diff --git a/scripts/audits/14096-tenancy-ledger-namespace-regexp-census.mjs b/scripts/audits/14096-tenancy-ledger-namespace-regexp-census.mjs
index 2f4013754f..4e71b194d9 100644
--- a/scripts/audits/14096-tenancy-ledger-namespace-regexp-census.mjs
+++ b/scripts/audits/14096-tenancy-ledger-namespace-regexp-census.mjs
@@ -94,7 +94,8 @@ const REPO_ROOT = resolve(HERE, '../..');
 // ---------------------------------------------------------------------------
 // §1. Discover every registered platform-namespace object by AST-lite census
 //     of `ObjectSchema.create(` calls — the same method the ledger's own
-//     header (`platform-object-tenancy.ts:32-37`) documents using.
+//     header (`packages/objectql/src/tenancy/platform-object-tenancy.ts`, its
+//     opening comment — lines 32-37 as measured) documents using.
 // ---------------------------------------------------------------------------
 
 /** Blank out `//` and `/* *\/` comments and string contents (chars only,
diff --git a/scripts/audits/14423-unbound-declarations-before-count.mjs b/scripts/audits/14423-unbound-declarations-before-count.mjs
index 28873a123c..1e9011ded3 100644
--- a/scripts/audits/14423-unbound-declarations-before-count.mjs
+++ b/scripts/audits/14423-unbound-declarations-before-count.mjs
@@ -59,7 +59,8 @@ const loadStandaloneActions = async () => [...namelessStandaloneOrphans, namedSt
 
 const { warnings, logger } = silentLogger();
 
