This is the demonstration layer of the Phase III central proposition of the research programme on the rigidity of geometric incidence. The predecessor framework supplies the quantum-measurement collapse arrow as a purely geometric datum — the causal structure and temporal orientability of Minkowski space — while keeping the global evolution unitary, and records the irreversibility of a measurement event as a causal monotone tied to the four-volume of the causal diamond between event and observer. The same data admit two readings: a geometric reading, in which the functional form of the irreversibility is rigidly determined by the modular geometry of the diamond (fixed up to an overall normalization) and is invariant under a change of geometry without recalibration; and an effective reading, in which the law is merely carried by a thermodynamic registration channel whose coefficient is free and must be recalibrated for each geometry. The paper poses the decision between them as a single theorem of empirical non-equivalence, built symmetrically so that either verdict — rigidity (new physics) or equivalence (reduction to einselection with an imported thermodynamic arrow) — would be a valid, publishable outcome. Working in algebraic quantum field theory, where the local algebras are type III₁, it establishes (on a reference state with geometric modular action — the vacuum, via Bisognano–Wichmann) that the irreversibility functional is determined by the diamond's modular data up to one normalization constant (Tomita–Takesaki, Casini–Huerta–Myers, Araki relative entropy), with the entire residual freedom localized in the modular status of the record-carrying state. Whether that normalization is itself geometric (the value π/24) is graded as a conjecture and is not load-bearing; the decisive content is invariance under a change of geometry, and the value is adjustable. In the genuine type III₁ continuum, both halves of the discriminating apparatus return a third verdict: the registered antisymmetric orientation is geometric and irreducible to any CPTP channel (in type III₁ there is no reduced density matrix to dilate, so no Stinespring dilation can readmit it "as only an ancilla"), yet it coincides with the Borchers–Wiesbrock spectral condition (energy positivity) — already-known QFT serving as the arrow, rather than a new irreducible mechanism. Any genuinely new positive is localized to two named sites: the interacting / non-geometric-flow sector (β ≠ 2π/a, where the Connes–Rovelli thermal-time reading must be rebutted, not assumed) and the orientation of the modular inclusion. A preregistered finite-N Gaussian discriminator returns equivalence in the simulable (type I) class — reported honestly as inconclusive for the fundamental reading, since the decisive type III₁ regime is not reached at any finite N. Scope: special-relativistic (Minkowski); gravity and quantum gravity are deferred. All results are explicitly graded (theorem / proposition / conjecture / computation / inherited), and the type III₁ instruments and verdict data are included with the deposit. Changes in v2 (2026-07-08). This version adds Phase IV, step 3 — the two-axis closure of the region left open by the free-sector verdict. Route (b), non-geometric Gaussian flow (multi-interval): the multi-interval vacuum modular flow is genuinely non-geometric (bilocal strength ρbi ∈ 0.004, 0.064) yet carries a single modular temperature on every channel (D ≡ 0 at βmod = 1 to 10-12, continuum-stable), with the independent-scales einselection foil detected at D = 0.148 — NULL. Route (a), perturbative non-Gaussianity (tree-level λφ3): the T-odd connected cumulant onsets at first connected order (|A3| ∝ λ) but its separating content is zero (D = 0, dD/dλ = 0), while the multi-scale and FDT-breaking foils fire at 0.138 and 0.177 — NULL / recalibrable; the third verdict extends to the perturbative interacting sector. Both axes were pre-registered before running (frozen 2026-06-26 and 2026-06-25 respectively); no v1 result is altered. New verified references: Peschel (2003), Casini–Huerta (2009), Peschel–Eisler (2009), Arias–Casini–Huerta–Pontello (2018).
Pablo Domínguez Delmás (Wed,) studied this question.