Detect significant interaction pairs (STi - Si > threshold). Characterize interaction types and implications.
<HARD-GATE>
- Parameters screened: >= 5
- Parameters eliminated (low mu*): >= 1
- Sobol indices computed: >= 3 parameters
- Interaction pairs identified: >= 1
</HARD-GATE><!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | 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". | | interaction-detection | Detect and characterize significant parameter interactions from Sobol decomposition results. | | morris-screening | Morris method screening — compute elementary effects to quickly identify important vs unimportant parameters. | | sobol-decomposition | Sobol variance decomposition — compute first-order and total-order sensitivity indices for precise variance attribution. |
<!-- 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 | 23,447 | 39,884 | +70% | 1 | 1 | 0% | 3,026 | 7,130 | +136% | 0 | 0 | — |
case-02 | pass→pass | 12,540 | 38,567 | +208% | 1 | 1 | 0% | 2,138 | 6,952 | +225% | 0 | 0 | — |
case-03 | pass→pass | 14,061 | 14,945 | +6% | 1 | 1 | 0% | 1,593 | 2,153 | +35% | 0 | 0 | — |
case-04 | pass→pass | 18,088 | 41,844 | +131% | 1 | 1 | 0% | 2,490 | 7,584 | +205% | 0 | 0 | — |
case-05 | pass→pass | 22,503 | 40,701 | +81% | 1 | 1 | 0% | 2,979 | 7,498 | +152% | 0 | 0 | — |
case-06 | fail→pass | 17,165 | 5,542 | -68% | 1 | 1 | 0% | 1,897 | 1,331 | -30% | 0 | 0 | — |
case-07 | fail→pass | 18,035 | 9,691 | -46% | 1 | 1 | 0% | 2,017 | 1,152 | -43% | 0 | 0 | — |
case-08 | fail→pass | 15,845 | 7,978 | -50% | 1 | 1 | 0% | 2,521 | 1,667 | -34% | 0 | 0 | — |
case-09 | fail→pass | 16,313 | 10,811 | -34% | 1 | 1 | 0% | 1,752 | 1,313 | -25% | 0 | 0 | — |
case-10 | fail→pass | 10,743 | 14,489 | +35% | 1 | 1 | 0% | 1,730 | 1,979 | +14% | 0 | 0 | — |
case-11 | pass→pass | 19,158 | 18,139 | -5% | 1 | 1 | 0% | 2,224 | 3,843 | +73% | 0 | 0 | — |
case-12 | pass→pass | 14,214 | 10,810 | -24% | 1 | 1 | 0% | 1,597 | 2,430 | +52% | 0 | 0 | — |
case-13 | pass→pass | 12,200 | 39,656 | +225% | 1 | 1 | 0% | 1,311 | 8,171 | +523% | 0 | 0 | — |
case-14 | pass→pass | 22,638 | 26,680 | +18% | 1 | 1 | 0% | 2,985 | 4,744 | +59% | 0 | 0 | — |
case-15 | pass→pass | 14,296 | 12,692 | -11% | 1 | 1 | 0% | 1,502 | 1,674 | +11% | 0 | 0 | — |
case-16 | pass→pass | 17,829 | 34,517 | +94% | 1 | 1 | 0% | 2,799 | 5,872 | +110% | 0 | 0 | — |
case-17 | pass→pass | 15,589 | 17,355 | +11% | 1 | 1 | 0% | 1,829 | 2,968 | +62% | 0 | 0 | — |
case-18 | pass→pass | 20,789 | 45,086 | +117% | 1 | 1 | 0% | 2,625 | 7,802 | +197% | 0 | 0 | — |
case-19 | pass→fail | 22,930 | 40,526 | +77% | 1 | 1 | 0% | 3,236 | 7,605 | +135% | 0 | 0 | — |
case-20 | pass→fail | 17,017 | 43,304 | +154% | 1 | 1 | 0% | 3,145 | 8,637 | +175% | 0 | 0 | — |
case-21 | pass→fail | 22,990 | 42,164 | +83% | 1 | 1 | 0% | 3,555 | 8,641 | +143% | 0 | 0 | — |
case-22 | fail→pass | 21,687 | 8,293 | -62% | 1 | 1 | 0% | 2,578 | 1,138 | -56% | 0 | 0 | — |
case-23 | fail→pass | 17,183 | 9,717 | -43% | 1 | 1 | 0% | 1,835 | 1,163 | -37% | 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 +17 percentage points is the difference between those two pass rates over the 23 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.