-// Exactly as ObjectQLPlugin.runGovernanceInventory calls it (plugin.ts:2516),
+// Exactly as ObjectQLPlugin.runGovernanceInventory calls it
+// (`packages/objectql/src/plugin.ts#runGovernanceInventory`),
 // with an empty registered-handler set and no object-embedded actions/registry
 // rung, so the ENTIRE declaration set comes from `loadStandaloneActions`.
 await runActionGovernanceInventory({
diff --git a/scripts/check-adr-links.mjs b/scripts/check-adr-links.mjs
index 09e41584ce..97d93e26ed 100644
--- a/scripts/check-adr-links.mjs
+++ b/scripts/check-adr-links.mjs
@@ -11,8 +11,9 @@
 // checked by nothing, and records link each other heavily: most carry a
 // `**Builds on**:` line with 3-10 relative links in it.
 //
-// That is not a theoretical hole. `0057-erp-authorization-core-business-units-
-// and-scope-depth.md:5` pointed at `./0010-metadata-protection.md`, a file that
+// That is not a theoretical hole.
+// `docs/adr/0057-erp-authorization-core-business-units-and-scope-depth.md`
+// (line 5 as measured) pointed at `./0010-metadata-protection.md`, a file that
 // has never existed (the record is `0010-metadata-protection-model.md`). It sat
 // there long enough that a triage comment inherited the wrong attribution and
 // repeated it. PD #13 sends the next author to "go read that decision"; a dead
diff --git a/scripts/check-changeset-no-major.mjs b/scripts/check-changeset-no-major.mjs
index b9c431b736..7f5799d91c 100644
--- a/scripts/check-changeset-no-major.mjs
+++ b/scripts/check-changeset-no-major.mjs
@@ -687,7 +687,8 @@ export function render(result) {
     // nothing final ships until `changeset pre exit`. Surface the introduced
     // majors for the RC curator, but do not fail. The guard re-arms once
     // pre-mode exits: `changeset pre exit` rewrites pre.json's mode to `"exit"`
-    // (@changesets/pre@2.0.2, changesets-pre.cjs.js:117), which is not `pre`.
+    // (@changesets/pre@2.0.2, its `changesets-pre.cjs.js` bundle, line 117 as
+    // pinned — a dependency file, outside this tree), which is not `pre`.
     case 'exempt':
       stdout.push(
         `✓ Changesets is in pre-release mode (tag: ${result.tag}) — ` +
diff --git a/scripts/check-cli-test-child-env.mjs b/scripts/check-cli-test-child-env.mjs
index d290b48b5d..22b37d1511 100644
--- a/scripts/check-cli-test-child-env.mjs
+++ b/scripts/check-cli-test-child-env.mjs
@@ -185,7 +185,8 @@
  * suite stays green, and the comment claiming the built entry stays plausible.
  *
  * `@oclif/core@4.13.3` skips its TypeScript path lookup only when `isProd()`,
- * which `lib/util/util.js:66` defines as a NEGATED membership test -- a leading
+ * which its `lib/util/util.js` (line 66 as pinned -- a dependency file, outside
+ * this tree) defines as a NEGATED membership test -- a leading
  * logical-NOT over `['development', 'test'].includes(process.env.NODE_ENV ?? '')`.
  * Measured directly against `Config.load()` on `packages/cli`, with
  * `packages/cli/dist` present, reading back the root plugin's `commandsDir`:
@@ -1056,7 +1057,8 @@ const BUILT_ENTRYPOINT = 'bin/run.js';
  * The `NODE_ENV` values that send `@oclif/core`'s command lookup to `src/`.
  *
  * `@oclif/core@4.13.3` skips its TypeScript path lookup only when `isProd()`,
- * which `lib/util/util.js:66` defines as a NEGATED membership test:
+ * which its `lib/util/util.js` (line 66 as pinned -- a dependency file, outside
+ * this tree) defines as a NEGATED membership test:
  * `!['development', 'test'].includes(process.env.NODE_ENV ?? '')`. So these two
  * values -- and only these two -- turn the reroute ON. `production` and unset
  * both leave it off, which is why the rule below refuses a SET of values rather
diff --git a/scripts/check-console-injection.mjs b/scripts/check-console-injection.mjs
index f4c1f4bb44..6b36a221a2 100644
--- a/scripts/check-console-injection.mjs
+++ b/scripts/check-console-injection.mjs
@@ -99,7 +99,8 @@
  *     esbuild re-emits `"…definition's…"`. Same characters, different bytes, so a
  *     literal substring search over source text misses it.
  *   - 19/44 constant-folded concatenation. Source splits a long description as
- *     `'… declares ' + '`_packageId`.'` (api/protocol.zod.ts:341-342 is one);
+ *     `'… declares ' + '`_packageId`.'` (`packages/spec/src/api/protocol.zod.ts`
+ *     is one, lines 341-342 as measured);
  *     the bundler folds it to one literal that exists in no source file.
  *
  * A missed probe here reads as "not expired" — a SILENT PASS, the same failure
diff --git a/scripts/check-dispatcher-error-vocabulary.mjs b/scripts/check-dispatcher-error-vocabulary.mjs
index 026f10fc93..ba3f271ec8 100644
--- a/scripts/check-dispatcher-error-vocabulary.mjs
+++ b/scripts/check-dispatcher-error-vocabulary.mjs
@@ -925,7 +925,7 @@ export const OBJECT_LITERAL_CODE_HELPER_BLINDNESS = Object.freeze({
  * for the D6 field-addressed validation catalog in `domains/automation.ts`, and
  * [#14626]'s two newly reached positions in that same file owe NO further rows:
  * they are the identical catalog ternary with the identical values, so they
- * derive the site keys the `:1125` instance already derives.
+ * derive the site keys the instance at line 1125 of that file already derives.
  *
  * ## [#14742] What moved this census, separated from what merely drifted
  *
@@ -1196,7 +1196,8 @@ export const INLINE_LITERAL_EXPRESSION_CENSUS = Object.freeze({
     }),
     // [#14626] The two the card predicted, recovered by the template-literal
     // mode. Same file, same genre, same two values — so they derive the site
-    // keys `:1125` already derives and owe NO further verdict rows. That is the
+    // keys the instance at line 1125 of that file already derives, and owe NO
+    // further verdict rows. That is the
     // measurement, not a convenience: `rowsOwed: 0` here is what says the
     // recovered positions surfaced nothing unregistered.
     Object.freeze({
diff --git a/scripts/check-doc-anchors.mjs b/scripts/check-doc-anchors.mjs
index 8ccfd1c4dc..d52ae3d5e2 100644
--- a/scripts/check-doc-anchors.mjs
+++ b/scripts/check-doc-anchors.mjs
@@ -16,7 +16,8 @@
 // heading lives in one file, the link in another, and until this script nothing
 // in CI related them: a rename that updates the heading and not the inbound
 // `#anchor` shipped green. #7465 exists because exactly that pairing had already
-// drifted once — `cli.mdx:458` pointed at a heading whose text had been correct
+// drifted once — `content/docs/deployment/cli.mdx` (line 458 as measured then)
+// pointed at a heading whose text had been correct
 // in 2024 — and PR #7483 could only catch its own rename by hand, with a slug
 // computed in a REPL and a repo-wide grep. Neither runs on anyone else's PR.
 //
diff --git a/scripts/check-durability-degradation-log-level.mjs b/scripts/check-durability-degradation-log-level.mjs
index 38738c52d4..87a4b61783 100644
--- a/scripts/check-durability-degradation-log-level.mjs
+++ b/scripts/check-durability-degradation-log-level.mjs
@@ -768,8 +768,10 @@ const FAILURE_PROPAGATION_SITES = new Map([
 // that fix — `67 read seam(s) … (7 … discriminated) (1 pass … through) (1
 // baselined)` on both trees, measured by ablating the fix back to its
 // pre-#8895 `catch { continue }` and re-running. And the seam is not merely
-// unseen: plant a `return []` in that same catch and the gate names it
-// (`engine.ts:10084 (in cascadeDeleteRelations())`, red). So it sits IN the
+// unseen: plant a `return []` in that same catch and the gate names it — red,
+// at `packages/objectql/src/engine.ts#cascadeDeleteRelations` (that catch was
+// on line 10084 when this was measured, a dated reading and not a pointer).
+// So it sits IN the
 // census throughout, reported clean while it was broken and reported clean now
 // that it is fixed — what decides visibility is the SHAPE of the exit, never
 // whether the seam is correct. A real fail-open on an integrity guard was
@@ -1176,24 +1178,28 @@ const FAILURE_PROPAGATION_SITES = new Map([
 // THE DISCRIMINATING RUN. Raising `MAX_READ_WRAPPER_DEPTH` from 2 to 6 while
 // leaving `walkSameTickInclusive` in place admits 5 of the same 8 (70 seams: +8
 // / -2, where the 2 are the SAME try lines re-attributed to a different
-// first-matching callee, engine.ts:9407 and :10572). Saturation checked at
+// first-matching callee, both in `packages/objectql/src/engine.ts`, on lines
+// 9407 and 10572 as measured). Saturation checked at
 // depth 50: 70 and 75, i.e. unchanged. So for those 5 the miss is the DEPTH
 // BOUND, not the callback boundary — `walkAll` merely masks the bound by
 // descending lexically through nested DECLARATIONS instead of counting call
 // hops, which reaches the read at depth 1 no matter how many awaits are between.
 //
-// THE 8 DELTA SEAMS, each read at its call site:
-//
-//   | # | seam (try line → wrapper)                                      | why today misses it | invoked now? |
-//   |--:|----------------------------------------------------------------|---------------------|--------------|
-//   | 1 | metadata-protocol protocol.ts:10243 getMetaItemCached→getMetaItem | depth bound       | yes — real   |
-//   | 2 | metadata-protocol protocol.ts:13559 saveMetaItem→getMetaItem      | depth bound       | yes — real   |
-//   | 3 | metadata-protocol protocol.ts:14535 migrateStoredMetadata→saveMetaItem | depth bound   | yes — real   |
-//   | 4 | metadata-protocol protocol.ts:17213 duplicatePackage→saveMetaItem  | depth bound       | yes — real   |
-//   | 5 | metadata-protocol sys-metadata-repository.ts:883 promoteDraft→dropPromotedDraftRow | CALLBACK | yes — real |
-//   | 6 | metadata-protocol sys-metadata-repository.ts:1353 close→terminate  | CALLBACK          | NO — FAKE    |
-//   | 7 | objectql engine.ts:9237 insert→applyAutonumbers                    | CALLBACK          | yes — real   |
-//   | 8 | objectql lifecycle-service.ts:625 sweep→reapObject                 | depth bound       | yes — real   |
+// THE 8 DELTA SEAMS, each read at its call site. ⚠️ The FILE is named as an
+// anchor and the try line sits beside it as DATA: each number is a reading
+// taken on the date above, not a pointer, and a `file:NNN` pointer written
+// here rots silently — the whole finding of #15765.
+//
+//   | # | file                                                  | try line | seam (wrapper)                     | why today misses it | invoked now? |
+//   |--:|-------------------------------------------------------|---------:|------------------------------------|---------------------|--------------|
+//   | 1 | `packages/metadata-protocol/src/protocol.ts`               |    10243 | getMetaItemCached→getMetaItem      | depth bound       | yes — real   |
+//   | 2 | `packages/metadata-protocol/src/protocol.ts`               |    13559 | saveMetaItem→getMetaItem           | depth bound       | yes — real   |
+//   | 3 | `packages/metadata-protocol/src/protocol.ts`               |    14535 | migrateStoredMetadata→saveMetaItem | depth bound       | yes — real   |
+//   | 4 | `packages/metadata-protocol/src/protocol.ts`               |    17213 | duplicatePackage→saveMetaItem      | depth bound       | yes — real   |
+//   | 5 | `packages/metadata-protocol/src/sys-metadata-repository.ts` |      883 | promoteDraft→dropPromotedDraftRow  | CALLBACK          | yes — real   |
+//   | 6 | `packages/metadata-protocol/src/sys-metadata-repository.ts` |     1353 | close→terminate                    | CALLBACK          | NO — FAKE    |
+//   | 7 | `packages/objectql/src/engine.ts`                          |     9237 | insert→applyAutonumbers            | CALLBACK          | yes — real   |
+//   | 8 | `packages/objectql/src/lifecycle/lifecycle-service.ts`     |      625 | sweep→reapObject                   | depth bound       | yes — real   |
 //
 // All 8 were decidable from the call site; none needed provenance. Seams 1-5,
 // 7 and 8 are genuine members the census does not count: every hop is an
@@ -1205,7 +1211,8 @@ const FAILURE_PROPAGATION_SITES = new Map([
 //
 // ⚠️ SEAM 6 IS A FAKE SEAM, AND IT IS THE REASON `walkAll` IS NOT THE FIX.
 // `close()`'s try calls `w.terminate()`. `terminate` resolves BY NAME to the
-// local const arrow at sys-metadata-repository.ts:1246 — a synchronous, void,
+// local const arrow in `packages/metadata-protocol/src/sys-metadata-repository.ts`
+// (line 1246 as measured) — a synchronous, void,
 // in-memory routine whose only call is `self.watchers.delete(subscription)` on
 // `private readonly watchers = new Set<...>()`. `calleeName` reads that as
 // `delete`, and the wrapper recursion resolves `delete` to THIS FILE'S
@@ -1254,8 +1261,10 @@ const FAILURE_PROPAGATION_SITES = new Map([
 //   | probe, with the `delete` wrapper hop refused      |         72 |
 //
 // ⚠️ THE ABLATION IS THE CONTROL, NOT THE FIX. Refusing the `delete` hop
-// outright drops TWO seams — sys-metadata-repository.ts:1353 `close`->
-// `terminate` (the FAKE) and :883 `promoteDraft`->`dropPromotedDraftRow` (a
+// outright drops TWO seams, both in
+// `packages/metadata-protocol/src/sys-metadata-repository.ts` — `close`->
+// `terminate` (the FAKE, line 1353 as measured) and
+// `promoteDraft`->`dropPromotedDraftRow` (line 883, a
 // REAL `await this.delete(ref, …)`). In a summary that is indistinguishable
 // from the correct outcome, which is why the pair is pinned in the self-test.
 //
@@ -1338,7 +1347,8 @@ const FAILURE_PROPAGATION_SITES = new Map([
 //   - THE ADMITTING STEP IS +6 / -0, not "+8 / -2". The 2 in the filing were
 //     two `try` lines re-attributed to a different first-matching callee, never
 //     seams leaving the population. That re-attribution still happens here —
-//     `engine.ts:9741` and `:10906` move from `resolveMasterDetailParent(s)` to
+//     two try lines in `packages/objectql/src/engine.ts` (9741 and 10906 as
+//     measured) move from `resolveMasterDetailParent(s)` to
 //     `mediaValueShapeStrictFor` — but at depth 4, one level ABOVE the level
 //     that admits anything, and it moves no count. The admitting step is clean.
 //
@@ -1347,14 +1357,16 @@ const FAILURE_PROPAGATION_SITES = new Map([
 // rule's vocabulary names. Chains as the recognizer actually resolves them —
 // two of them are NOT the tails the filing predicted:
 //
-//   | # | seam (try line -> first wrapper)                    | chain to the read |
-//   |--:|-----------------------------------------------------|-------------------|
-//   | 1 | metadata-protocol protocol.ts:10497 getMetaItemCached | getMetaItem -> findDraft -> lookup -> engine.findOne |
-//   | 2 | metadata-protocol protocol.ts:13874 saveMetaItem      | getMetaItem -> findDraft -> lookup -> engine.findOne |
-//   | 3 | metadata-protocol protocol.ts:14850 migrateStoredMetadata | saveMetaItem -> refuseUnmintableMetaType -> metaTypeNamespaceExists -> engine.findOne |
-//   | 4 | metadata-protocol protocol.ts:16502 publishPackageDrafts  | promoteDraftForPublish -> lockWriteRefusal -> getEffectiveLock -> engine.findOne |
-//   | 5 | metadata-protocol protocol.ts:17565 duplicatePackage      | saveMetaItem -> refuseUnmintableMetaType -> metaTypeNamespaceExists -> engine.findOne |
-//   | 6 | objectql lifecycle-service.ts:664 sweep                   | reapObject -> archiveObject -> archivePass -> hot.find |
+//   (the FILE is the anchor, the try line beside it is a dated reading)
+//
+//   | # | file                                                  | try line | first wrapper         | chain to the read |
+//   |--:|-------------------------------------------------------|---------:|-----------------------|-------------------|
+//   | 1 | `packages/metadata-protocol/src/protocol.ts`           |    10497 | getMetaItemCached     | getMetaItem -> findDraft -> lookup -> engine.findOne |
+//   | 2 | `packages/metadata-protocol/src/protocol.ts`           |    13874 | saveMetaItem          | getMetaItem -> findDraft -> lookup -> engine.findOne |
+//   | 3 | `packages/metadata-protocol/src/protocol.ts`           |    14850 | migrateStoredMetadata | saveMetaItem -> refuseUnmintableMetaType -> metaTypeNamespaceExists -> engine.findOne |
+//   | 4 | `packages/metadata-protocol/src/protocol.ts`           |    16502 | publishPackageDrafts  | promoteDraftForPublish -> lockWriteRefusal -> getEffectiveLock -> engine.findOne |
+//   | 5 | `packages/metadata-protocol/src/protocol.ts`           |    17565 | duplicatePackage      | saveMetaItem -> refuseUnmintableMetaType -> metaTypeNamespaceExists -> engine.findOne |
+//   | 6 | `packages/objectql/src/lifecycle/lifecycle-service.ts` |      664 | sweep                 | reapObject -> archiveObject -> archivePass -> hot.find |
 //
 // (3 and 5 reach a read through `saveMetaItem`'s OWN precondition check, not
 // through the `getMetaItem` tail the filing assigned them. Same verdict, and
@@ -1378,8 +1390,9 @@ const FAILURE_PROPAGATION_SITES = new Map([
 //   walkAll @ depth 2 admits   7   (was 8 — #12358's guard deleted the fake seam)
 //   depth 3 admits             6
 //   BOTH                       5   <- the claim, reproduced exactly
-//   depth only                 1   protocol.ts:16502 publishPackageDrafts
-//   walkAll only               2   sys-metadata-repository.ts:883, engine.ts:9571
+//   depth only                 1   `packages/metadata-protocol/src/protocol.ts`, line 16502
+//   walkAll only               2   `packages/metadata-protocol/src/sys-metadata-repository.ts`, line 883
+//                                  and `packages/objectql/src/engine.ts`, line 9571
 //
 // So the NUMBER 5 survives and its DENOMINATOR does not: read it as "5 of 7",
 // the eighth having been the fake seam #12358 removed. And one correction the
diff --git a/scripts/check-engine-double-contract.mjs b/scripts/check-engine-double-contract.mjs
index e1ea8c1dc2..d5466a15d8 100644
--- a/scripts/check-engine-double-contract.mjs
+++ b/scripts/check-engine-double-contract.mjs
@@ -101,8 +101,9 @@
 //     NO on #7620 for these doubles specifically. So it asks each independent
 //     double a behavioural question instead of handing it an implementation.
 //   - a scoped repository that declares NO repository-only member. Measured on
-//     the corpus this landed against: `packages/runtime/src/action-body-identity
-//     .test.ts:71` is a real scoped facade (`createContext().object(name)`)
+//     the corpus this landed against:
+//     `packages/runtime/src/action-body-identity.test.ts` (line 71 as measured)
+//     is a real scoped facade (`createContext().object(name)`)
 //     spelling only `find`/`count`/`insert`/`update`/`delete`, and it stays in
 //     the ledger. Seeing it would mean reading its parameter NAMES, and `o` is
 //     ambiguous in exactly this repo: `o: string` is the object name in twelve
diff --git a/scripts/check-live-db-isolation.mjs b/scripts/check-live-db-isolation.mjs
index b15e16b175..a9a5b93e83 100644
--- a/scripts/check-live-db-isolation.mjs
+++ b/scripts/check-live-db-isolation.mjs
@@ -214,8 +214,8 @@ const SOLE_INTERPOLATION = /^\$\{\s*([A-Za-z_$][\w$]*)\s*\}$/;
  * Source with block and line comments BLANKED -- replaced space-for-space
  * rather than deleted, so prose is never a hit and the line numbers this gate
  * reports are still the line numbers in the real file. Deleting the comments
- * was the first spelling and it reported `…live-mysql.test.ts:45` for a
- * statement that lives on line 81; a gate that points at the wrong line is a
+ * was the first spelling and it reported the live-mysql isolation test at line
+ * 45 for a statement that lives on line 81; a gate that points at the wrong line is a
  * gate the next author stops believing.
  */
 export function codeOf(source) {
diff --git a/scripts/check-overlay-whitelist-table.mjs b/scripts/check-overlay-whitelist-table.mjs
index 85b6d7890e..ac2d332e24 100644
--- a/scripts/check-overlay-whitelist-table.mjs
+++ b/scripts/check-overlay-whitelist-table.mjs
@@ -173,7 +173,8 @@
  * lines across 13 files. Of those, ~5 are genuine "<type> is <bool>" assertions;
  * the rest are the schema-field reference row, a code sample using an invented
  * type, and sentences ABOUT the flag rather than about any type
- * (`metadata-lifecycle.mdx:121` itself is one). Covering them mechanically
+ * (`content/docs/concepts/metadata-lifecycle.mdx` itself is one, line 121 as
+ * measured). Covering them mechanically
  * means a co-occurrence regex over English plus a per-site allowlist of ~17
  * exceptions to catch ~5 assertions -- a worse gate than none, and the kind
  * whose baseline gets bulk-updated to green. They are left out on purpose, and
diff --git a/scripts/check-platform-checklist.mjs b/scripts/check-platform-checklist.mjs
index 9065f9fed4..71ad13dedc 100644
--- a/scripts/check-platform-checklist.mjs
+++ b/scripts/check-platform-checklist.mjs
@@ -1282,7 +1282,8 @@ export const NEIGHBOURING_MAP: Readonly<Record<string, string>> = Object.freeze(
 // half, and the cheap half is the one that removes a false signal today.
 //
 // A citation is a colon-then-digits reached one of two ways, because the ledger
-// spelled it both ways: anchored to a source filename (`manifest.zod.ts:158`),
+// spelled it both ways: anchored to a source filename (`<file>.ts:158`, the
+// placeholder spelling `scripts/symbol-anchors.mjs#ANCHOR_GRAMMAR` uses),
 // or BARE, continuing a filename named earlier in the same sentence
 // (`ManifestSchema id :140 and version :202`). The bare half is why a plain
 // "filename followed by a colon" rule is not enough — and the bare half is the
diff --git a/scripts/check-runtime-services-index.mjs b/scripts/check-runtime-services-index.mjs
index 8845c0d3a7..69640bc252 100644
--- a/scripts/check-runtime-services-index.mjs
+++ b/scripts/check-runtime-services-index.mjs
@@ -28,7 +28,8 @@
 // time, and nothing reads them: `check:docs-audit-scope` derives WHICH pages the
 // docs-accuracy audit covers, never whether an index enumerates them. #9604
 // measured the result -- `services.sms` had a page, a `meta.json` entry, a
-// registered slot (`sms-plugin.ts:181`) and a canonical-source row, and was
+// registered slot (`packages/services/service-sms/src/sms-plugin.ts`, line 181
+// as measured) and a canonical-source row, and was
 // still missing from BOTH index lists. It shipped that way and every gate was
 // green. #9588 was the same page drifting on a different line.
 //
@@ -124,7 +125,8 @@
 // This gate deliberately did NOT read that list until now, and the reason is
 // worth keeping: it was a superset by exactly one row on purpose-of-record.
 // `Security: packages/spec/src/contracts/security-service.ts` named a real,
-// registered slot (`security-plugin.ts:1157`) for a service this chapter has no
+// registered slot (`packages/plugins/plugin-security/src/security-plugin.ts`,
+// line 1157 as measured) for a service this chapter has no
 // page for, while `services.security` was documented NOWHERE under
 // `content/docs/`. Whether that row should become a page, move to another
 // chapter, or be dropped was a product-surface question, and encoding ANY of the
@@ -256,7 +258,8 @@ const IS_TEST_PATH = (rel) =>
 
