diff --git a/plans/abstractions.md b/plans/abstractions.md
index 3c37946f..d68a17f9 100644
--- a/plans/abstractions.md
+++ b/plans/abstractions.md
@@ -12,6 +12,12 @@ Anything short → _Watching_ (what's missing) or _Rejected_ (why).
 
 ## Last scan
 
+- **Date:** 2026-06-07 (radar loop, pass 20)
+- **Pass 20 — honest empty pass.** 3 new census recurrences since p17 (normalize/index ×2,
+  query ×3) — all **name collisions** (same noun, domain-specific op), added to the table.
+  Recorded the meta-pattern: the fleet shares vocabulary, not structure. Most subsystems
+  quiescent (acl/mod/search/commerce static ~9-15 passes = API-stable); only events/
+  identity/content/fed-sx still committing domain features. No new gate-clearer.
 - **Date:** 2026-06-07 (radar loop, pass 19)
 - **Pass 19 — honest empty pass.** Scanned 10 active subsystems. content/index.sx is a
   blog index/tag-cloud listing (presentation, not full-text search — no search reinvention)
@@ -135,6 +141,9 @@ subsystems/modules appear. Map:
 | `wire` ×2 | content(SX serializer) / mod(pipe-format) | Rejected — divergent |
 | `schema`,`engine` ×2 | acl/mod | substrate-twin parallels (Datalog vs Prolog); only audit (W4) is liftable |
 | `catalog`,`batch` ×2 | commerce/persist, mod/persist | name collisions, unrelated |
+| `normalize` ×2 | content(tree-prune)/feed(record-coerce) | name collision (pass 20) |
+| `index` ×2 | content(listing)/search(inverted index) | name collision (pass 20) |
+| `query` ×3 | content(doc-block)/search(bool AST)/persist(stream-read) | 3-way name collision (pass 20) |
 | `store` ×2 | content(on persist) / flow(workflow records) | related concept, divergent |
 | `rank` ×2 | feed/search | different domains (activities vs docs), ≤2 |
 **acl⇄mod are structural twins** (decision engine over a logic substrate, Datalog vs
@@ -142,6 +151,13 @@ Prolog) — they parallel across engine/schema/explain/audit/fed, but only the *
 is substrate-agnostic and liftable (→ W4); the rest are substrate-idiomatic. Next passes:
 re-test gates (W2/W3/W8) as consumers mature, watch new modules — not re-census.
 
+**Meta-pattern (pass 20):** new module names keep *recurring* but the operations keep
+*colliding* — same noun, domain-specific op (normalize, index, query, catalog, batch,
+notify, page, store all proved to be collisions). This is *why* genuine extraction
+candidates are rare: the fleet shares vocabulary, not structure. The real shared assets
+are the **substrate subsystems** (persist, flow, acl, fed-sx) that app domains *adopt*
+(W1/W2/W4/W7/W8), not hand-rolled libs to extract.
+
 **Scanning-method note (learned the hard way, passes 5/12/14/15):** a filename census
 for *cross-subsystem* recurrence MUST restrict to each subsystem's OWN namespace —
 `X/lib/X/*.sx` — never `X/lib/*/`. The merged substrate libs (`lib/prolog`, `lib/persist`,