evidence-collection

name	evidence-collection
description	Gather evidence for causal claims
execution	strategy
used-by	causal-modeling

Evidence Collection

Systematically gather and attach evidence pages to the causal claims in the model, creating supported_by edges for confirming evidence and contradicts edges for disconfirming evidence. A causal model without evidence provenance is speculation; this strategy converts it into a structured, auditable knowledge artifact.

Guiding Focus

CC should treat evidence collection as adversarial by default: for every supported_by edge added, actively search for contradicting evidence before moving on. The contradicts edges are as important as the supported_by edges — a model that only records confirming evidence is biased. Evidence pages must record the source, study design (where applicable), and a brief assessment of quality. Quantity matters less than coverage across independent sources.

Available Tactics

• evidence-weighing — score each evidence page by study design strength, sample size, and independence from other sources
• counterfactual-reasoning — use "what evidence would change this claim?" to direct the search toward the most informative gaps
• feedback-loop-detection — flag evidence patterns that suggest bidirectional causation rather than clean one-way support

Budget Slice

Metric	S	M	L
Evidence pages created	8	20	45
supported_by edges	10	25	50
contradicts edges flagged	2	5	10

State Ledger Template

| Metric                   | Target | Current | Status |
|--------------------------|--------|---------|--------|
| Evidence pages created   | S:8 / M:20 / L:45  | 0 | ⬜ |
| supported_by edges       | S:10 / M:25 / L:50 | 0 | ⬜ |
| contradicts edges flagged | S:2 / M:5 / L:10  | 0 | ⬜ |

Cannot exit until 80% of budget met. Print state ledger before each iteration decision.