 // MULTILINE-AWARE ON PURPOSE. `\s*` spans newlines, so a registration whose key
 // sits on the line after the `(` is still found -- `plugin-audit` really is
-// written that way (`audit-plugin.ts:117`), and a `^.*registerService\('x'` style
+// written that way (`packages/plugins/plugin-audit/src/audit-plugin.ts`, line
+// 117 as measured), and a `^.*registerService\('x'` style
 // line-at-a-time grep reports a confident ZERO for it. An absence measured with
 // the wrong tool is the failure mode this whole check exists to rule out, so the
 // self-test pins the split-line form explicitly.
diff --git a/scripts/check-scripts-symbol-anchors.mjs b/scripts/check-scripts-symbol-anchors.mjs
index 23751124f5..19c4914e44 100644
--- a/scripts/check-scripts-symbol-anchors.mjs
+++ b/scripts/check-scripts-symbol-anchors.mjs
@@ -7,6 +7,7 @@
  *
  *   node scripts/check-scripts-symbol-anchors.mjs
  *   node scripts/check-scripts-symbol-anchors.mjs --list
+ *   node scripts/check-scripts-symbol-anchors.mjs --list-unresolvable
  *   node scripts/check-scripts-symbol-anchors.mjs --self-test
  *
  * ⚠️ THE MECHANISM IS NOT HERE. The grammar, the extractor, the comment-prose
@@ -54,15 +55,31 @@
  *
  * ## What this corpus judges, and what it declines to
  *
- * `judgeUntrackedLineAnchors: false`. The 96 citations that name no tracked
- * file are 66 bare filenames inside census tables (`engine.ts` at some line --
- * an abbreviation no resolver can bind to one of this tree's several files of
- * that name), 22 continuations inheriting no path, 11 directory-qualified
- * illustrations or sibling-repo paths, and 1 tilde form. They are a real defect
- * class and are recorded as a follow-up, exactly as the 1,056 bare paths under
- * `checkBarePaths` were for `docs/adr/**` -- but this gate cannot tell their
- * author how to fix them, and a gate whose only remedy is "stop writing that"
- * is the permanently-red gate this repo retired.
+ * `judgeUntrackedLineAnchors: false`. A citation that names no tracked file is
+ * SEEN, COUNTED and ENUMERATED, and judged by nothing: this gate cannot tell
+ * its author how to fix it, and a gate whose only remedy is "stop writing that"
+ * is the permanently-red gate this repo retired. They were a real defect class
+ * and were carded as a follow-up (#15809), exactly as the 1,056 bare paths
+ * under `checkBarePaths` were for `docs/adr/**`.
+ *
+ * ⭐ THAT FOLLOW-UP IS MOSTLY DONE, AND THE REST IS A LIST, NOT A NUMBER. When
+ * the corpus was registered it declined 96 citations; #15809 migrated 81 of
+ * them by the same method PR #15806 used on the tracked-target ones -- the file
+ * as a file-level anchor, the number kept beside it as data, the placeholder
+ * spelling `scripts/symbol-anchors.mjs#ANCHOR_GRAMMAR` defines for an
+ * illustration, and prose naming the repo for a third-party or dependency
+ * source no in-repo resolver could ever check. ⛔ Not one digit was repaired or
+ * repointed; every number that was in an anchor is still on its page, as data.
+ *
+ * The residual when that landed was 15, and every one of them was a file
+ * another lane held OPEN at that moment -- ⛔ not one was an ambiguity. That
+ * count is a DATED reading; `--list-unresolvable` is the live one. ⚠️ It is why the
+ * flag is still `false`: the fence #15809 was dispatched under is that it flips
+ * only when the residual is ZERO and a self-test pins the flip, and a residual
+ * of 15 would make this gate permanently red for the length of somebody else's
+ * pull request. `--list-unresolvable` prints the residual so the next author
+ * inherits a worklist rather than a count; the day it prints nothing, the flag
+ * is a one-line change with a case to pin it.
  *
  * `checkBarePaths: false`, for the same reason and on a measurement: judging
  * every bare path code span in this corpus produces 1,617 findings, nearly all
@@ -213,8 +230,32 @@ export function runCheck(root = process.cwd()) {
 }
 
