spec(t27mcp): spec_brief -- brief spec row fields and why a spec fails (Closes #10090)

[gHashTag]

Closes #10090

Part of the epic #10066 (learnability: fewer tokens and round trips for cheap models). Stacked on #10083 (mcp-t27-examples), which stacks on #10080.

What

specs/t27mcp/spec_brief.t27 (module T27mcpSpecBrief): two decisions of the t27 MCP server (gHashTag/t27-mcp) that were hand-written JavaScript.

• BRIEF_FIELDS / ONE_FIELDS: which manifest fields a short spec row keeps (t27_specs rows and t27_spec). Emitted by t27c gen-js; the server copies those keys and nothing else.
• why_mask(is_fail, nodes, nfailed, tc, loss): why a health: "fail" spec fails, as a bit set (W_NOT_PARSED, W_BACKEND, W_TYPECHECK, W_LOSS, W_UNKNOWN). A spec that is not failing gets 0. WHY_TEXT holds one template per bit; {} is the count or the backend list the server fills in.

specs/OWNERS.md extends the t27mcp/ row.

Evidence (Railway lab, t27c 0.5.2)

• t27c test-report specs/t27mcp/spec_brief.t27: 5 of 5 pass, 0 vacuous.
• Negative controls: a spec that is not failing has no reason; a failing spec with no counter is W_UNKNOWN alone. With the not-parsed rule inverted, 3 of 5 tests fail.
• t27c gen-c plus zig cc -target wasm32-wasi lowers it to a 185-byte wasm; the commands and sha256 are in wasm/BUILD of the server PR.
• The reasons the wasm gives equal the JavaScript it replaces on all 2402 specs of the published manifest (113 failing, 0 differences).

Known limit

.len on a const table is constant-folded: a test that asserted table lengths ran 0 runtime asserts and was reported vacuous, so the spec checks the bit set through why_mask instead of table lengths.

🤖 Generated with Claude Code