33.53 turns merging clusters into a retainable calibration field: after source-end calibration, chromatic stripping, shared time and frequency standards, and environment-only prediction cards built from TSP/TSC, eventness, convergence / shear, lensing–X-ray misalignment, non-thermal radio, and tumbling proxies, Δt_common or PER-like residuals must stay non-dispersive across bands, align in sign and strength across at least two probe classes within matched spatial and time windows, rank monotonically or by plateau/threshold with environment tier, and beat randomized baselines in blinded holdouts; under V08/V09-compatible retain, this remains one merger-cluster PER/common-term calibration ledger rather than a global redshift or ontology verdict.
Use this section as a compact machine-readable EFT reference.
33.53 says merging clusters are useful only if they function as a rigid calibration field. A PER-like or Δt_common residual that survives ordinary corrections should be easiest to align and hardest to fake where the environment is steep, phase-ordered, and auditable. The chapter therefore turns dramatic mergers into a hard calibration court rather than a generic story about unexplained redshift residue.
The measurable object is a paired template. On one side sit phase, eventness, TBN proxies, STG proxies, convergence / shear, and lensing–X-ray misalignment tiers. On the other side sit sign-stable, frequency-independent common-term residuals extracted from at least two probe families, such as strong-lensing time-delay residuals, lensed FRB image-to-image shifts, same-source line ensembles, or a 21 cm line-cube residual. The calibration claim is that these residuals should follow the environment template in direction and strength.
Execution keeps the calibration court blind and cross-probe. The cluster list, source lists, phase aperture, and alignment framework are preregistered. Each probe completes its own standard subtraction before extracting the common term, while environment-only prediction cards publish direction, strength tier, band-edge non-dispersive expectations, and phase windows. Arbitration then scores hit, wrong-direction, and null rates under holdouts without allowing card revision.
Null pressure is destructive by design. Sightlines that traverse the merging cluster should outrank nearby non-traversing ones. Randomizing the environment template or the phase bins must break the relation, and random rescaling or splicing of probe data may not reproduce a comparable non-dispersive residual. Persistent disagreement between probe classes counts as alignment failure rather than support.
Support requires three simultaneous outcomes: non-dispersiveness survives frequency and band-edge pressure, the residual tracks phase and environment monotonically or by plateau/threshold, and normalized probes agree on sign and relative ranking under one template. Falsification follows from dispersion-like behavior, failed environment stratification, probe contradiction, or holdout performance near random. The named adversaries are source-end drift, macro-model leakage, geometric misalignment, and mismatched beam or time windows.
So the chapter closes, at most, one retained merger-cluster calibration ledger aligned with V08-8.4 / 8.5 and V09-9.6. It may not settle global redshift ontology or let one probe family stand in for the whole court. Its clean onward value is to hand a rigid cross-probe calibration grammar to 33.54–33.56.