 function list(root = process.cwd()) {
-  const { findings, counts } = sweepCorpus(CORPUS, root);
-  console.log(JSON.stringify({ counts, findings, allowances: HELD_FILE_ALLOWANCES }, null, 2));
+  const { findings, counts, declined } = sweepCorpus(CORPUS, root);
+  console.log(JSON.stringify({ counts, findings, declined, allowances: HELD_FILE_ALLOWANCES }, null, 2));
+}
+
+/**
+ * The citations this corpus DECLINES to judge, enumerated.
+ *
+ * ⚠️ A listing, never a verdict: this exits 0 whatever it prints, exactly as
+ * `--list` does. `runCheck()` is still the only arm that can fail.
+ *
+ * The count alone (`unresolvableLineCitation`, printed by the green line) says
+ * a residual exists without saying where, so nobody can work it down and
+ * nobody can tell a residual that SHRANK from one that moved. This prints
+ * `file:line -- citation` per row and a tally by shape, which is what a
+ * follow-up card needs to be closed rather than re-measured.
+ */
+function listUnresolvable(root = process.cwd()) {
+  const { counts, declined } = sweepCorpus(CORPUS, root);
+  const byShape = {};
+  for (const d of declined) byShape[d.shape] = (byShape[d.shape] ?? 0) + 1;
+  for (const d of declined) console.log(`${d.doc}:${d.line}  ${d.raw}   [${d.shape}]`);
+  console.log(
+    `\n${declined.length} citation(s) name no tracked file and are not judged `
+      + `(counter: ${counts.unresolvableLineCitation}) — `
+      + Object.entries(byShape).sort().map(([k, v]) => `${v} ${k}`).join(', '),
+  );
 }
 
