contract: the page agrees with the tree, the pins get a freshness gate, six baselines committed (#574, #461, #484)

[rowan-claude]

Three lanes of the C++ column's CONTRACTION pass: the pages state what the tree carries, the pins under testdata/ get the freshness gate generated/ has had all along, and the six corpora that keep printing the missing-baseline notice get their baselines.

1. CLAUDE.md states the present tree (#574)

internal/parser/parser.go carries scanner.KwTable as a declaration and docs/SPEC-TABLES.md is a normative page beside docs/SPEC.md, while CLAUDE.md:68-73 said the table layer left the language and the parser refuses table by name. The horizon section now says what the tree does. message and object are still refused by name (parser.go:246, :255) and contexts is refused at file scope, so the protocol bullet keeps its substance and states the frozen projection tokens for what they do today, which is holding every existing unit's id stable. Glenn's scope sentence of 2026-08-25 is verbatim in both places it appears.

2. Page versus tree

Each line measured on main first.

• docs/VERSIONING.md's owed list. The table wire goes uniformly 64-bit before 3.0.0: every id fnv1a64 (fields, variants, arms, tables), every length, count and node index u64 #435, Reserve one escape kind on the table wire: length-prefixed opaque, skip me #434 and Design lock-in before the sweep: the cold read's shortest list #523's message form and unbounded array are ✅ for C++ on ROADMAP.md, so each row now names what is built and what is owed, in the form the tables: the wstring kind on the id-table wire #522 and Retain-unknown: an opt-in that keeps what this build cannot name (ruling 12 on #523) #525 rows already use. The Design lock-in before the sweep: the cold read's shortest list #523 doc-comment row leaves the list: a tag parses at every line kind and the doc and tags descriptor columns are ✅ in all nine ports.
• SPEC-TABLES.md §6.1, the block-identity backend status. It cited schema#301, which is closed on the standing ruling that generated output records no compiler version. The page says the status, says no open issue carries the row, and says why the identity is over the block's own text.
• SPEC-TABLES.md §6.6. TableRetain and the three verbs were listed as OWED to §11's claimed set. internal/tablenames/cpp.go:13, internal/check/check.go:3646 and :3692 claim them and internal/codegen/cpptable/retain.go emits them, so the bullet says they are claimed. What still lands with the Dart port is TableRetain in that backend's library-scope registry.
• SPEC-TABLES.md §6.1, the backend-status paragraph. "Every language carries the table wire" now says which FORM: §3's id-table form is the C++ reference and internal/tablewire, and the eight ports write the one that preceded it (Go: the table wire's id-table form (#435) #511 to Rust: the table wire's id-table form (#435) #518).
• SPEC-TABLES.md §15, the map and list rows. compiler/cook.go:60 and :150 refuse a map or a []T at Cook and Uncook by name, and internal/tablecook/check.go:212 refuses a map slot while the scan carries §7.4's element-array clause. The map row said cook-check's clause was "the one piece still owed"; both rows now state the tool's whole cook status. ROADMAP's maps cell is left alone: it is a per-language cell and C++ does carry maps, so the cell and the page do not disagree.
• Makefile:185 cited a round log the tree no longer holds.

Nothing normative changed.

3. A freshness gate for the pins (#461)

New pins job, the generated job's shape one door over: clone ../serialize at the shared SERIALIZE_TAG, run the tree's own re-pinning targets conformance-pin and update-goldens in place, then fail on a dirty testdata/, tracked or untracked. Naming the targets rather than the files means a rule that starts writing a new pin is covered the moment it joins one of them.

Its negative control is live material rather than a planted byte. On main the job goes red and names six files, testdata/golden/tables/maps/*Table.h, which carry neither the retain walk's arm and enum-reference case nor the map entry loader's present clamp. They were pinned on the branch that became #662, cut before #658's retain-unknown walk landed and merged after it, and the two diffs do not touch the same lines, so the merge was clean and nothing in CI read them. The first commit of this pair re-pins them; with them committed the job is green.

Wall time, measured locally on arm64 with the build tree deleted first: 21 s cold, 7 s warm. The CI number is on this pull request's own run.

4. Six baselines (#484)

for d in tables/*/; do ./bin/schema check $d; done named six units with no tables.baseline. Each is one unit in one directory, so each gets the file the tool writes, generated and never hand-written:

./bin/schema tables-baseline --update --reason "first baseline" tables/<d>

blobs, block, blockhome, scalars, stream, vocab9. The loop now prints the notice zero times.

What still prints it, and it is not a baseline this branch can write: make check prints eight, seven from test/tables and one from bench/corpus. test/tables holds seven units of different packages in one directory, which is the multi-unit case #484 itself carves out ("either split those units into directories or teach the notice to say so"). bench/corpus holds one unit and could take a baseline; it is outside the six and outside #484's own list of eight.

Verification

• go test ./... green.
• make check exit 0.
• The new job's command run locally: make -j"$(nproc)" conformance-pin update-goldens then git diff --exit-code -- testdata/, green, 21 s cold.
• make test stops before the dotnet legs, at tables-json-list-walk, and that stop is main's, not this branch's: a clean clone of dcdd5cdd fails the same gate with the same line. tables/maps/Spans.schema declares map[uint8][]Item, so every unit of the maps package emits the list half, while the gate still names build/tables-generated/maps/*Table.cpp in its list-free set. Filed separately rather than fixed here.

Closes #574
Closes #461
Closes #484

🤖 Generated with Claude Code

From the cold read

The pins job's first run found six stale maps goldens on the main of that hour and this branch re-pinned them (6d7386c); #668 then re-pinned the same six independently, so main is pin-fresh today and that commit merges to nothing (the reader materialised the merge tree and ran the gate on it: clean). The history is kept as it happened rather than rewritten. The pins job measured 1m44s and 1m21s on this PR's two runs, inside the two-minute rule, with big-endian the binding constraint at 4m27s. The message and object refusals are at parser.go:246 and :254.