This paper is also available as a print-ready PDF preprint (7 pages, with the register and the machine verifications as an appendix), and has been independently deposited with a timestamp: DOI 10.5281/zenodo.21908485. The site remains the publication of record.
We perform an exhaustive, pre-registered search over the Wedderburn-classified finite real spectral triples of noncommutative geometry, ranking candidates by frozen prefix-free description-length measures. Bare complexity minimization fails to select the Standard Model: the minimal admissible geometries yield the wrong unbroken gauge group under all three frozen measures. Adding frozen observer conditions (a light uncoloured charged fermion must exist; charged nuclei; reachability of SU(3) × U(1)em), the search uniquely reconstructs the one-generation Standard Model on C ⊕ H ⊕ M₃(C), with the hypercharges solved from the anomaly conditions. A cheaper Pati-Salam class initially wins, but a theorem from the exact first-order condition forces quark-lepton Yukawa equality in every embedded-lepton diagram; the measured bottom-tau ratio then eliminates the entire class, including its SU(5) extension. With neutrino mass as a second data gate, the unique survivor is the Standard Model with right-handed neutrinos. Three pre-registered attempts to derive the fine-structure constant fail quantitatively, locating the boundary between derivable structure and environmental parameters; the generation number resists derivation at a quantified price of 10 bits. Predictions: Majorana neutrinos, normal mass ordering (95-99.9%), mass sum near 0.06 eV, mββ of order 1-10 meV, no new coupled particles, and, under a Bayesian accounting with the universal prior, a θ-solving axion-like sector. All criteria were frozen before computation; all failures are reported.
The finite geometry F of the spectral action approach (Connes-Chamseddine) has always been an input: the algebra C ⊕ H ⊕ M₃(C) is chosen because it yields the Standard Model. This paper asks whether F can instead be selected, and by what. Our method is deliberately austere: a frozen, prefix-free description-length measure on the classified space of finite real spectral triples; hard admissibility conditions (chirality, anomaly freedom) that are membership requirements rather than penalties; observer conditions frozen before any scan; and measured low-energy data as final referee. Every criterion in this paper was written down and frozen before the corresponding computation was run, and we report the outcomes that went against us with the same prominence as those that did not. Three of our own derivation attempts are falsified below; the selection result survives them.
The algebra C ⊕ H ⊕ M₃(C) and its Standard Model realization go back to Connes and Chamseddine; Krajewski classified the finite triples; the Pati-Salam extension was developed by Chamseddine, Connes and van Suijlekom, who found the first-order condition obstructs it on the standard fermion space (our Theorem S′ shows why, and how the minimal diagram evades their obstruction structurally). Description-length selection descends from Solomonoff's universal prior and MDL; observer conditioning has anthropic antecedents (Weinberg; Tegmark), from which this work differs by freezing gates before computation and by using measured low-energy data as exclusion criteria rather than post-hoc accommodation.
Three complexity measures (prefix-free algebra coding, real dimension, gauge-group dimension) are carried in parallel; a selection claim is made only when all three agree, and only ordinal comparisons are ever used. Reformulation-dependence is handled as in algorithmic information theory: description length is defined up to the additive constant of the invariance theorem, which cancels in the chain comparisons used here.
Finite real spectral triples with AF a direct sum of matrix algebras Mn(K), K ∈ {R, C, H} (Wedderburn), Hilbert space and Dirac operator encoded by Krajewski diagrams, KO-dimension 6. Declared universe: up to three summands, n ≤ 4 (extended to n ≤ 5 and analytically to four summands in Sec. 8). Complexity of a candidate F = (algebra, diagram):
K_F = K_A + K_diag K_A = cost(s) + Σ over summands [cost(n) + log₂(3)] K_diag = cost(r) + r·log₂(B) + Σ over blocks cost(m) cost(n) = 1 + 2·floor(log₂ n) (Elias-gamma, prefix-free)
with s the number of summands, r the number of block types, B the available block types of the algebra, m block multiplicities. Prefix-free lengths add under composition, so K qualifies as a resistance functional in the sense of the underlying selection framework (Appendix A); the weight per bit is ln 2 (universal prior).
Admissible diagrams are chiral on the physical representation content (after solving for the anomaly-free U(1) direction, with pseudoreal/real equivalences) and free of all anomalies: cubic SU(n≥3), mixed factor²-Y, Y³, unimodularity, and the Witten condition for every Sp/SU(2)-like factor.