 /* ─────────────────────────────── self-test ─────────────────────────────── */
@@ -242,11 +283,14 @@ function assert(cond, msg) { if (!cond) { console.error(`❌ check-scripts-symbo
 // row was retired, then 30 → 26 when the `scripts/check-adr-0087-registration.mjs`
 // row was retired and the array went EMPTY (#15765). Any other drop is cases that
 // STOPPED RUNNING; find what stopped registering instead of moving the number.
+// 26 → 29 when the declined citations gained an ENUMERATION beside their count
+// (#15809): three cases hold `declined` equal to `unresolvableLineCitation`, to
+// the file/line/text a residual list needs, and to the shape classification.
 // ⚠️ An empty array does NOT make the allowance battery vacuous: its five
 // fixture cases run off `scripts/bad.mjs`, never off the live rows, so they are
 // not part of this arithmetic and must never fall out of the count.
 const SELF_TEST_BATTERIES = Object.freeze({
-  'check-scripts-symbol-anchors self-test': 26,
+  'check-scripts-symbol-anchors self-test': 29,
 });
 
 // DELETING an entry silences that battery's floor exactly as effectively as
@@ -316,7 +360,7 @@ export function selfTest() {
     execFileSync('git', ['init', '-q'], { cwd: tmp });
     execFileSync('git', ['add', '-A'], { cwd: tmp });
 
-    const { findings, counts } = sweepCorpus(CORPUS, tmp);
+    const { findings, counts, declined } = sweepCorpus(CORPUS, tmp);
     const kinds = findings.map((f) => f.kind);
     const count = (k) => kinds.filter((x) => x === k).length;
 
@@ -338,6 +382,19 @@ export function selfTest() {
       'a citation naming no tracked file must not be a finding under judgeUntrackedLineAnchors: false');
     check(counts.unresolvableLineCitation === 2,
       `both declined citations must be SEEN and counted, got ${counts.unresolvableLineCitation}`);
+    // ...and ENUMERATED, not merely counted (#15809). A count says a residual
+    // exists without saying where, so nobody can work it down and a residual
+    // that MOVED reads identically to one that shrank. `--list-unresolvable`
+    // prints this array; these three cases are what keep it equal to the
+    // counter rather than a second, drifting instrument.
+    check(declined.length === counts.unresolvableLineCitation,
+      `the declined ENUMERATION must equal the declined COUNT, got ${declined.length} vs ${counts.unresolvableLineCitation}`);
+    check(declined.every((d) => d.doc.includes('declined.mjs') && Number.isInteger(d.line) && d.raw),
+      'every declined row must carry the file, the line and the citation text it was declined for');
+    check(
+      declined.map((d) => d.shape).sort().join(',') === 'bare-filename,directory-qualified',
+      `the declined rows must be classified by SHAPE, got ${declined.map((d) => d.shape).sort().join(',')}`,
+    );
 
     // 2. The allowance mechanism, in both directions, against the fixture.
     const fake = [{ file: 'scripts/bad.mjs', dated: '2026-01-01', heldBy: 'PR #1', why: 'fixture' }];
@@ -462,6 +519,7 @@ if (isEntrypoint(import.meta.url)) {
       );
       process.exit(1);
     }
-  } else if (process.argv.includes('--list')) list();
+  } else if (process.argv.includes('--list-unresolvable')) listUnresolvable();
+  else if (process.argv.includes('--list')) list();
   else runCheck();
 }
diff --git a/scripts/check-settings-bind-window.mjs b/scripts/check-settings-bind-window.mjs
index e813f6563e..e5fc5ec5ea 100644
--- a/scripts/check-settings-bind-window.mjs
+++ b/scripts/check-settings-bind-window.mjs
@@ -91,9 +91,11 @@
  *
  * Nested function bodies are deliberately NOT entered: a closure defined inside
  * a hook and handed to a collaborator runs when that collaborator calls it, not
- * during the hook. `plugin-audit`'s `getLocale` (`audit-plugin.ts:200`) is the
- * measured case — it is passed to `installAuditWriters` and invoked from
- * `resolveWriteLocale` on CRUD writes (`audit-writers.ts:781`), i.e. long after
+ * during the hook. `plugin-audit`'s `getLocale`
+ * (`packages/plugins/plugin-audit/src/audit-plugin.ts`, line 200 as measured)
+ * is the measured case — it is passed to `installAuditWriters` and invoked from
+ * `packages/plugins/plugin-audit/src/audit-writers.ts#resolveWriteLocale` on
+ * CRUD writes (line 781 as measured), i.e. long after
  * the window closed. `packages/rest`'s `settingsServiceProvider` and
  * `ObjectQLPlugin`'s `getSettings` are the same shape.
  *
diff --git a/scripts/check-tenant-audit-census.mjs b/scripts/check-tenant-audit-census.mjs
index d32d4a2435..6eaa604ecb 100644
--- a/scripts/check-tenant-audit-census.mjs
+++ b/scripts/check-tenant-audit-census.mjs
@@ -75,8 +75,9 @@
  * calls subtracted -- is present, dated, and NOT compared.
  *
  * Adopted by maintainer ruling on 2026-08-31 (option A), following the precedent
- * the sibling `check-system-context-census.mjs` set at `:77-84`. The full ruling,
- * the falsified alternative at `:30`, and the measurement that draws the line
+ * the sibling `scripts/check-system-context-census.mjs#UNENFORCED_TEXT_COUNTS`
+ * set (lines 77-84 there as measured). The full ruling, the falsified
+ * alternative (line 30 there, the ⭐ line quoted below), and the measurement that draws the line
  * where it is drawn are all at `UNENFORCED_SCALE_ROWS` below.
  *
  * Check B is not redundant with A. The generated region cannot lie, because it
@@ -419,7 +420,8 @@ export const PROSE_COUNTS = [
  * ## ⛔ What was NOT adopted, and why it is closed
  *
  * Asserting against a RECORDED BASELINE SHA instead of the live tree was proposed
- * and is falsified. The sibling's `:30` refutes it in one line:
+ * and is falsified. `scripts/check-system-context-census.mjs` refutes it in one
+ * line (line 30 there as measured):
  *
  *   > *"a gate that only checks what the page already says can never find what
  *   > the page failed to say."*
diff --git a/scripts/check-workflow-status-functions.mjs b/scripts/check-workflow-status-functions.mjs
index 0fa4ced881..25d4175a34 100644
--- a/scripts/check-workflow-status-functions.mjs
+++ b/scripts/check-workflow-status-functions.mjs
@@ -373,7 +373,8 @@ function selfTest() {
   try {
     // ── 1. A violating sample must go red ────────────────────────────────────
     //
-    // Verbatim shape of publish-smoke.yml:98 as #5343 found it.
+    // Verbatim shape of `.github/workflows/publish-smoke.yml` as #5343 found
+    // it, at line 98 then.
     battery('1. A violating sample must go red');
     const violating = makeRoot({
       '.github/workflows/publish-smoke.yml': `name: Publish Smoke
diff --git a/scripts/doc-line-anchors.mjs b/scripts/doc-line-anchors.mjs
index 3f6e6aeb51..e9aac66c0c 100644
--- a/scripts/doc-line-anchors.mjs
+++ b/scripts/doc-line-anchors.mjs
@@ -4,7 +4,7 @@
 /**
  * doc-line-anchors -- the ONE reader for `file:line` anchors written in docs prose.
  *
- * A docs page that cites source by `` `security-plugin.ts:1560` `` has created a
+ * A docs page that cites source by `` `<file>.ts:1560` `` has created a
  * two-sided invariant with no owner: the line lives in one tree, the citation in
  * another, and nothing relates them. Measured on
  * `content/docs/permissions/system-context.mdx` over 19 days, **101 of its 111
@@ -20,16 +20,21 @@
  *
  * A naive reader finds only the first and undercounts by a third.
  *
- *   FULL          `` `objectql/src/engine.ts:10501` ``   path + line
- *   CONTINUATION  `` `:1409` ``                          line only; the file is
- *                                                        the nearest FULL anchor
- *                                                        to its left, which is
- *                                                        how a row cites four
- *                                                        sites in one file
- *   RANGE_END     `` `2630` ``                           a bare number whose only
- *                                                        separation from the
- *                                                        anchor on its left is a
- *                                                        dash: `:2483`--`2630`
+ *   FULL          `` `<dir>/<file>.ts:10501` ``        path + line
+ *   CONTINUATION  a backticked `:` and a line number,   line only; the file is
+ *                 `1409`                                the nearest FULL anchor
+ *                                                       to its left, which is
+ *                                                       how a row cites four
+ *                                                       sites in one file
+ *   RANGE_END     `` `2630` ``                          a bare number whose only
+ *                                                       separation from the
+ *                                                       anchor on its left is a
+ *                                                       dash -- a continuation
+ *                                                       for `2483`, then `2630`
+ *
+ * ⚠️ The angle brackets and the spelled-out continuations are deliberate: this
+ * file is itself inside the `scripts/**` symbol-anchor corpus, so an
+ * illustration written anchor-shaped would be a citation of its own.
  *
  * The measurement that makes this list non-negotiable: `grep -c` for anchor-shaped
  * text over the previous edition of that page answered **64**, because it counts
@@ -39,7 +44,7 @@
  * ## Resolution is by unique suffix, and ambiguity is an ERROR
  *
  * An anchor names as little of the path as it can and still be unique --
- * `read-audit.ts:556` where the basename is unique, `objectql/src/engine.ts:10501`
+ * `<file>.ts:556` where the basename is unique, `<dir>/<file>.ts:10501`
  * where it is not. That is a property a gate can hold: resolve the spelling
  * against the tracked file list by path-suffix and REFUSE when two files match.
  * The previous edition had **41 of 111** anchors whose bare basename matched two
@@ -60,11 +65,11 @@ import { extname } from 'node:path';
 /** Extensions an anchor may name. Anything else is prose, not a citation. */
 export const ANCHOR_EXTENSIONS = ['.ts', '.tsx', '.mts', '.cts', '.mjs', '.cjs', '.js', '.jsx'];
 
