Falsifying a finite model: controls, calibration and predictions kept separate

State a claim that a calculation or experiment can contradict

A finite model is more specific than a broad narrative about gauge structure. The singlet construction supplies a Hilbert space, a root state, four oriented links, hopping t and penalty Δ. The channel example supplies registers, a preparation and a map. Each predicts numerical outputs once its inputs and measurement protocol are fixed. Their mathematical consequences can be checked independently, and a proposed physical realization can be compared with observations.

The source is explicit about this separation: the static Hamiltonian proofs require neither a temporal source nor a completion interpretation. Its operational reconstruction adds source, carrier, reference and dynamical interfaces. Temporal completion, v20, opening of Part I, p.8; compare Completed records, v21, §§5 and 9, pp.9–10 and 18.

A useful falsification statement therefore has the form: given these operators, this preparation, these independently calibrated parameters and this test, the output must lie in a specified range. An unrestricted claim that some completion mechanism could reproduce the output makes a different, much weaker commitment.

Separate exact predictions from approximation hypotheses

The first comparison requires the same root-nonsinglet penalty and spectator dual on both sides. The second cannot inherit validity at arbitrary t/Δ simply because the exact ground state remains isolated. The source's static gap theorem holds for unrestricted hopping, whereas its explicit weak-hopping remainder requires |t|/Δ<1/4. V20, Theorem 2.1, pp.9–10, and Theorem 2.5, pp.11–12.

A failed leading-order fit at stronger hopping is therefore not by itself a failure of the exact model. Conversely, exact agreement between two implementations does not establish that nature selects the implemented Hamiltonian.

Calibrate on one data family and reserve another for prediction

The coherent identity-loop return offers a concrete synthetic calibration. Use u=τ/ℏ, set t≥0 by convention, and write p_I(u)=1−a₂u²+a₄u⁴+O(u⁶). The source's moments give

The last expression requires t≠0 and the stipulated Δ>0 selects its positive square root. For generated data a₂=0.08 and a₄=0.009333333333, one recovers t=0.2 and Δ=1. These values follow from v20, the moment calculation in Theorem 20.4, p.61; they are not physical measurements.

Now reserve nontrivial holonomies and later return times as predictions. The same parameters must determine their curves. In real data, extracting a fourth-order coefficient is sensitive to finite-time terms and noise, so report the fit interval and parameter uncertainty. A separate energy or spectral calibration can be preferable. The essential principle is to avoid refitting Δ and t independently at every loop or time.

Use an equal-real-trace pair as a decisive response test

A model that predicts only κ[3−ReTrW] assigns identical responses to any two SU(3) loops with equal real trace. The exact singlet model makes a stronger prediction that can differ between them. Take the pair constructed in the full SU(3) domain: eigenangles (0,2π/3,−2π/3) and (θ,θ,−2θ), where cosθ=(√3−1)/2. Both have ReTr=0.

At independently fixed Δ=1 and t=0.5, the exact shifts are 0.0373563397744 and 0.0379782291064. Their difference is approximately 0.000621889332. A protocol with total uncertainty larger than that difference cannot resolve the proposed discrimination; agreement within such an error bar is inconclusive. These numbers come from the full finite Hamiltonian and its equivalent secular equation. V20, equations (2)–(5), pp.8–10.

The two-parameter domain matters because the original balanced one-angle slice cannot realize every such comparison. Nor can an arbitrary angle-dependent “effective coefficient” be introduced after observing the difference while still claiming to test the original one-coefficient trace functional. That modification changes the tested model.

Use controls that test structural assumptions directly

Several exact controls require no small-angle or weak-hopping expansion. Frame-transform every link and the corresponding carrier data; the energy and return probabilities must remain unchanged. Reverse the loop; the supplied static energy remains the same because its secular function depends on cosθ_a. Set t=0; ground energy is zero, the auxiliary gap is Δ and return probability is one. For m≥2 the computed full gap must obey Δ/(4m). V20, §2.1 and Theorem 2.1, pp.8–10.