Calibrations (all passed, all mandatory): the one-generation Standard Model content passes with all anomalies vanishing and even doublet count; leptons-only fails on exactly Y³, SU(2)²-Y and Witten; and the search, run on C ⊕ H ⊕ M₃(C), must rediscover the SM diagram with the correct hypercharge ratios solved from the anomaly system. An earlier version of the filter failed the third calibration through three identifiable bugs (chirality tested on block indices rather than physical content; Witten applied only to quaternionic summands; the calibration itself not enforced); we document this because the audit trail is part of the method.
Of the 54 algebras with K_A ≤ K_A(C ⊕ H ⊕ M₃(C)) = 12.75 passing the admissibility gates, the K_F-minimal survivors are, in order: H ⊕ H ⊕ M₃(C) (K_F = 28.1, a left-right quark sector with no U(1) at the unification scale), C ⊕ C ⊕ M₂(C) (28.8, no colour), and only then diagrams on the SM algebra, whose own minimal diagram is a Y = 0 variant rather than the SM (the true SM content sits at K_F = 42.3). A breaking analysis under maximally generous rules (the scalar sector bounded above by the one-index rule, generic vevs) shows none of the K-minimal candidates can reach SU(3) × U(1)em: the failure criterion F1*, frozen at the start, fires unanimously under all three measures. Bare complexity does not select the Standard Model.
We then freeze three observer conditions: (O1a) a chirally protected, uncoloured, charged fermion state must exist (an electron; without it there is no atomic scale separate from the nuclear scale and no chemistry); (O1b) a chirally protected coloured charged state (charged nuclei); (O1c) SU(3) × U(1)em reachable under the frozen breaking rules. Result: of the ten surviving algebras, nine admit no observer-compatible diagram at all; the bottleneck, in every single case, is the electron. The tenth is C ⊕ H ⊕ M₃(C), and its minimal observer-compatible diagram is block-for-block the one-generation Standard Model, with hypercharges emerging from the anomaly system. Within the declared universe the SM content is not merely minimal but unique.
Unification-type algebras with an M₄(C) summand evade the electron gate: after SU(4) breaking, the fundamental splits as 4 → 3 + 1 and the singlet component is a charged lepton. Under frozen breaking amendments (Pati-Salam pattern), 24 such candidates beat the SM content, the cheapest being H ⊕ H ⊕ M₄(C) at K_F = 30.3, whose diagram is exactly the Pati-Salam fermion content plus a neutral block that supplies the SU(4)-breaking scalar. The exact first-order condition [[D,a],b°] = 0, imposed component-by-component (numerically exact, by rank computation), admits precisely the required scalar sector: the Majorana block (dim 7), the breaker (2L, 4̄) (dim 8), the bidoublet (dim 4), and nothing with no shared index. Notably this diagram evades the classical obstruction that forced the literature's spectral Pati-Salam model to abandon the first-order condition: its neutral sector sits in its own block outside SU(4).
Let (A, H, D, J, γ) be an admissible finite triple in which every charged uncoloured fermion state arises as the singlet component of the branching n → (n−1) + 1 of a fundamental representation of a simple summand Mn(C), n ≥ 4. Then for every Dirac component allowed by the first-order condition [[D,a],b°] = 0, the induced Yukawa coupling is proportional to the identity on the Mn(C) index, and consequently the down-type quark and charged-lepton Yukawa matrices coincide at the breaking scale of that summand.
Proof. Writing the fermion blocks as rectangular matrix spaces with left action by the row summand and J-induced right action by the column summand, the first-order condition applied with independent algebra elements on either side forces every component connecting blocks that share no index to vanish, and every component connecting blocks sharing one index to be a module map for that index, i.e. of the form φ ⊗ identity on the shared factor. A mass component for the block containing the embedded lepton necessarily shares its Mn(C) index with the quark components of the same block; being the identity on that index, it assigns equal couplings to the colour components (quarks) and the singlet component (lepton). ∎ The component spaces were additionally computed numerically-exactly (SVD rank over all 1600 algebra basis pairs), reproducing precisely the module-map dimensions and nothing else.
The Pati-Salam candidate passes a two-scale coupling consistency test (g₄ = gL = gR at Λ; SM two-loop below the intermediate scale; matching αY⁻¹ = αR⁻¹ + (2/3)α₄⁻¹): an exact triple crossing exists at MPS = 3.7×10¹³ GeV, Λ = 1.3×10¹⁵ GeV, αU⁻¹ = 44.5, with MPS inside the pre-declared seesaw window. But Theorem S′ bites: running the measured Yukawas up gives yb/yτ = 0.65 at MPS (two-loop QCD), a 35% violation of the forced equality, outside the frozen 25% band. Realistic threshold corrections cannot bridge it: finite matching contributions scale as (coupling²/16π²) times short logarithms (the MR-to-MPS window spans only 1.2 e-folds), giving a declared estimate below 3%, and the diagram's own breaker-induced mixing works in the wrong direction. We register explicitly the escape branch: a demonstrated threshold correction above 13% from a concrete spectrum would reinstate the Pati-Salam candidate; producing one is an open challenge to the reader, not an option we can exclude by fiat. A sweep confirms every observer-compatible diagram in the M₄(C) flank (and the M₅(C)/SU(5) flank, 136 algebras, where gauge-mediated proton decay additionally applies) has embedded leptons, and a search for hybrid diagrams with separate leptons returns empty. The entire embedding class is eliminated by the measured bottom and tau masses.
