Remove the extracted lithoglyph tree — it existed in two places

[hyperpolymath]

Closes the defect the extraction existed to fix. hyperpolymath/lithoglyph#4 merged on 2026-07-21, but the 819 files it imported were left here as well, so the whole database existed twice with no canonical copy.

1878 → 1058 tracked files. lithoglyph/ is now a pointer README.

Nothing is lost, and that is checkable

Every removed file is preserved at its original path in the annotated tag split-history/lithoglyph, which is on origin:

$ git ls-tree -r split-history/lithoglyph --name-only | wc -l
819
$ git checkout split-history/lithoglyph -- <path>   # recover anything

⚠️ Never prune that tag, the _split_lithoglyph branch, or their glyphbase/gnpl equivalents. The imports into the standalone repos were squash merges, so these refs hold the only granular history that exists.

Destinations verified live before removing: lithoglyph 700KB, glyphbase 238KB, gnpl 330KB — all populated. The nested lithoglyph/glyphbase/ (130 files) and lithoglyph/gql-dt/ (125) were confirmed byte-identical to their own repos during the #4 import. lithoglyph/core-factor/gql* was superseded rather than migrated (see #56) and is not "pending".

Documentation corrected, rather than the code bent to fit it

REGISTRY.adoc claimed "verisimdb and lithoglyph are git submodules of this coordination repo". That was never true of either — git ls-files --stage shows both as ordinary tracked files, not gitlinks, and no .gitmodules has ever existed here.

I was going to add submodules to make the sentence true. That would force submodules: recursive into every workflow touching the tree — real cost for a coordination repo that only needs to point at its members. The sentence now says what the repo does. verisimdb/ is still 714 tracked files awaiting the same extraction.

CLAUDE.md's transitional note listed lithoglyph/ among directories "being extracted". It's done; the note says so and points at the tag.

placement-guard: lithoglyph/ promoted from warning to error

The GRANDFATHER list exempts legacy database directories from failing the guard — they warn instead — because they held content not yet moved. lithoglyph/ no longer does.

Leaving it grandfathered would mean a new file there raised only a warning, letting the duplication this PR removes grow back silently. It's out of the list, so it now fails. Move each directory out as its extraction completes.

Also: .github/workflows/instant-sync.yml removed

Malformed YAML on origin/main before this branch — expected <block end>, but found '-' at line 15 — so it produced a startup_failure on every push and no check run. It is also one of the two workflows rsr-template-repo deleted upstream in 09f94c5. Dead and unparseable.

All 12 remaining workflows parse.

Related: the stuck rebase and the mangled remote

Both resolved outside this PR, on the local checkout:

• Origin was mangled to git@github.com:hyperpolymath/lithoglyph-reassembly.git — a repo that does not exist. That was a worktree name leaking into .git/config, which worktrees share with the main checkout. Reset to nextgen-databases (HTTPS — SSH port 22 is unreachable from this host).
• An interactive rebase was stuck mid-flight, conflicted on lithoglyph/MOVED.md, with three "Mistral Vibe" sweeps queued. Aborted, and main reset to origin/main. All four commits are preserved on backup/sweeps-mistral-vibe-nextgen.

The sweeps were re-measured rather than trusted, and each confirmed the earlier ruling to drop them: sweep2 adds a mise.toml with 31 latest pins; sweep3 adds a 661-line AGPL-3.0 text to an MPL-2.0 repo; sweep4 has real files but names metadatastician as owner/funding in a hyperpolymath repo, plus an ARCHITECTURE.md describing a layout this project doesn't have. 2393737 is superseded by #56.

🤖 Generated with Claude Code

---

CI: 3 failures, all pre-existing — verified, not assumed

governance / Language / package anti-pattern policy, Validate K9 contracts and Validate A2ML manifests all fail on this branch. All three already fail on main — checked directly with gh run list --branch main — so none is a regression from this PR.

The anti-pattern gate is the interesting one, and this PR substantially improves it without closing it:

.res files tracked	count
origin/main (before)	140
this branch (after)	44

96 of the 140 were the duplicated Lithoglyph copy. What remains is verisimdb/ (40) and typeql-experimental/ (4) — both still-pending extractions, both out of scope here. Also remaining: 0 .ts/.tsx, 1 package.json.

So the gate stays red, but for a 69%-smaller and correctly-attributed reason. It closes when verisimdb/ gets the same treatment this PR gives lithoglyph/.

The log also shows the gate's own AffineScript drift-check is inert on the runner — ::notice::affinescript compiler unavailable on runner — skipping drift check — which is the same affinescript#56 blocker recorded in RESCRIPT-TS-AFFINESCRIPT-MIGRATION.adoc.