Purpose
Deterministic deltas on same-metric, same-market pairs (#2.1); cross-definition comparisons flagged, never blended.
When to use
After research-retrieve in every run.
Inputs
question.json + config/research-index.json.
Steps
- Run scripts/trend_compare.py --question question.json --index research-index.json --out compared.json
- Quote deltas with both study ids and periods; state methods when they differ.
Output
compared.json - the {retrieved, trends, contradictions} hop.
Grounding requirements
Every delta cites both studies.
Constraints
- Like-for-like only; the model never converts metrics to force a comparison.
Escalation / uncertainty
Metric-definition mismatch -> flagged comparison, not a blended number.