33.71 turns synchronized beat and propagation readouts into a retainable antisign court: after one frozen subtraction rule, the beat residual y and the normalized propagation residual u derived from Δt_common/t_model must stay opposite in sign within matched windows, flip together under tight/loose inversion, survive local-only / path-only / joint scans and multiple link lengths, and beat label / path / instrument / dispersion nulls; under V08/V09-compatible retain, J remains only a preregistered tight/loose state label rather than a new field ontology.
Use this section as a compact machine-readable EFT reference.
33.71 converts beat-rate and link-propagation anomalies into one antisign court. The admissible claim is not merely that both channels react to environment, but that after one frozen subtraction prescription the beat residual y and the normalized propagation residual u derived from Δt_common/t_model read the same matched-window state with opposite signs: tight conditions drive y negative and u positive, while loose conditions drive y positive and u negative.
The measurement ledger keeps only the hard observables. y is the fractional frequency residual from ν_obs against a frozen ν_model after gravitational, kinematic, and known environmental subtraction. u is the normalized propagation residual built from the dispersionless common-term time residual Δt_common over the modeled travel time t_model. Samples are stratified by an independent tight/loose label J drawn from a preregistered gradient G, a same-window base-map indicator T, or an event-template grade. The verdict metrics are S_anti, the sign of Corr(y,u), and the tight-versus-loose stratified consistency of the expected sign pairs.
The workflow is synchronized and separation-first. Beat and propagation channels share one time reference and matched windows, their calibration chains are frozen before extraction, and controlled tight/loose scans include polarity reversal so the sign pair must flip self-consistently. Three separation routes run in parallel: change only the local clock environment, change only the link-path environment, and change both together. At least two link lengths or path configurations are required so normalization by modeled travel time does not hide a one-baseline artifact.
Controls are built to break fake anti-correlation. With J near zero, both channels should collapse toward zero mean or a small calibratable constant rather than a stable antisign bias. Dispersion controls must show that Δt_common does not rescale like inverse frequency or wavelength squared. Label permutation must drive S_anti back toward chance, path permutation must move u without spuriously dragging y, and independent instrumentation swaps must preserve the verdict instead of tying it to one hardware chain.
Support requires a negative y–u relation that beats permutation baselines, a self-consistent sign flip under tight/loose inversion, and clean separation from near-zero, dispersion, label, path, and common-mode controls across multiple distance configurations. Falsification follows if the relation trends to chance in large samples, if inversion does not flip both channels coherently, if dispersive media behavior explains u, or if reasonable subtraction choices overturn the sign verdict. The main systematic ledgers are residual clock-environment effects, residual media and hardware delays, and synchronized triggering or time-base drift that can inject a shared artifact.
So the chapter closes only one retained clock–link antisign ledger aligned with the V08/V09 matched-window and null-separation lane. J may survive only as a preregistered state label for this audit, not as a one-chapter ontology owner. Its clean onward value is to feed the cross-era continuity court of 33.72 and the unified cross-probe scorecard of 33.73.