# Reference fixtures Golden MSR files captured from fixed `muSRFlow` GUI configurations. Each `.generated.msr` is the byte-for-byte output of `buildMSR()` for a deterministic configuration and is verified hermetic by `tests/reference-fixtures.test.js` (no `musrfit` required). The naming records the workflow: - `asymmetry.generated.msr` — single-run Asymmetry fit with bounds and an FFT block (the canonical export path). - `bounds.generated.msr` — Asymmetry fit with tight explicit bounds and the FFT block disabled. - `t0background.generated.msr` — Asymmetry fit with fixed forward/backward t0 and fixed background values. - `singlehist.fitted.msr` — a real SingleHist **fit result** (fittype 0, `GaussianCos` model, 16 fitted parameters, converged `STATISTIC` block with `maxLH/NDF 1.04011`, NDF 3644) captured from production. It is the anchor for the reverse-parsing path: `tests/reference-fixtures.test.js` verifies that muSRFlow reads it back into GUI state (`parseMSRFitStatistic`, `parseMSRParameters`, `parseMSRTheoryModel`, `parseMSRGUISettings`). - `global.generated.msr` — global-asymmetry fit (`AsymmetryGLB`, fittype 2) over runs 2452-2460 with the `GaussianCos` model and a shared phase `Phi`. The generator emits a single-RUN template (run 2452) that `msr2data` later expands to all nine runs; the FITPARAMETER already carries the shared + per-run set. - `global.fitted.msr` — the real **global fit result** over runs 2452-2460: 37 fitted parameters (one shared `Phi` + 9 runs x `Alpha`/`Asy`/`Sgm`/`Frq`), converged `chisq/NDF 1.034138` (NDF 8198). Because the run-family THEORY shares parameters through `mapN`, `parseMSRTheoryModel` deliberately refuses to reconstruct it (safe guard, editor-owned) — the test asserts that refusal. - `singlehist.generated.msr` — SingleHist detector-normalization workflow (fittype 0): four histogram groups each with its own `No`/`NBg` normalization and a detector-resolved phase, so the RUN blocks carry `norm` + `map`. - `functions.generated.msr` — global-asymmetry workflow where the frequency is expressed through a custom field parameter (`fun1 = gamma_mu * Pfield`), so the generated MSR carries a `FUNCTIONS` block alongside THEORY. > `missingdata` was on the original pending list but is **not** a `.generated.msr` > workflow: `buildMSR()` has no "missing data" output. Missing data is a > run-time concern (`classifyMissingDataOutput` reads musrfit output that says a > data file was not found) and belongs to the integration/runner side, not the > hermetic generation goldens. ## Regenerating a fixture Edit the corresponding `config` in `tests/reference-fixtures.test.js`, write the new `buildMSR()` output to `.generated.msr`, and re-run the test. Only do this when the generator change is intentional; otherwise the test guards against unexpected drift. The real-fitted counterparts (`.fitted.msr`) used by the later integration checks are produced by running `musrfit` on the generated MSR against the configured data root. ## Integration checks `tests/integration-musrfit.test.js` is gated on the presence of the real `musrfit` binary and the fake data tree. When both are present it stages each single-run `.generated.msr` fixture (rewiring its data path to the on-disk fake data), runs `musrfit -e`, and asserts the fitted MSR's `STATISTIC` block matches the recorded converged value (asymmetry `39.776426`, bounds `60.869627`, t0/background `5.602355`, all NDF 406). It also round-trips every `.fitted.msr` fixture through `parseMSRFitStatistic`. Without the tools it prints a skip notice and exits successfully so the hermetic suite stays runnable on a bare CI box.