The bare SM content has no room for neutrino mass and is excluded by oscillation data. The unique geometry passing all gates: admissibility, observer conditions, b-τ, neutrino mass, is the Standard Model with right-handed neutrinos (K_F = 47.2). Robustness: extending the universe to n = 5 adds nothing (all 25 O-survivors have embedded leptons and fall under S′; the hybrid escape is empty), and four-summand algebras are excluded by an analytic cost bound (K_F ≥ 52.3 for any separate-lepton diagram).
Multiplicities. With a baryogenesis gate (CP violation via either the CKM route, requiring three generations, or leptogenesis, requiring two right-handed neutrinos; phase counting N(nR − 1), calibrated on known cases), the K-minimal observer-compatible multiplicity is one generation plus two νR: the measured three-generation world costs exactly 10 bits more (2⁻¹⁰ = 1/1024). The generation number is thereby an environmental ("address") fact with a quantified price, not a derived one; any future explanation must supply exactly those 10 bits. We verify that the dynamical term of the selection functional cannot: its naive vacuum reading would charge ~10²²⁸ bits per generation and is refuted by our own existence; renormalized, it is flat across candidates.
The observer-conditioned measure factorizes per member into discrete structure F (weight 2−K(F)) and continuous parameters with conditional density. For observed data, total surprise decomposes additively: −log₂ P = K(F) + [−log₂ P(data-window | F)]. Model comparison at fixed data must therefore use the sum: K is the prior, typicality the likelihood, and bare K-minimization is the special case of equal likelihoods. All accounting in Secs. 8-10 is this posterior comparison; an earlier prior-only registration of the dark-sector question, and its correction by the posterior analysis, are documented in the working records as a methods note.
Pre-registered and failed: (i) the unification route (SM content plus spectral boundary condition 5g₁² = 3g₂² = 3g₃², two-loop): predicted α⁻¹(me) = 152.2 vs 137.04 measured, 11% outside the frozen ~10% criterion; (ii) the bare selection route (Sec. 4); (iii) a parameter-free cutoff-function route (f(x) = e⁻ˣ forced by the framework's own weight, f₀ = 1): predicted αU⁻¹ ∈ [1.3, 5.1] vs a required 46: a factor 9. The pattern across all computations: countable structure (charges, ratios, mixing angle 3/8, group content) is derivable; continuous strengths are not. Typicality is then the correct question: within the literature habitability window (α⁻¹ ∈ [59, 685]) the measured value sits at log-position 0.34, comfortably interior. The residual "surprise ledger" of the framework: α ~5 bits (conditioning), generations 10 bits, cosmological constant 3.5-9 bits (band over both literature windows): 18-25 bits in total.
Root diagnosis and meta-stake (added 12 August). The three falsifications share one root with a fourth miss outside this paper (a Zipf demonstration): each time the selection measure, which weighs a-locally over structures, was put in a local chair: as boundary condition, as cutoff function, or as empirical law inside a single member. Formally: K counts bits, so e⁻ᴷ ranks countable alternatives, but singling out a point from a continuum costs infinitely many bits; continuous strengths are therefore address. Because this boundary was formulated after the falsifications, we register it as a frozen meta-stake itself: no first-principles derivation of a dimensionless continuous constant of nature from pure structure selection will ever succeed, by anyone. A parameter-free derivation of α (or another continuous constant) accepted by the community falsifies this paper's address reading. Reductions to other continuous constants, or frameworks introducing new continuous parameters, do not count as counterexamples.
Delegation stake (added 12 August, night). The level distinction points to one deriver that is not the measure: local fixed-point dynamics (asymptotic safety), which runs on exactly the structure selected here (Standard Model up to the Planck scale, no unification) and carries one pre-registration trophy (the 126 GeV Higgs mass, predicted in 2009). Registered as a weak stake: local self-consistency bounds continuous address coordinates rather than fixing them. For α the SM-only upper bound on the abelian coupling (Eichhorn-Versteegen) translates to α⁻¹ ≳ 75-90, shrinking the habitability window from below: a gain of 0.15-0.27 bits in the ledger above. The strong form (unique fixing) is empirically dead for SM-only (the interacting fixed point would put α ~35% too high) and closed via a GUT by Theorem S′. Delegation stake and meta-stake cover each other's flanks: exactly one can win.
