Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Select portfolios that perform well across multiple future scenarios using Minimax regret, Robust optimization, Scenario planning, and Info-gap methods.
.claude/skills/yogsoth-ai-robustness-under-uncertainty/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 307% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -4% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 164% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 289% | 0% |
Select a portfolio that performs acceptably well across a range of plausible futures, rather than optimizing for a single expected scenario. Prioritizes resilience over peak performance.
| Dimension | Target | |-----------|--------| | Candidates evaluated | 8-20 | | Scenarios constructed | >=3 distinct futures | | Performance metrics | 2-4 per scenario | | Robustness threshold | acceptable in all scenarios |
| Field | Type | Description | |-------|------|-------------| | candidates | list | All candidates with scenario-dependent performance | | scenarios | list | Distinct future scenarios | | performance_matrix | matrix | Candidate performance per scenario | | regret_matrix | matrix | Regret vs best-in-scenario for each candidate | | robust_portfolio | list | Portfolio minimizing worst-case regret |
| Tactic | When | |--------|------| | scenario-stress-testing | Core tactic — evaluate across scenarios | | pareto-frontier-construction | Trade off robustness vs expected value |
| SOP | Purpose | |-----|---------| | scenario-construction | Build distinct future scenarios | | portfolio-evaluation-per-scenario | Evaluate portfolio in each scenario | | portfolio-synthesis | Synthesize robust recommendation | | objective-definition | Define robustness criteria | | optimization-run | Find minimax-regret or robust solutions |
yamlstrategy: robustness-under-uncertainty selected_portfolio: - candidate: <name> worst_case_performance: <value> best_case_performance: <value> robustness_score: <0-1> max_regret: <value> vulnerable_scenarios: - scenario: <name> performance: <value> gap_to_best: <value> method_used: <minimax-regret|robust-optimization|info-gap>
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | pareto-frontier-construction | Build the Pareto frontier from multi-objective optimization, visualize trade-offs, and select a portfolio from non-dominated solutions. | | scenario-stress-testing | Construct distinct future scenarios, evaluate portfolio performance under each, and identify vulnerabilities and robustness characteristics. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | objective-definition | Define optimization objectives, constraints, and trade-off preferences from context and candidate information. | | optimization-run | Execute multi-objective optimization on candidates to produce a Pareto front of non-dominated solutions. | | portfolio-evaluation-per-scenario | Evaluate a specific portfolio's performance metrics and vulnerabilities under a given scenario. | | portfolio-synthesis | Synthesize all per-scenario evaluations into a final portfolio recommendation with robustness score and actionable guidance. | | scenario-construction | Construct distinct future scenarios spanning key uncertainties for portfolio stress testing. |
<!-- 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-06 | pass→fail | 10,099 | 43,109 | +327% | 1 | 1 | 0% | 1,742 | 9,105 | +423% | 0 | 0 | — |
case-01 | fail→pass | 13,203 | 33,291 | +152% | 1 | 1 | 0% | 1,630 | 6,635 | +307% | 0 | 0 | — |
case-02 | fail→fail | 20,004 | 30,713 | +54% | 1 | 1 | 0% | 3,822 | 9,138 | +139% | 0 | 0 | — |
case-03 | pass→pass | 18,083 | 27,065 | +50% | 1 | 1 | 0% | 2,848 | 5,655 | +99% | 0 | 0 | — |
case-04 | pass→pass | 38,642 | 44,626 | +15% | 1 | 1 | 0% | 8,259 | 8,670 | +5% | 0 | 0 | — |
case-05 | fail→fail | 18,356 | 45,445 | +148% | 1 | 1 | 0% | 2,243 | 9,108 | +306% | 0 | 0 | — |
case-07 | fail→fail | 27,501 | 43,595 | +59% | 1 | 1 | 0% | 4,446 | 9,105 | +105% | 0 | 0 | — |
case-08 | pass→pass | 40,916 | 35,752 | -13% | 1 | 1 | 0% | 7,999 | 7,162 | -10% | 0 | 0 | — |
case-09 | fail→pass | 38,157 | 32,516 | -15% | 1 | 1 | 0% | 8,261 | 7,926 | -4% | 0 | 0 | — |
case-10 | fail→pass | 17,629 | 36,278 | +106% | 1 | 1 | 0% | 2,797 | 7,394 | +164% | 0 | 0 | — |
case-11 | pass→fail | 29,875 | 46,249 | +55% | 1 | 1 | 0% | 5,018 | 9,101 | +81% | 0 | 0 | — |
case-12 | fail→pass | 24,644 | 28,465 | +16% | 1 | 1 | 0% | 4,504 | 5,833 | +30% | 0 | 0 | — |
case-13 | fail→pass | 16,503 | 38,039 | +130% | 1 | 1 | 0% | 1,965 | 7,638 | +289% | 0 | 0 | — |
case-14 | pass→pass | 19,801 | 29,698 | +50% | 1 | 1 | 0% | 2,716 | 5,894 | +117% | 0 | 0 | — |
case-15 | fail→pass | 16,999 | 41,833 | +146% | 1 | 1 | 0% | 2,015 | 8,731 | +333% | 0 | 0 | — |
case-16 | fail→fail | 19,119 | 41,763 | +118% | 1 | 1 | 0% | 2,367 | 9,099 | +284% | 0 | 0 | — |
case-17 | fail→pass | 13,579 | 34,396 | +153% | 1 | 1 | 0% | 2,372 | 6,698 | +182% | 0 | 0 | — |
case-18 | pass→pass | 22,192 | 32,332 | +46% | 1 | 1 | 0% | 3,217 | 6,170 | +92% | 0 | 0 | — |
case-19 | fail→pass | 23,037 | 29,293 | +27% | 1 | 1 | 0% | 2,934 | 6,832 | +133% | 0 | 0 | — |
case-20 | pass→fail | 19,755 | 18,858 | -5% | 1 | 1 | 0% | 2,830 | 3,481 | +23% | 0 | 0 | — |
case-21 | pass→fail | 20,684 | 43,548 | +111% | 1 | 1 | 0% | 3,512 | 8,724 | +148% | 0 | 0 | — |
case-22 | pass→pass | 21,632 | 25,391 | +17% | 1 | 1 | 0% | 3,935 | 5,502 | +40% | 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 +6 percentage points is the difference between those two pass rates over the 22 comparable cases. 7 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.