11 — Route / UI Recommendation (VERIFIED v2)

Decision: KEEP /knowledge/pivot canonical · CONVERGE /knowledge/registries onto it · OPTIONAL alias.

Why not a new page

A third route (/knowledge/registries-pivot) would add a counting surface and more drift. The clean architecture already exists at /knowledge/pivot (render-shell over pivot_results). Build nothing new; reconcile.

End state

1. /knowledge/pivot — canonical counting surface (Tab Pivot 9 rows + pivot_matrix). KEEP. Add: visible refreshed_at; point the DOT row at PIV-007 (309) not CAT-DOT (307).
2. /knowledge/registries — the list-of-lists navigator over meta_catalog (169). KEEP its matrices (Taxonomy 6x6 / Species Matrix / Entity Matrix already present), but: (a) remove hardcoded phantom rows + in-page gap math (Đ28, doc 03/07); (b) source numbers from the pivot path; (c) link each list row to its pivot + the entity's own profile (Đ38/44).
3. /knowledge/registries-pivot — OPTIONAL alias/redirect → same registries component. No logic fork.

Responsibility split (non-overlapping)

• registries = "what lists exist" (navigation / list-of-lists / profile links).
• pivot = "how many, by which dimension" (counts / matrix).
• Both read the same pivot truth; neither computes in Nuxt.

UI standard (OS Agency / AgencyOS, data-driven)

• Layers/species/lists/numbers all from data. Caveat: the 6-layer reference table does not exist yet (doc 06) — layer grouping must wait on R6 (build composition_levels) rather than hardcode the 6.
• Every count shows pivot_code + refreshed_at (honest staleness).
• Consistent tokens with existing surfaces (#639922 / #ef9f27 / #e24b4a).
• A list with no pivot is shown as "no pivot" — never a hardcoded number.

Sequencing

Lowest-risk first: P1 (read-only contract) → P4 (freshness) → P2/P3 (DOT dedupe + record≠actual) under Đ32 → R6 build composition_levels → R5/P7 de-hardcode UI. No UI fork until the count truth is single.