Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Sensitivity Analysis Campaign — identify which assumptions are most critical by measuring their impact on conclusions. 5 strategies (parameter-screening, variance-decomposition, assumption-criticality, uncertainty-propagation, decision-sensitivity), 3 tactics, 11 subagent SOPs.
.claude/skills/yogsoth-ai-deep-insight-sensitivity-analysis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 145% | 0% |
| case-20 | ✓→✗ | ▼ Worse | 184% | 0% |
| case-21 | ✓→✗ | ▼ Worse | -6% | 0% |
Identify which assumptions are most critical — rank by impact on conclusions.
This campaign is a strategy book — CC reads, internalizes, and autonomously constructs an approach.
| Signal | Strategy | |--------|----------| | quick screening, Morris method, OAT, preliminary elimination | → parameter-screening | | variance decomposition, Sobol indices, contribution, interaction effects | → variance-decomposition | | assumption criticality, perturbation, negation, re-derivation | → assumption-criticality | | uncertainty propagation, Monte Carlo, distributions, Bayesian | → uncertainty-propagation | | decision sensitivity, EVPI, influence diagrams, tornado diagrams | → decision-sensitivity |
Import (5): web-search, web-research, paper-overview, paper-search, paper-research
Subagent (11): morris-screening, sobol-decomposition, interaction-detection, assumption-extraction, negation-definition, re-derivation, conclusion-sensitivity-measurement, distribution-assignment, monte-carlo-sampling, critical-path-identification, sensitivity-synthesis
Shared (1): assumption-surfacing
context-checkpoint after each strategy completes.
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Strategy | When to use | | --- | --- | | assumption-criticality | Measure how much conclusions change when each assumption is negated. Ranks assumptions by their impact on the final result. | | decision-sensitivity | Identify which uncertainties would actually change the research direction decision. Compute EVPI to prioritize uncertainty reduction. | | parameter-screening | Quick Morris method screening to identify which parameters have large effects and which can be safely ignored. | | uncertainty-propagation | Propagate input uncertainties through the model via Monte Carlo sampling. Identifies which input uncertainties contribute most to output uncertainty. | | variance-decomposition | Sobol variance decomposition — compute first-order and total-order sensitivity indices to quantify each parameter's contribution to output variance. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | context-checkpoint | Append research process and results to the current Phase's context file. Covers both process and results with genuine substance. Use this skill at plan-designated checkpoint points — typically after each strategy completes or at key decision nodes within a research Phase. | | context-init | Create a new context file for a research Phase. Called once at Phase start to initialize the file that subsequent context-checkpoint calls will append to. Use this skill whenever a new research Phase begins and a fresh context file is needed. |
<!-- 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 | 21,121 | 28,020 | +33% | 1 | 1 | 0% | 3,631 | 5,427 | +49% | 0 | 0 | — |
case-02 | pass→pass | 16,964 | 10,196 | -40% | 1 | 1 | 0% | 2,603 | 2,355 | -10% | 0 | 0 | — |
case-03 | pass→pass | 12,583 | 13,602 | +8% | 1 | 1 | 0% | 2,266 | 3,060 | +35% | 0 | 0 | — |
case-04 | fail→pass | 32,816 | 20,588 | -37% | 1 | 1 | 0% | 3,050 | 4,008 | +31% | 0 | 0 | — |
case-05 | pass→pass | 15,311 | 7,234 | -53% | 1 | 1 | 0% | 2,480 | 1,896 | -24% | 0 | 0 | — |
case-06 | pass→pass | 29,098 | 15,430 | -47% | 1 | 1 | 0% | 2,666 | 3,300 | +24% | 0 | 0 | — |
case-07 | pass→pass | 15,534 | 5,433 | -65% | 1 | 1 | 0% | 1,380 | 1,417 | +3% | 0 | 0 | — |
case-08 | pass→pass | 9,868 | 3,610 | -63% | 1 | 1 | 0% | 1,505 | 1,303 | -13% | 0 | 0 | — |
case-09 | fail→fail | 12,067 | 3,379 | -72% | 1 | 1 | 0% | 1,828 | 1,198 | -34% | 0 | 0 | — |
case-10 | fail→pass | 5,787 | 2,238 | -61% | 1 | 1 | 0% | 882 | 984 | +12% | 0 | 0 | — |
case-11 | pass→pass | 17,572 | 23,203 | +32% | 1 | 1 | 0% | 2,863 | 4,424 | +55% | 0 | 0 | — |
case-12 | pass→pass | 6,619 | 3,903 | -41% | 1 | 1 | 0% | 995 | 1,259 | +27% | 0 | 0 | — |
case-13 | pass→pass | 13,508 | 4,555 | -66% | 1 | 1 | 0% | 2,119 | 1,343 | -37% | 0 | 0 | — |
case-14 | pass→pass | 6,204 | 5,017 | -19% | 1 | 1 | 0% | 965 | 1,548 | +60% | 0 | 0 | — |
case-15 | pass→pass | 9,802 | 3,440 | -65% | 1 | 1 | 0% | 1,458 | 1,233 | -15% | 0 | 0 | — |
case-16 | pass→pass | 12,846 | 5,911 | -54% | 1 | 1 | 0% | 2,032 | 1,752 | -14% | 0 | 0 | — |
case-17 | fail→pass | 7,472 | 12,423 | +66% | 1 | 1 | 0% | 1,215 | 2,978 | +145% | 0 | 0 | — |
case-18 | pass→pass | 11,238 | 3,185 | -72% | 1 | 1 | 0% | 1,814 | 1,215 | -33% | 0 | 0 | — |
case-19 | pass→pass | 9,973 | 2,262 | -77% | 1 | 1 | 0% | 1,532 | 1,038 | -32% | 0 | 0 | — |
case-20 | pass→fail | 12,378 | 43,413 | +251% | 1 | 1 | 0% | 2,414 | 6,848 | +184% | 0 | 0 | — |
case-21 | pass→fail | 22,888 | 21,855 | -5% | 1 | 1 | 0% | 4,543 | 4,288 | -6% | 0 | 0 | — |
case-22 | pass→pass | 12,002 | 17,347 | +45% | 1 | 1 | 0% | 2,265 | 4,076 | +80% | 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 +5 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 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.