-/** `path/to/file.ts:1234` -- the whole span, nothing else. */
+/** `<dir>/<file>.ts:1234` -- the whole span, nothing else. */
 const FULL_RE = /^([A-Za-z0-9_.\-/]+)\.([a-z]+):(\d+)$/;
-/** `path/to/file.ts` -- a cited file with no line. */
+/** `<dir>/<file>.ts` -- a cited file with no line. */
 const PATH_ONLY_RE = /^([A-Za-z0-9_.\-/]+)\.([a-z]+)$/;
-/** `:1409` -- a continuation of the anchor to its left. */
+/** A backticked `:` and a line number, `1409` -- a continuation of the anchor to its left. */
 const CONTINUATION_RE = /^:(\d+)$/;
 /** `2630` -- a bare number; only an anchor when a dash joins it to one. */
 const BARE_NUMBER_RE = /^(\d+)$/;
@@ -295,8 +300,9 @@ export function blankAnchorLineNumbers(rawText) {
     if (a.start < cursor) continue;
     out += rawText.slice(cursor, a.start);
     // The LAST run of digits in the span is the line number in all three shapes --
-    // `path.ts:1234`, `:1234` and a bare `1234` -- and a path that itself carries
-    // digits (`0112-codes.ts:55`) keeps them, because the lookahead requires that
+    // `<file>.ts:1234`, a backticked `:` and `1234`, and a bare `1234` -- and a
+    // path that itself carries digits (`0112-<name>.ts:55`) keeps them, because
+    // the lookahead requires that
     // nothing but non-digits follows.
     out += rawText.slice(a.start, a.end).replace(/\d+(?=\D*$)/, '#');
     cursor = a.end;
diff --git a/scripts/objectui-changeset-digest.mjs b/scripts/objectui-changeset-digest.mjs
index bd4aaf9d2e..f51a5a29ff 100644
--- a/scripts/objectui-changeset-digest.mjs
+++ b/scripts/objectui-changeset-digest.mjs
@@ -2153,7 +2153,8 @@ function selfTest() {
     );
 
     // --- #6494: the ADR-0087 disposition scaffold ---------------------------
-    // The gate's own marker pattern (check-adr-0087-registration.mjs:416),
+    // The gate's own marker pattern (`scripts/check-adr-0087-registration.mjs`,
+    // line 416 as measured),
     // copied rather than imported: that module's CLI dispatch is top-level, so
     // importing it would RUN the gate. The copy is belt-and-braces — the REAL
     // gate binary judges a real artifact in the round trip below, and that, not
diff --git a/scripts/pr-labels.mjs b/scripts/pr-labels.mjs
index c6b9cab431..0335425712 100644
--- a/scripts/pr-labels.mjs
+++ b/scripts/pr-labels.mjs
@@ -21,13 +21,14 @@
  * Both label writers this file replaces did exactly that, read out of their
  * pinned sources rather than inferred from their docs:
  *
- *   * codelytv/pr-size-labeler@v1.10.4 -- src/github.sh:68-91
- *     (`github::add_label_to_pr`): GETs the PR, greps its OWN size family out
+ *   * codelytv/pr-size-labeler@v1.10.4 -- `src/github.sh` in THAT repo, lines
+ *     68-91 as pinned (`github::add_label_to_pr`): GETs the PR, greps its OWN size family out
  *     of the result, appends the new size label, then
  *     `curl -X PUT .../issues/$pr_number/labels` with the whole set. No
  *     mitigation of any kind: the window is the entire round trip.
- *   * actions/labeler@v7.0.0 -- src/labeler.ts:56,111-133 plus
- *     src/api/set-labels.ts: snapshots `preexistingLabels` at run start, unions
+ *   * actions/labeler@v7.0.0 -- `src/labeler.ts` in THAT repo, lines 56 and
+ *     111-133 as pinned, plus its
+ *     `src/api/set-labels.ts`: snapshots `preexistingLabels` at run start, unions
  *     in the config matches, re-reads the live label list once and carries
  *     forward whatever appeared in between, then calls
  *     `client.rest.issues.setLabels` -- which IS the PUT. The re-read NARROWS
@@ -67,12 +68,14 @@
  *
  * ## What is deliberately preserved from the retired actions
  *
- *   * The size buckets compare with `<`, NOT `<=` -- `labeler.sh:50-60` uses
+ *   * The size buckets compare with `<`, NOT `<=` -- codelytv's `src/labeler.sh`,
+ *     lines 50-60 as pinned, uses
  *     `[ "$total" -lt "$max" ]`. A 10-line PR is `size/s`, not `size/xs`. The
  *     thresholds arrive in the same env names the action's inputs used, so the
  *     workflow diff is auditable value-for-value.
  *   * `files_to_ignore` is a space-separated list matched against the WHOLE
- *     path, the way `[[ $filename == $pattern ]]` did in github.sh:36.
+ *     path, the way `[[ $filename == $pattern ]]` did in codelytv's
+ *     `src/github.sh`, line 36 as pinned.
  *   * Path labels are never removed, matching `sync-labels: false`. The path
  *     half issues POST only -- it has no DELETE at all.
  *   * The size family IS owned by this writer, so a stale `size/*` is removed
@@ -80,7 +83,8 @@
  *     thing by grepping the family out of its PUT payload; the difference is
  *     that its version also carried -- and could drop -- every bystander label.
  *
- * One behaviour deliberately DIVERGES: github.sh:23 reads
+ * One behaviour deliberately DIVERGES: codelytv's `src/github.sh`, line 23 as
+ * pinned, reads
  * `pulls/{n}/files?per_page=100` and never paginates ("NOTE: this code is not
  * resilient to changes w/ > 100 files"), so a 400-file PR was sized off its
  * first 100 files. This paginates. A PR over 100 files may therefore get a
@@ -196,7 +200,8 @@ export function matchGlob(pattern, filePath) {
 }
 
