How divergence is computed
Every time the markets scorer writes a new score for a category, the pairer looks backward up to nine hours for the most recent editorial score for the same category. If one exists, it computes a signed gap: editorial − rescaled-markets, both on the 0–100 axis. The markets scorer's 1–10 bucket is lifted onto 0–100 via midpoint mapping (1 → 5, 5 → 45, 10 → 95) so the two are directly comparable.
The rationale text is written by an LLM that sees both editorial and markets rationales plus the gap, and is asked to commit to the most plausible reason for the divergence — drawing only from the two source rationales, no outside facts.
When markets have no coverage for a category, the pair is recorded with a null gap and rendered as a "no coverage" card. Only categories with both editorial and markets scores produce a comparable gap.
What this view is for: forming your own view faster. Editorial reads narrative, markets read price. When they line up, that's reinforcement. When they don't, the gap and its explanation give you a structured place to start your own read — not a recommendation, not a trade signal, not a prediction about which side is right.