Step 12. Registered workflow & publishing checklist

Plain English: this step is about discipline. Before touching data, you freeze your plan (pre-register it), and after the run you publish enough artifacts for anyone to re-run your analysis bit-for-bit. Here’s the minimal, “no-wiggle-room” checklist the paper advocates.

A. Pre-register (freeze before analysis)

• Hypotheses & channel

  • Lever A (flux→drift): \(H_0:\lambda=0\); \(H_1:\lambda=G/(c^4 r)\) with correct sign and zero-lag.
  • Lever B (visibility): bound \(S_\Phi\) via both Markovian and filter-weighted estimators.

• Geometry & timing

  • Definition of \(r\) (areal/straight-line), clock separation \(\ll r\), sampling rates, record lengths, trigger windows (e.g., SNEWS ±6 h).

• Instrument config

  • Clock lines, servo bandwidths, time-transfer method (GPSCV/TWSTFT/fiber), detector configs.

• Data conditioning

  • Detrend, band/line notches, resampling, PSD method (e.g., Welch params), whitener freeze.

• Analysis templates

  • Lever A: template \(h(t)=-G L(t)/(c^4 r)\); estimator (GLS/matched filter), sign-lock and 1/r scaling tests.
  • Lever B: measured \(f(t)\) and derived \(F(\Omega)\); visibility integral \(-\ln V=\tfrac{\omega^2}{2}\!\int S_\Phi|F|^2\).

• Controls & nulls

  • Shuttered source / cold dump, off-band, time-slides, cable swaps, cross-site desyncs.

• Injections

  • Software phase injections (with forced \(1/r\) ladder); optional safe hardware injections; success metrics.

• Calibration plan

  • Power \(L(t)\) (traceable meters), distance \(r\) (survey), timebase, neutrino-count→\(L_\nu(t)\) mapping (if astro).

• Pass/fail rules

  • Exact criteria for “detection-like” template recovery vs “publish bound only.”

B. Publish (make it reproducible)

• Raw & conditioned data

  • Raw time series, conditioning scripts, PSDs, whiteners, measured \(f(t)\), computed \(F(\Omega)\).

• Code & environment

  • Version-pinned repo, environment file (e.g., lockfile/containers), run scripts, checksums.

• Derived products

  • Lever A: \(\hat\lambda\) with \(\mathrm{Var}_{\rm stat}\) and total variance (incl. calibration/systematics), 1/r regression plot, zero-lag cross-corr.
  • Lever B: \(V(T)\) with errors; two bounds on \(S_\Phi\): Markovian \(S_\Phi(0)\) and filter-weighted band figure.

• Uncertainty budget

  • Table separating statistical, calibration (\(r,L\)), and model mismatch \(\Delta_S\).

• Sanity figures

  • Injection recovery, source-off baseline, noise budget (Newt/shot/vacuum), lag scan.

• Registration proof

  • Timestamped prereg doc & hash; DOI (e.g., Zenodo) for the full bundle; license.

Resulting posture: either (i) you recover the template with the right sign, zero-lag, and 1/r within stated errors, or (ii) you publish a bound with every ingredient to let others challenge or improve it.

Glossary (for this step)

Ready for Step 13 — Theoretical backbone recap: why these observables fall straight out of lapse-first GR (constraint-projected \(\Phi\), no extra DOF) and how the sign-lock emerges.