 // ---------------------------------------------------------------------------
-// Size bucketing -- codelytv/pr-size-labeler src/labeler.sh:50-60, verbatim.
+// Size bucketing -- codelytv/pr-size-labeler `src/labeler.sh`, lines 50-60 as
+// pinned, verbatim.
 // ---------------------------------------------------------------------------
 
 /**
@@ -216,7 +221,8 @@ export function sizeLabelFor(total, buckets) {
   return fallthrough.label;
 }
 
-/** Total modifications, mirroring github.sh:5-52 with pagination added. */
+/** Total modifications, mirroring codelytv's `src/github.sh` lines 5-52 as
+ *  pinned, with pagination added. */
 export function totalModifications(files, filesToIgnore) {
   let total = 0;
   for (const file of files) {
@@ -719,7 +725,8 @@ function selfTest() {
   const staleRead = [];
   const liveSet = ['skip-changeset'];
 
-  // Transcribed from github.sh:68-91 -- take the read, drop its OWN family,
+  // Transcribed from codelytv's `src/github.sh`, lines 68-91 as pinned -- take
+  // the read, drop its OWN family,
   // append the new size label, PUT the whole thing.
   const retiredPlan = [
     {
diff --git a/scripts/symbol-anchors.mjs b/scripts/symbol-anchors.mjs
index ff836fccbd..2e94f189f3 100644
--- a/scripts/symbol-anchors.mjs
+++ b/scripts/symbol-anchors.mjs
@@ -32,7 +32,9 @@
  * 16,000-line file has an expected lifetime measured in days.
  *
  * ⭐ And the rot can hide a SEMANTIC INVERSION. ADR-0113 cited
- * `sql-driver.ts:4901` for `if (field.required) col.notNullable()`; the
+ * `packages/drivers/driver-sql/src/sql-driver.ts` (line 4901 as the ADR wrote
+ * it -- a dated reading, ⛔ not a pointer) for `if (field.required)
+ * col.notNullable()`; the
  * mechanism moved and now keys off `storage.notNull` -- the opposite predicate.
  * With the anchor rotted, the ADR's pre-decision Context row reads to a fresh
  * reader as a description of today. That is carded separately as #14193 and is
@@ -96,8 +98,11 @@
  *   (c) READS NATURALLY. `#symbol` is the fragment syntax a reader already
  *       knows from URLs, and it survives copy-paste into a GitHub link.
  *
- * ⛔ A LINE NUMBER IS NOT AN ANCHOR FORM. `file.ts:4901`, `file.ts:341-400`,
- * `file.ts:459–463` (en dash) and a bare continuation `:2933` are all findings.
+ * ⛔ A LINE NUMBER IS NOT AN ANCHOR FORM. `<file>.ts:4901`, `<file>.ts:341-400`,
+ * `<file>.ts:459–463` (en dash) and a bare continuation -- a backticked `:`
+ * followed by a line spec, `2933` -- are all findings. ⚠️ The angle brackets
+ * are deliberate: this file is itself inside the `scripts/**` corpus, so an
+ * illustration written path-shaped would be a citation of its own.
  *
  * ### The one escape hatch, and who may use it
  *
@@ -209,10 +214,10 @@ const CONTINUATION_SPAN = new RegExp('`#(?<symbol>' + SYMBOL + ')`', 'g');
 /* ⛔ The forms the migration deleted. Matched on the SAME text the anchor
  * extractor sees, so a line number cannot hide behind a spelling the extractor
  * normalises away. Both the full form and the bare continuation. */
-/* ⭐ A LINE SPEC is not just `:123`, and assuming it was is how a gate
- * false-greens. The #13556 corpus spells one position NINE ways, and the
- * census's own extractor missed the comma forms exactly as a naive rule here
- * would:
+/* ⭐ A LINE SPEC is not just a colon and one run of digits, and assuming it was
+ * is how a gate false-greens. The #13556 corpus spells one position NINE ways,
+ * and the census's own extractor missed the comma forms exactly as a naive rule
+ * here would:
  *
  *     :4901        :341-400      :459–463 (en dash)     :2214+
  *     :610,1028    :13,346-389   :29-39,147-152         :2956/2991
@@ -224,18 +229,19 @@ const CONTINUATION_SPAN = new RegExp('`#(?<symbol>' + SYMBOL + ')`', 'g');
  * version matched only a span that was EXACTLY an anchor, and it called a
  * corpus still carrying rot clean — three ways at once: a bare anchor in
  * running prose, an anchor sharing a span with other text
- * (`` `Builder.io SDK: packages/sdks/src/types/builder-block.ts:42` ``), and
+ * (`` `Builder.io SDK: <dir>/<file>.ts:42` ``), and
  * that slash list. A line number is rot wherever it is written; the only thing
  * excluded is a FENCED block, which is quoted material, not an anchor. */
 const LINE_SPEC = '\\d+(?:\\s*[-–—]\\s*\\d+)?(?:\\s*[,/]\\s*\\d+(?:\\s*[-–—]\\s*\\d+)?)*\\+?';
 const LINE_ANCHOR = new RegExp('(?<![\\w/.-])(' + PATHISH + '):(' + LINE_SPEC + ')(?![\\w-])', 'g');
 const LINE_CONTINUATION = new RegExp('`\\s*[:,]\\s*(' + LINE_SPEC + ')`', 'g');
 
-/* `// packages/foo/bar.ts:378` — a comment that is nothing but a path. */
+/* `// <dir>/<file>.ts:378` — a comment that is nothing but a path. */
 const FENCED_HEADER = new RegExp('^\\s*(?://|#|\\*|/\\*)\\s*' + PATHISH + ':\\d');
 