These controls diagnose different failures. A frame-dependent output suggests inconsistent endpoint orientation or transformations. An incorrect zero-hopping result suggests a penalty or state-preparation error. A gap below the proved bound can reveal an omitted root-nonsinglet penalty or a wrong spectral branch.

A physical comparison must also check measurement timing. The source predicts different results for uninterrupted evolution, a forced circuit and intermediate projections. A change caused by deliberately adding intermediate measurements tests that intervention, not the untouched coherent-return theorem. V20, §20.4, pp.61–62. Specifying controls makes the proposed contradiction informative rather than merely numerical.

Challenge the retained-reference premise with more than marginals

The Bell-reference example gives a complementary channel test. In the label-erasure protocol, all AB observables are unchanged. In reference-A depolarization, XX=s, YY=−s and ZZ=s, while both marginals remain I₂/2. A measurement of the marginals alone cannot discriminate them or establish pointwise fixation of the full carrier algebras.

With trusted preparation and measurement axes, estimate s from one correlation and reserve the other two, the Bell overlap and the matrix spectrum as predictions. If independent process calibration supports exp(−γt), fit γ on one time interval and predict another. A simultaneous failure of the simple exponential need not be described as erased gauge information; it may instead indicate a different noise channel or preparation mixture.

The exact source statement concerns a unital completely positive map that fixes both full local algebras for all observables. It then fixes their generated joint algebra. V21, Proposition 4.7, p.8. Actual carrier depolarization fails that premise for positive exposure. Testing the premise on an operator basis is more revealing than declaring a common reference absent after observing a mixed local state.

Do not replace a proved bound by an attractive fit

The source gives the raw-energy remainder B=2048Δε⁶/(1−16ε²) for |ε|<1/4. Subtracting E(I) from E(W) introduces two such remainders, so the direct bound for the normalized response is 2B. At Δ=1 and ε=0.1, that is approximately 0.00487619. A central SU(3) loop has leading response 0.0003, much smaller than this conservative error certificate. V20, Theorem 2.5 and its qualification, pp.11–12.

The approximation can be numerically much better than the bound. That observation should be reported as a numerical result, not as a tighter theorem that was never proved. Independent exact matrix calculations are useful precisely where a general analytic bound is too conservative to resolve a desired comparison.

Keep numerical error, truncation error, detector statistics and calibration drift separate. For a difference of independently estimated probabilities, variances add; for bounded unknown systematic errors, a worst-case absolute budget adds magnitudes. A confidence claim requires its sampling assumptions. A residual without a defined uncertainty and comparison model cannot establish either falsification or confirmation.

A useful negative result narrows the next research task

A reproducible discrepancy should include the exact source version/page, parameter values, link matrices or eigenangles, preparation, measurement schedule, expected result and observed or independently computed result. This allows collaborators to distinguish an implementation error from a violated model assumption. A new term or channel may repair the fit, but its introduction should be named and tested as an extended model.

Successful checks support increasingly specific claims: correct algebra, correct finite implementation, and perhaps a calibrated physical realization. They do not by themselves select the six nested inner generative levels, seventh completion, retained carrier ranks, determinant resource or spatial dynamics. The operational manuscript lists these additional interfaces and supplies flat countermodels under weaker assumptions. V21, §§5, 8–9, pp.9–10 and 16–18.

The collaboration page invites precisely scoped contributions: a sharper remainder, an independent return calculation, a channel counterexample with its quantifiers stated, or a physical preparation that makes a new prediction possible. Such results can improve the programme without inflating a finite auxiliary success into an unconditional Standard Model derivation or continuum mass-gap theorem.

Continue reading

• Exact return curves and measurement schedules
• A same-real-trace stress test
• Reference retention and degradation
• The gauge claim-status ledger
• Propose a focused collaboration