yogsoth-ai Mechanism Extraction
Extract structured causal mechanisms from a theory description, providing a skeleton for variable identification and hypothesis construction.
HARD-GATE
Not satisfied → stop, return error: the theory description is too vague to extract a mechanism.
Pipeline
- Precondition check: verify completeness of the theory description
- Causal chain identification: identify X → Y causal-relation statements in the theory text
- Mediator extraction: identify intermediate variables (mediators) between X and Y
- Mechanism naming: give each mechanism chain a short descriptive name
- Mechanism diagram construction: represent as a directed graph (text ASCII or JSON nodes/edges)
- Output a structured mechanism list
Output Format
[
{
"name": "Mechanism name",
"chain": "X → mediator → Y",
"variables": {
"cause": "X description",
"mediator": "mediator description (null if direct)",
"effect": "Y description"
},
"source_theory": "Theory name this was extracted from",
"direction": "positive | negative | conditional",
"notes": "Any caveats or conditions"
}
]
Extract at least 1 mechanism chain per theory; at least 2 in total.