Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic: Core of the deductive path — start from theory to extract mechanisms, variables, and relationships, generating hypothesis candidates
.claude/skills/yogsoth-ai-theory-mechanism-extraction/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -17% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 64% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -12% | 0% |
Core of the deductive path — starting from existing theory, systematically extract mechanisms, identify variables, and specify relationships, providing a structured basis for deductive hypothesis generation.
The starting point of deductive reasoning is theory, not data. This tactic forces the CC to first identify relevant theories in the domain, then unpack layer by layer: theory → mechanism → variable → variable-to-variable relationship. Each layer is the prerequisite for the next; skipping steps is not allowed.
The final output is not the hypotheses themselves, but the "raw material" for hypotheses — each mechanism corresponds to at least one hypothesis candidate (with variables and direction), for the upstream strategy to further formalize.
| SOP | Responsibility | When to call | |-----|------|---------| | theory-identification | Identify existing theories relevant to the gap/insight (including theory name, core claim, scope of applicability) | Required in all modes, executed first | | mechanism-extraction | Extract operationalizable causal mechanisms from each theory (mechanism = the process linking cause and effect) | Required in all modes, after theory-identification | | variable-identification | Identify independent, dependent, moderating, and control variables from each mechanism | Required in all modes, after mechanism-extraction | | relationship-specification | Specify the directional relationships between variables (positive/negative/nonlinear/moderation/mediation), generating hypothesis candidates | Required in all modes, executed last |
Simplified (S tier, 1 theory)
Standard (M tier, 2-3 theories)
Deep (L tier, ≥3 theories)
After execution, report to the calling strategy:
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | hypothesis-formation-variable-identification | SOP: identify variables and their roles within a hypothesis | | mechanism-extraction | SOP: extract causal mechanism chains from a theory | | relationship-specification | SOP: specify the direction and form of relationships between variables | | theory-identification | SOP: Identify theoretical frameworks relevant to a research gap |
<!-- 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 | 51,910 | 45,969 | -11% | 1 | 1 | 0% | 8,304 | 8,011 | -4% | 0 | 0 | — |
case-02 | fail→fail | 48,351 | 41,037 | -15% | 1 | 1 | 0% | 8,296 | 7,914 | -5% | 0 | 0 | — |
case-08 | fail→fail | 21,395 | 19,970 | -7% | 1 | 1 | 0% | 3,269 | 3,775 | +15% | 0 | 0 | — |
case-03 | fail→pass | 48,272 | 39,329 | -19% | 1 | 1 | 0% | 7,557 | 7,189 | -5% | 0 | 0 | — |
case-04 | fail→pass | 25,759 | 19,237 | -25% | 1 | 1 | 0% | 3,483 | 3,434 | -1% | 0 | 0 | — |
case-05 | pass→pass | 24,077 | 29,667 | +23% | 1 | 1 | 0% | 4,334 | 5,050 | +17% | 0 | 0 | — |
case-06 | fail→pass | 47,264 | 34,813 | -26% | 1 | 1 | 0% | 7,404 | 6,132 | -17% | 0 | 0 | — |
case-07 | pass→fail | 47,635 | 49,109 | +3% | 1 | 1 | 0% | 7,347 | 8,235 | +12% | 0 | 0 | — |
case-09 | pass→pass | 18,214 | 17,372 | -5% | 1 | 1 | 0% | 2,601 | 3,480 | +34% | 0 | 0 | — |
case-10 | fail→fail | 10,296 | 10,365 | +1% | 1 | 1 | 0% | 848 | 1,506 | +78% | 0 | 0 | — |
case-11 | fail→pass | 22,049 | 27,279 | +24% | 1 | 1 | 0% | 2,837 | 4,651 | +64% | 0 | 0 | — |
case-12 | pass→pass | 51,565 | 33,173 | -36% | 1 | 1 | 0% | 6,973 | 5,588 | -20% | 0 | 0 | — |
case-13 | fail→fail | 29,310 | 27,440 | -6% | 1 | 1 | 0% | 4,274 | 4,964 | +16% | 0 | 0 | — |
case-14 | fail→pass | 53,988 | 39,542 | -27% | 1 | 1 | 0% | 8,234 | 7,281 | -12% | 0 | 0 | — |
case-15 | fail→fail | 29,769 | 26,675 | -10% | 1 | 1 | 0% | 3,451 | 4,399 | +27% | 0 | 0 | — |
case-16 | pass→pass | 28,279 | 43,300 | +53% | 1 | 1 | 0% | 5,344 | 6,450 | +21% | 0 | 0 | — |
case-17 | fail→pass | 26,437 | 27,544 | +4% | 1 | 1 | 0% | 3,459 | 6,044 | +75% | 0 | 0 | — |
case-18 | pass→pass | 29,054 | 23,990 | -17% | 1 | 1 | 0% | 4,273 | 5,205 | +22% | 0 | 0 | — |
case-19 | fail→pass | 55,432 | 28,057 | -49% | 1 | 1 | 0% | 8,225 | 5,896 | -28% | 0 | 0 | — |
case-20 | fail→pass | 27,719 | 26,432 | -5% | 1 | 1 | 0% | 4,865 | 5,802 | +19% | 0 | 0 | — |
case-21 | fail→pass | 38,767 | 42,821 | +10% | 1 | 1 | 0% | 6,047 | 7,873 | +30% | 0 | 0 | — |
case-22 | fail→pass | 50,494 | 64,217 | +27% | 1 | 1 | 0% | 8,234 | 9,061 | +10% | 0 | 0 | — |
case-23 | fail→pass | 48,483 | 31,041 | -36% | 1 | 1 | 0% | 7,772 | 5,715 | -26% | 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 +43 percentage points is the difference between those two pass rates over the 23 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.