Analysis Ecosystem

All-vs-all correlation across the user's metrics — the Signal tab payload.

Unlike Build, each metric contributes only its own measurements (no subtree aggregation): a parent aggregated over its children would be trivially correlated with every one of them, which is definitional noise in an all-vs-all view. Container metrics with no direct measurements drop out via the empty-series filter.

Per-metric missing policy defaults to each metric's saved policy, else its type default (:func:_metric_default_missing), and an explicit missing map overrides per metric. On top of the matrix payload (r/p/n/ci_low/ci_high) this adds:

• q: Benjamini–Hochberg FDR-adjusted p-values — edge visibility should key on these, not raw p (all-vs-all is a multiple-comparisons machine).
• order: seriation permutation (indices) from hierarchical clustering, so a heatmap renders with similar metrics adjacent.
• groups: cluster id per metric (aligned with metric_ids).
• observed: non-empty buckets per metric (node sizing).
• category_ids: category per metric (node coloring / grouping).
• dismissed: [metric_id_a, metric_id_b] pairs the user has marked unimportant (see /analysis/dismissed-correlations) — the client hides these by default rather than the server dropping them, so a "show dismissed" toggle needs no extra round trip.

Multi-scale mode (scales: ["D","W","M"] in the body): computes the matrix on each requested calendar grid, pools every scale's p-values into ONE Benjamini–Hochberg family (per-scale families would give each pair several shots at significance), and combines them into best-scale top-level matrices — each pair reported at the scale with the highest confidence-adjusted strength (LCB), ties to the finer scale. The daily grid reads a trailing DAILY_RECENCY_CAP_DAYS window unless custom_window: true; coarse scales read the full window and are skipped (never partially computed) when the data span can't reach MIN_CORR_N buckets. Adds best_scale (per-pair scale matrix), scales (computed codes), scale_meta (window/buckets/needed/computed per requested scale), and per_scale (full matrix set per computed scale, on the same union id order — the edge sheet's scale profile). insights are generated at each pair's chosen scale and carry a scale field. Without scales the endpoint behaves exactly as before (single resample).

Free plans get an outline, not the matrix. Everything above describes the Pro payload. A free caller receives the topology-only projection built by :func:_project_ecosystem — opaque node indices, cluster ids, bucketed sizes and index-pair edges — with no labels, statistics or insights. The endpoint itself stays open (rather than returning 402) precisely so the locked Signal page can still draw the user's real network shape as the thing they are being offered.