docs(tooling): rewrite metadata's DEBT note to its measured composition

[claude]

Closes#10723

DEBT['@objectstack/metadata'] recorded errors: 89 while its own note itemised
34 + 24 + 34 = 92 and stated "Re-measured 92 at 5ab0842, up from 87". The entry was
evidently lowered 92 -> 89 without the note being rewritten alongside it — the same class
PR #10721 repaired for @objectstack/service-automation an hour earlier.

Prose only. errors: 89 is unchanged — no count moves.

Measured, not adopted

Re-derived with the gate's own invocation (tscErrorCount()'s), from the repo root
against a built dependency closure, at base 4b84834a32:

$ ./node_modules/.bin/tsc --noEmit --pretty false -p packages/metadata/tsconfig.json

Code	Count	Tier
TS7006	33	noise
TS2345	30	code
TS2835	25	config
TS6133	1	noise
Total	89

89 exactly — matching the field.TS2322, which the old note itemised at x4, is
measurably absent (0 occurrences in the output).

The measurement is valid, not a silent zero: the output carries none of the codes in
the gate's own TSC_SETUP_ERROR set (TS5058/5083/6053/18003/5012/2688), and tsc's exit 2
came with 89 recognisable diagnostics, so the gate's "refusing to record 0" guard does not
fire.

What moved — and it is fully attributable

Against the composition the note itself recorded at 92:

Tier	Note said (92)	Measured (89)	Delta
code	34 (TS2345 x30, TS2322 x4)	30 (TS2345 x30)	-4
config	24 (TS2835)	25 (TS2835)	+1
noise	34 (TS7006 x33, TS6133)	34 (TS7006 x33, TS6133)	0

-4 +1 +0 = -3, and 92 - 3 = 89. The delta closes exactly, so the new note states the
attribution rather than falling back on the file's own escape hatch at :167-173 ("when
the delta cannot be attributed, say so in the note rather than inventing composition").

Read across the entry's history, the code tier is the only churning one: TS2353 gave way
to TS2322, and TS2322 has now gone too. TS2345 x30 is its only lasting resident.

Two inherited claims re-derived — one of which never held

Per the dispatch, neither the tier split nor the per-file claim was carried forward
unchecked.

• The per-file counts survive exactly.metadata.test.ts = 34 and
  register-notifies-watchers.test.ts = 16 today, the same two numbers the note carried at
  92. The whole -3 landed in the other ten files.
• The fraction did not. 34 + 16 = 50 of 89 is 56% — over half, not the "two thirds"
  the note claimed. It was not true at 92 either (50/92 = 54%), so this was a wrong claim
  at both endpoints rather than a stale one. The rewrite says "over HALF" and names both
  numbers it was wrong at.

The residue restated as what it actually is

The 89 is three mechanical repairs, not 89 problems:

• All 30 TS2345 are one defect thirty times over — metadata.test.ts between lines 608
  and 945, every one the same mock PluginContext object literal missing
  registerServiceFactory and getServiceScoped. Verified rather than assumed: 30/30 of
  those diagnostics carry the identical "is not assignable to parameter of type
  'PluginContext'" head and the identical "missing the following properties ...
  registerServiceFactory, getServiceScoped" continuation. One shared fixture closes the
  code tier outright.
• The 25 TS2835 are the widest spread — 12 files — and are one codemod: a relative
  import wanting an explicit .js extension under node16 resolution.
• The remaining 34 are noise: implicit-any parameters, plus one unused import.

Mutation verified, not assumed

This is a prose edit inside a multi-line string concatenation, exactly where a zero-match
edit reads as a clean no-op. Repo-wide phrase counts across scripts/, before -> after:

Phrase	Before	After
Re-measured 92 at 5ab08428	1	0
code-tier 34 (TS2345 x30, TS2322 x4)	1	0
Two thirds of the pile	1	0
Re-measured 89 at 4b84834a32	0	1
Re-measured (every ledger note)	30	30
ledger entries named @objectstack/metadata...	13	13

Re-measured holding at 30 is the neighbour check — one occurrence was swapped for one, so
no other entry's note moved. The replacement itself asserted a match count of exactly 1
before writing. git diff --stat is 1 file, +12/-4, and no errors: line appears in the
diff at all.

Deliberately not in this PR

• The general rule at :167-173 speaks only about counts going up; this entry drifted
  down. Extending that wording, and adding a note-vs-field consistency check, are [finding] nothing reads a DEBT/TEST_DEBT note, and --lower rewrites only the digits — so the advertised way to lower an entry desynchronises its prose by construction #10722
  — which names the mechanical cause: lowerLedgerEntries() rewrites only the errors:
  digits and never the note, so --lower mints a fresh instance of this class every time it
  runs.
• [finding] A package can pass check:type-check-coverage with a whole source directory outside tsc — packages/objectql/scripts/** is invisible to its own typecheck #10756 is open against this same file about ledger completeness. Different aspect,
  untouched here.
• No other DEBT or TEST_DEBT entry is touched.

Gates

Derived with node scripts/pm/dispatch-gates.mjs (no paths — merge base 4b84834a3,
1 committed path, 0 working-tree, 0 untracked), then run at 464fe41468:

Gate	Result
check:type-check-coverage	OK — 64/77 packages type-checked, 13 in DEBT; self-test 23+24+25+28+9 cases hold
check:type-check-debt (--re-measure)	OK — 33 ledger entries re-measured in 277.1s, 1912 raw errors, none above its recorded number
check:cross-package-test-inputs	OK — 13 packages read outside themselves, all declared; 90 self-test cases passed
check:nul-bytes	OK — 6235 text files scanned, no raw ASCII control bytes

The re-measure half refused correctly on the first attempt — "43 workspace
dependenc(ies) of the ledgered packages have no built type entry point on disk ... measuring
now would not fail, it would silently measure a DIFFERENT WORLD" — and was re-run after
pnpm exec turbo run build --filter='./packages/*' --filter='./packages/*/*' (70/70
successful). That refusal is the evidence the number came from a real closure.

Its one informational line is @objectstack/plugin-auth TEST_DEBT 109 recorded vs 97
measured — pre-existing, unrelated to this card, explicitly not an error, and already owned
by #6376 / #10615. @objectstack/metadata is not among the surplus entries, which is
the gate's own independent confirmation that it measures its recorded 89 exactly.

No changeset: gate-ledger prose publishes nothing, so this carries skip-changeset.

---

Generated by Claude Code