fit on train · test on holdout · ship outputs, not recipes · train ∩ test = ∅
RMS_before · typical
low 30s
ns · window-dependent
RMS_after · holdout
17–22
ns · not remainder = 0
Δ_RMS · typical
~45%
not a guarantee on your data
Holdout-validated timing evaluation.
Post-processing for long-baseline GNSS clock comparison. You get corrected outputs and RMS-class summary statistics—the correction recipe stays internal.
For people who already know what “30 s final CLK” means: we publish the evaluation protocol and typical holdout numbers. We do not publish the fit.
Typical performance · held-out
HOME_SPEC_v0 · train ∩ test = ∅| Primary metric | RMS on agreed comparison/residual seriesSame definition before and after |
|---|---|
| Uncorrected (typical) | low 30s nsWindow- and baseline-dependent |
| Corrected (typical holdout) | 17–22 nsTested configurations; not remainder = 0 |
| Improvement (typical) | ~45% RMS reductionNot a guarantee on your panel until you run the pilot |
| Next step | Evidence room → — pinned runs, null tests, strict-window example |