Comparison Detail

The curated cross-dataset "Compared with" section (correlation-feature-proposal.md).

Deliberately no ?search= or arbitrary-pair form -- pair_key is one of a small, hand-registered set (databeat/comparisons.py's PAIR_KEY), enumerable and finite by design (proposal §3.3): if this endpoint cannot answer "compare arbitrary series A to arbitrary series B", nobody can build a discovery UI on top of it later without a deliberate decision to add one.

r, rho and the leave-one-out range are computed here, not in the browser, so any future consumer (an alert email, a newsletter) gets the number and its qualification together rather than reconstructing them.

404 for an unknown key OR a known one with too little data to say anything: the status is the same on purpose, matching renderCompanion()'s existing "a bonus, never a blocker" rule (the frontend and the MCP client both branch on !ok and degrade either case to nothing shown). The two are told apart by the error envelope's code only (#589 S2 / #1120, EA-approved distinction, built non-breaking): "unknown_pair" for a key that is not a registered comparison, "insufficient_data" for a registered pair with too little data right now, so an agent can tell "never will exist" from "check back later".