 /* ⭐ A ninth spelling, and the census counted none of them: a bare backticked
- * number, tilde-prefixed — `` ~`326` `` for "about line 326". It carries no
+ * number, tilde-prefixed — a `~` then a backticked line number, `` ~`NNN` ``,
+ * written for "about line 326". It carries no
  * path, so it cannot be resolved to anything; ADR-0056 alone held 13.
  *
  * ⚠️ Only the TILDE form is judged. A bare `` `403` `` or `` `4096` `` is an
@@ -370,7 +376,7 @@ export function extractAnchors(markdown) {
      * and the census counted those among its 343 — so skipping fences wholesale
      * would leave five rotted anchors behind. Everything else in a fence is
      * quoted material (sample code, CI output) and is NOT judged: an example
-     * that happens to contain `foo.ts:12` is not an anchor. */
+     * that happens to contain `<file>.ts:12` is not an anchor. */
     if (inFence) {
       if (!FENCED_HEADER.test(text)) return;
       for (const m of text.matchAll(LINE_ANCHOR)) {
@@ -450,7 +456,7 @@ export function extractAnchors(markdown) {
  * corpus unusable. Measured over the 216 tracked `scripts/**` `.mjs` files on
  * `5315098df`: 251 live line citations raw against 128 through this projection.
  * The 123 that vanish are not rot -- they are a gate's own self-test FIXTURES,
- * string literals like `'p.ts:2'` and `'content/docs/other.mdx:1'` written to
+ * string literals like `'<file>.ts:2'` and `'<dir>/<file>.mdx:1'` written to
  * provoke that gate's own line-reporting. Judging those would red a gate for
  * testing itself, which is the fabrication direction
  * `scripts/js-comment-mask.mjs` exists to close.
@@ -523,7 +529,26 @@ function trackedFiles(root) {
 }
 
 /**
- * Sweep one corpus. Returns findings and the counts a report needs.
+ * The SHAPE of a declined citation -- why no resolver could bind it. Kept here
+ * rather than in a corpus so every corpus that waives citations reports the
+ * same four words, and a residual list can be grouped without re-deriving the
+ * classification from the raw text.
+ *
+ * Order is load-bearing: a continuation and a tilde form BOTH inherit whatever
+ * path preceded them on the line, so they must be named by how they were
+ * written, never by the path they borrowed.
+ */
+export function declinedShape(la) {
+  if (la.tilde) return 'tilde';
+  if (la.continuation) return 'continuation';
+  if (la.path?.includes('/')) return 'directory-qualified';
+  return 'bare-filename';
+}
+
+/**
+ * Sweep one corpus. Returns findings, the counts a report needs, and the
+ * citations a corpus declined to judge (`declined`, empty unless the corpus
+ * sets `judgeUntrackedLineAnchors: false`).
  *
  * Finding kinds:
  *   line-anchor        a `path:NNN` survived the migration                (RED)
@@ -535,6 +560,7 @@ function trackedFiles(root) {
 export function sweepCorpus(corpus, root = process.cwd()) {
   const tracked = trackedFiles(root);
   const findings = [];
+  const declined = [];
   const counts = { docs: 0, anchors: 0, symbol: 0, fileLevel: 0, declaration: 0, literal: 0, crossRepo: 0, exempt: 0, continuation: 0, unresolvableLineCitation: 0 };
   const sourceCache = new Map();
   const readTarget = (p) => {
@@ -560,19 +586,26 @@ export function sweepCorpus(corpus, root = process.cwd()) {
        * and a gate whose only remedy is "stop writing that" is the
        * permanently-red gate this repo retired.
        *
-       * Measured on `5315098df` over `scripts/**` `.mjs` comment prose: 128
-       * live citations in all, of which 32 name a tracked file and 96 do not --
-       * 66 bare filenames (`engine.ts:9407`, an abbreviation inside a census
-       * table that no resolver can bind to one of this tree's several
-       * `engine.ts`), 22 continuations inheriting no path of their own, 11
-       * directory-qualified paths that are illustrations or sibling-repo files
-       * (`path/to/file.ts:1234`, `src/github.sh:68-91`), and 1 tilde form. The
-       * 96 are a real defect class and are recorded as a follow-up, exactly as
-       * the 1,056 bare paths under `checkBarePaths` were -- but they are not
-       * the cross-file rot #15765 measured, and folding them in would bury this
-       * gate's signal under a cleanup nobody ruled on. */
+       * Measured on `scripts/symbol-anchors.mjs` over `scripts/**` `.mjs`
+       * comment prose, on the date recorded in `check-scripts-symbol-anchors`'s
+       * own census constant: 128 live citations in all, of which 32 named a
+       * tracked file and 96 did not -- bare filenames (an abbreviation inside a
+       * census table that no resolver can bind to one of this tree's several
+       * files of that name), continuations inheriting no path of their own,
+       * directory-qualified paths that are illustrations or sibling-repo files,
+       * and one tilde form. That population is a real defect class and was
+       * recorded as a follow-up (#15809), exactly as the 1,056 bare paths under
+       * `checkBarePaths` were -- but it is not the cross-file rot #15765
+       * measured, and folding it in would bury this gate's signal under a
+       * cleanup nobody ruled on.
+       *
+       * ⭐ Declining to JUDGE is not declining to SEE. Every declined citation
+       * is counted AND recorded in `declined`, so a corpus can enumerate what
+       * it waived (`--list-unresolvable`) and the residual stays a list rather
+       * than a number nobody can act on. */
       if (!corpus.judgeUntrackedLineAnchors && !tracked.has(la.path ?? '')) {
         counts.unresolvableLineCitation += 1;
+        declined.push({ doc: rel, line: la.line, raw: la.raw, path: la.path ?? null, shape: declinedShape(la) });
         continue;
       }
       /* A marker whose CLASS is unrecognised is its own finding, never a
@@ -645,7 +678,7 @@ export function sweepCorpus(corpus, root = process.cwd()) {
       counts[cls] += 1;
     }
   }
-  return { findings, counts };
+  return { findings, counts, declined };
 }
 
 export function formatFindings(findings) {
@@ -678,8 +711,9 @@ function assert(cond, msg) { if (!cond) { console.error(`❌ symbol-anchors --se
 // The count is a FLOOR, not an equality — adding cases is ordinary work and must
 // not red. A battery BELOW its floor means cases stopped running; the remedy is
 // to find what stopped registering.
+// 63 → 67 when `declinedShape` gained a case per arm (#15809).
 const SELF_TEST_BATTERIES = Object.freeze({
-  'symbol-anchors self-test': 63,
+  'symbol-anchors self-test': 67,
 });
 
 // DELETING an entry silences that battery's floor exactly as effectively as
@@ -885,6 +919,18 @@ export function selfTest() {
   try { defineCorpus({ id: 'x', label: 'x', docRoots: ['a'], docProjection: 'commentProse' }); } catch { projThrew = true; }
   check(projThrew, 'defineCorpus must refuse a docProjection that is not callable — a skipped projection sweeps raw source while reading as though it did not');
 
+  // 11. ⭐ The declined SHAPE, whose ordering is the whole rule (#15809): a
+  //     continuation and a tilde form inherit whatever path preceded them on
+  //     the line, so classifying by the path first would file them under the
+  //     borrowed path and a residual list would name a file the author never
+  //     wrote. All four arms, in one case each.
+  check(declinedShape({ tilde: true, path: 'a/b.ts' }) === 'tilde',
+    'a tilde form is a TILDE even when a path preceded it on the line — it borrowed that path, it did not cite it');
+  check(declinedShape({ continuation: true, path: 'a/b.ts' }) === 'continuation',
+    'a continuation is a CONTINUATION even when a path preceded it on the line');
+  check(declinedShape({ path: 'a/b.ts' }) === 'directory-qualified', 'a path with a slash is directory-qualified');
+  check(declinedShape({ path: 'b.ts' }) === 'bare-filename', 'a path with no slash is a bare filename — the shape no resolver can bind');
+
   // ── The floor: every declared battery RAN, and ran its cases (#13489) ────
   //
   // Evaluated after every battery has had its chance and BEFORE the verdict, so