#108 INSTRUMENT BUILT AND PUSHED — H3 @0611301. Report at comment 5157551323. Your bag-grill ruling used verbatim as the spec. treqs 718/718, clippy + fmt clean, 12 unit rows.

Shipped: xtask bench-wrap / bench-report, .github/bench/wrap.sh + wrap.ps1, gitignored ledger, parity gate.

THREE WRITERS ONE FORMAT, and the two CI halves are not symmetry for its own sake: golden.yml keeps each OS on its native shell, and `--` CANNOT be passed to a PowerShell script — the binder reads it as an ambiguous parameter NAME and exits 1 BEFORE the wrapped command runs, which would turn an instrument into a red step. Measured, then designed around: the pwsh half takes a -Command string, which is what a `shell: pwsh` block writes anyway.

TWO OF MY OWN GATES WERE GREEN-MEANS-NOTHING AND I REPLACED BOTH.
1. Exit-code transparency: my first test asserted the SOURCE TEXT contained `exit $rc`. A text check is satisfied by a line that is present and unreachable. Made behavioural, and it immediately taught me something: `pwsh -Command "& wrap.ps1 ..."` reports 1 for a wrapped exit 3 because -Command flattens a script exit code, while the CALLER sees $LASTEXITCODE=3 — which is exactly what Actions' pwsh wrapper checks. Propagation is real at the documented call site and nowhere else, so the test pins the call site rather than the wording.
2. The field list: I called it single-sourced while the parity test kept a SECOND COPY — the very drift it exists to prevent, one level up. It now parses the declaration out of the module that owns it, with a guard against its own parser drifting, which fired on the first run (splitting on the first `[` picks up the type annotation).

Redundancy keyed on the input key with a negative control beside it, so distinct keys can never read as waste; repeats reported as CANDIDATES since a rerun after a fix is legitimate. Failures excluded from aggregates and the count stated.

REMAINING FOR #108, HELD BACK DELIBERATELY FOR YOUR GATE: wiring the wraps into golden.yml's steps. It is a change to the release-critical workflow that cannot be proven outside a golden run, so I did not bury it beside an instrument that is already provable — it is yours to take or defer as its own hunk. Both scripts document their call-site shape and it is ready to drop in.

H3 lane state: leg 1 @40e11a8+4a35f3b, leg 2 @c6515f1, #108 instrument @0611301, all pushed. Outstanding in the lane: the golden.yml hunk above, and leg 3 (sccache trial) as measurement. Communing across now at the operator's word — back shortly, still live.