wasm: one build record for t27f.wasm, check:t27f compares the vendored copy with it

[gHashTag]

Refs gHashTag/t27#10229 (t27 PR: gHashTag/t27#10241)

wasm/t27f.wasm was rebuilt by ten hand-typed lab commands and the only record of a build was a hand-pasted list of sha256 lines. This adds one build record and a cheap check.

• wasm/t27f.build: # input lines (closure digest of the exporter specs and their imports, zig version, t27c sha256) then <sha256> <file> lines for export.zig, the generated Zig and JavaScript, and the wasm.
• npm run check:t27f: grep -v '^#' wasm/t27f.build | shasum -a 256 -c - -- exits 1 and names the file if a vendored copy drifted from the record. Verified OK on this branch.
• wasm/BUILD points at the rule in gHashTag/t27 specs/t27f/build.t27 instead of a hand-pasted sha block.

Note for review: the vendored spec_graph_view.zig, spec_graph_view.mjs and t27f.wasm were rebuilt from current t27 master with a deterministic build (lab; it reproduces trinity's vendored wasm byte for byte), so the view now carries master's dark theme. The wasm size is unchanged (39273 bytes), the sha differs.

No new hand-written code: the record is data, the check is one package.json script line using shasum.

Measured on the Railway lab: cold rebuild 1111 ms, REUSE hit 218-243 ms.

🤖 Generated with Claude Code

[tri-claw]

Summary by t27-bees

This PR adds a build record (wasm/t27f.build) and a verification script (npm run check:t27f) so the vendored t27f.wasm can be checked against its inputs. The wasm and its JS/Zig sources were rebuilt from t27 master (dark theme colors), changing their hashes but not the wasm size.

Worth a look

• package.json:14 — check:t27f runs shasum -c on every line in t27f.build, but the record lists export.zig, reel.zig, deck.zig, spec_graph_view.zig, spec_graph_export.zig, spec_graph_outline.zig which are not vendored in wasm/ (only the .mjs counterparts and t27f.wasm are). The check will fail with "No such file or directory" for those missing files.
• wasm/t27f.build:4–15 — The record includes 11 artifact lines but only 6 of those files exist in wasm/ per package.json:files. Either the record should list only vendored files, or the check should filter to existing files.
• wasm/BUILD:42–55 — The comment directs rebuilds to specs/t27f/build.t27 in gHashTag/t27, but this repo has no automation to run that rule; the manual commands listed (t27c gen …, zig build -Dtarget=wasm32-wasi) assume the full t27 source tree. Contributors without it cannot reproduce the build.
• wasm/spec_graph_view.zig:547,562,575,681,693 — Test expectations were updated to the new dark-theme colors. If the wasm is rebuilt again with different colors (e.g., upstream theme change), these tests will fail until expectations are updated again — which is correct behavior, but worth noting as a future maintenance point.
• wasm/t27f.wasm — Binary changed (sha differs) but size unchanged (39273 bytes). The new hash 7b2a0f0c… in t27f.build matches the rebuilt wasm; the old hand-pasted list in BUILD had deedd9c1…. No issue, but verify the vendored wasm is indeed the one just built.

---

_{Read: 6 files, +42 -36, the whole diff. Head 072e6fa. Model: openai-compatible/nvidia/nemotron-3-ultra-550b-a55b. Commands: @t27-bees review | summary | help | pause | resume. Free for every repository.}