Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Strategy: Pearl Three-Step counterfactual — Abduction (fit model to evidence), Action (intervene on factor), Prediction (derive counterfactual outcome).
.claude/skills/yogsoth-ai-structural-counterfactual/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 146% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -8% | 0% |
Pearl Three-Step: Abduction, Action, Prediction applied to artifact validation.
| Parameter | S | M | L | |---|---|---|---| | Factors modeled | 5 | 10 | 20 | | Interventions per factor | 1 | 2 | 4 | | Prediction depth | 2 | 4 | 8 |
causal-claim-extraction → factor-enumeration
→ [for each factor]:
single-factor-removal (abduction + action)
→ counterfactual-scenario-construction (prediction)
→ necessity-evaluation + sufficiency-evaluation
→ load-bearing-identification → fragility-measurement<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| Tactic | When to use | | --- | --- | | causal-necessity-testing | Tactic: Extract causal claims, evaluate probability of necessity (PN) and sufficiency (PS) for each, classify into necessity-sufficiency quadrants. | | systematic-factor-ablation | Tactic: List all factors, remove one at a time, assess conclusion stability, rank factors by load-bearing importance. |
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | causal-claim-extraction | Extract all causal claims (X causes Y, X leads to Y, X enables Y) from an artifact, producing a structured list of cause-effect pairs. | | counterfactual-scenario-construction | Construct precise, internally consistent counterfactual scenarios where specified factors are altered, then reason about the resulting conclusion. | | factor-enumeration | List all key factors, conditions, and assumptions that support or enable the artifact's conclusion. | | fragility-measurement | Compute a fragility index from flip-point distances and degradation scores, summarizing how robust the conclusion is. | | load-bearing-identification | Identify which factors are "load-bearing walls" — factors whose removal would collapse the conclusion. | | necessity-evaluation | Evaluate the probability of necessity (PN) for a causal factor — would the conclusion fail if this factor were absent? | | single-factor-removal | Remove one specified factor from the artifact's support structure and reason about how the conclusion changes. | | sufficiency-evaluation | Evaluate the probability of sufficiency (PS) for a causal factor — would this factor alone be enough to produce the conclusion? |
<!-- 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-04 | fail→pass | 44,617 | 12,726 | -71% | 1 | 1 | 0% | 1,219 | 2,994 | +146% | 0 | 0 | — |
case-17 | fail→pass | 38,340 | 6,966 | -82% | 1 | 1 | 0% | 1,586 | 1,039 | -34% | 0 | 0 | — |
case-01 | pass→pass | 26,325 | 26,922 | +2% | 1 | 1 | 0% | 4,189 | 5,076 | +21% | 0 | 0 | — |
case-02 | pass→fail | 71,166 | 46,153 | -35% | 1 | 1 | 0% | 5,625 | 7,817 | +39% | 0 | 0 | — |
case-03 | pass→pass | 20,409 | 19,875 | -3% | 1 | 1 | 0% | 3,869 | 3,388 | -12% | 0 | 0 | — |
case-05 | fail→pass | 19,727 | 7,384 | -63% | 1 | 1 | 0% | 2,297 | 1,123 | -51% | 0 | 0 | — |
case-06 | fail→pass | 24,560 | 13,032 | -47% | 1 | 1 | 0% | 3,018 | 2,203 | -27% | 0 | 0 | — |
case-07 | pass→pass | 11,434 | 13,060 | +14% | 1 | 1 | 0% | 1,780 | 3,163 | +78% | 0 | 0 | — |
case-08 | pass→pass | 9,276 | 7,535 | -19% | 1 | 1 | 0% | 1,434 | 1,948 | +36% | 0 | 0 | — |
case-09 | fail→pass | 16,472 | 5,443 | -67% | 1 | 1 | 0% | 1,765 | 1,629 | -8% | 0 | 0 | — |
case-10 | fail→pass | 12,222 | 10,823 | -11% | 1 | 1 | 0% | 1,162 | 1,693 | +46% | 0 | 0 | — |
case-11 | fail→fail | 12,793 | 14,790 | +16% | 1 | 1 | 0% | 2,103 | 2,460 | +17% | 0 | 0 | — |
case-12 | fail→fail | 20,662 | 21,726 | +5% | 1 | 1 | 0% | 2,668 | 3,612 | +35% | 0 | 0 | — |
case-13 | pass→pass | 21,047 | 17,605 | -16% | 1 | 1 | 0% | 2,500 | 3,813 | +53% | 0 | 0 | — |
case-14 | fail→fail | 12,171 | 9,398 | -23% | 1 | 1 | 0% | 2,001 | 1,468 | -27% | 0 | 0 | — |
case-15 | fail→fail | 16,324 | 8,115 | -50% | 1 | 1 | 0% | 1,679 | 1,296 | -23% | 0 | 0 | — |
case-16 | fail→fail | 37,098 | 10,121 | -73% | 1 | 1 | 0% | 2,157 | 1,671 | -23% | 0 | 0 | — |
case-18 | pass→pass | 21,432 | 6,999 | -67% | 1 | 1 | 0% | 2,504 | 1,106 | -56% | 0 | 0 | — |
case-19 | fail→pass | 12,477 | 3,284 | -74% | 1 | 1 | 0% | 1,230 | 1,100 | -11% | 0 | 0 | — |
case-20 | pass→pass | 15,665 | 13,417 | -14% | 1 | 1 | 0% | 1,543 | 2,142 | +39% | 0 | 0 | — |
case-21 | fail→pass | 16,942 | 14,347 | -15% | 1 | 1 | 0% | 1,803 | 3,181 | +76% | 0 | 0 | — |
case-22 | fail→pass | 25,283 | 22,380 | -11% | 1 | 1 | 0% | 3,452 | 4,015 | +16% | 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 +36 percentage points is the difference between those two pass rates over the 20 comparable cases. 1 case got worse with the skill loaded, and it is 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.