docs(deed): correct the descriptor census in the mapping doc

1-formats/deed/mappings/launcher-standard-to-praxis-deed.adoc

@@ -361,16 +361,62 @@ descriptors under `metadatastician/` and `_ESOTERIC_GROUP _SET/`. These are
 per-application *instances* of the metadata block, a different family from this
 standard, and they need their own spec before anything touches them.
 
-⚠ This was first recorded as **10 files**. That is an undercount by more than
-5×. Measured 2026-09-22 (`rg --files -g '*.launcher.a2ml'`, excluding
-`.claude/worktrees/` and `developer/worktrees/`): **23 distinct descriptors
-across 52 on-disk paths** — the path count exceeds the descriptor count because
-several repos are cloned more than once — and **39 of those paths cite the
-deleted `launcher-standard.a2ml`**. Two of the 23 (`_DATABASE`,
-`_ESOTERIC_GROUP`) are set-level rather than per-application, which is why an
-app-name census returns 21. The figure decays as clones are added, so it is
-dated rather than stated as standing fact; `hyperpolymath/standards#960` carries
-the decision on what happens to these descriptors.
+⚠ This was first recorded as **10 files**, which was an undercount.
+
+⚠ **CORRECTED 2026-09-22 — the correction itself was wrong.** This paragraph
+previously read "an undercount by more than 5×… **23 distinct descriptors across
+52 on-disk paths**… **39 of those paths cite the deleted
+`launcher-standard.a2ml`**", and explained the gap against `standards#960`'s own
+figure of 21 by asserting that two set-level directories (`_DATABASE`,
+`_ESOTERIC_GROUP`) inflate the descriptor census. **All three numbers were
+wrong, the multiplier was wrong, and the explanation was a rationalisation of a
+control that was correctly firing.** The sweep was run from `developer/` without
+a scope restriction, so it counted two populations that are not live repositories:
+
+* **9 paths** under `archive/` quarantine snapshots
+  (`_QUARANTINE-2026-09-02`, `_QUARANTINE-2026-09-03-halted-rebase`,
+  `2026-09-08-drain/orphaned-worktrees`);
+* **7 paths** under `tools/opt/cargo/git/checkouts/` (`burble` ×5,
+  `gossamer` ×2) — vendored cargo git checkouts, not clones.
+
+* a further **5 paths** under nested `.claude/worktrees/` — linked worktrees
+  inside the repos themselves, which duplicate every file they contain.
+
+`standards#960`'s figure of **21 was right**; this document's 23 was wrong, and
+so was the set-level explanation offered for the difference. 10 → 21 is roughly
+**2×**, not "more than 5×".
+
+⚠ The worktree exclusion is worth stating precisely, because getting it wrong
+is silent: a ripgrep `-g '!.claude/worktrees/**'` glob **only matches at the
+search root** and excludes nothing when the worktrees are nested inside repos.
+The exclude must be `-g '!**/.claude/worktrees/**'`, and the sweep needs a
+positive control on its own filter.
+
+Re-measured 2026-09-22, scoped to `hyper-repos` and `meta-repos` only, with
+nested worktrees verified absent from the result:
+
+[cols="3,1"]
+|===
+| metric | value
+
+| `*.launcher.a2ml` paths on disk | 31
+| **distinct descriptor basenames** | **21**
+| paths citing the deleted `launcher-standard.a2ml` | 26
+| distinct descriptors citing it | 17
+| **distinct upstream repositories affected** | **18**
+|===
+
+All **66** citing lines are `#` comments, in six distinct shapes — so the
+citation text is fixable independently of the file extension.
+
+The path counts decay as clones and worktrees are added, so they are dated
+rather than stated as standing fact. **The distinct counts do not**, and that
+is the invariant worth trusting: when the worktree exclusion was corrected the
+path figures moved (36→31, 29→26, 71→66) while distinct descriptors stayed 21,
+distinct citing descriptors stayed 17 and distinct upstream repositories stayed
+18 — which is exactly what removing duplicates should do.
+`hyperpolymath/standards#960` carries the decision on what happens to these
+descriptors.
 
 == 8. What this spec does not claim