1. Neutrinos are Majorana particles; the νR Majorana mass is the unique gauge-invariant Dirac component and the MR reproducing the measured neutrino mass scale (1.2×10¹⁴ GeV) falls mid-window. Definitive exclusion of neutrinoless double beta decay (LEGEND-1000/nEXO class) falsifies the entire chain. Because the outcome of that test depends on the mass ordering, we freeze the decision tree before JUNO reports: detection confirms regardless of ordering; an emptied inverted-ordering band refutes; a normal-ordering funnel outcome is registered in advance as undecidable rather than reinterpreted.
1b. Treating the neutrino sector's continuous parameters as generic draws from the measure reproduces the anarchy scenario, whose successful prediction of large θ₁₃ predates this work. Frozen forecasts: normal mass ordering at 95-99.9% (JUNO, within years); the mass sum peaked near 0.06 eV (DESI/Euclid class cosmology, this decade; current data already lean this way); mββ of order 1-10 meV, small but nonzero, so that null results at next-generation double-beta experiments are the expected outcome without undermining the Majorana claim. Sharpening (12 August): under the anarchy measure δ_CP is uniform and independent of the angles, hence frozen in advance: P(|sin δ_CP| > 0.5) = 66.7%, P(|sin δ_CP| > 0.25) = 83.9%, median Jarlskog invariant 0.024; an eventual measurement with |sin δ_CP| < 0.25 (DUNE, Hyper-K) is booked in advance as a miss.
2. No fourth generation; no WIMPs; no new coupled particles beyond νR under bare K-minimality.
3. Under the combined accounting of Sec. 8b, the strong-CP zero (34 bits as an address fact, given the measured CKM phase and θbare = 0 from the parity-even heat kernel) is cheaper to buy with structure: a θ-solving axion-like sector (10-20 bits) is favoured by ~14 bits, and doubles as a dark-matter candidate. Axion-search experiments are thus referees for, not against, the framework.
4. Structure stakes following directly from the chain (registered 12 August): (i) the proton does not decay: the selection eliminated precisely the decay-carrying embeddings (Pati-Salam, SU(5)), so a single confirmed proton decay (Hyper-Kamiokande, from 2027) breaks the full chain of Secs. 5-8; (ii) no light sterile neutrinos: the three heavy νR are the complete extension, and a fourth, light neutrino is structure without gate necessity; (iii) no Lorentz violation: the (3,1) signature is a singleton under the gates and a preferred frame costs bits without gate yield. An expectation calendar 2026-2030, listing per stake the referee and the meaning of the other outcome, is kept in the registry.
Further frozen stakes, with their referees, are maintained on the testing page of this site, which serves as the living registry of this paper's predictions.
The computation localizes the boundary between law and environment inside particle physics, and does so with falsifiable stakes on both sides of the line; since 12 August the line itself is a stake as well (Sec. 9). What is perhaps most notable methodologically: every wrong turn (three falsifications, an invalidated first scan, two conflicting registrations) is part of the published record, and the final selection emerged from, not despite, that discipline.
The weight P = e⁻ˢ follows uniquely from composition and additivity of resistance together with the probability codomain [0,1]: continuity is redundant, since an additive function bounded above on a set of positive measure is linear (Ostrowski), and the pathological solutions are unbounded on every such set. Prefix-free description length qualifies as a resistance term with weight ln 2 per bit (universal prior). The philosophical framework is deliberately confined to this appendix and to the rest of this site: the results of the main text stand independently of it.
All frozen protocols, outputs and the thirteen analysis scripts (anomaly and Krajewski search, breaking analysis, observer and data gates, Pati-Salam and SU(5) flank scans, endgame, generation round, coupling pipelines, first-order rank computation, b-τ running at one and two loops, αU point test, Sdyn cross-check) are archived with the working records of this site and are available on request; every number in this paper regenerates deterministically from frozen seeds.
Connes & Chamseddine (spectral action; Why the Standard Model); Krajewski (classification of finite triples); Chamseddine, Connes & van Suijlekom (Pati-Salam beyond the SM); Gilkey (heat kernel invariants); Foreman-Magidor-Shelah is not used here (see the mathematics page for the companion programme); Solomonoff; Li & Vitányi (universal prior, MDL); Weinberg (anthropic Λ); Tegmark (dimensionality); Hall, Murayama & Weiner; de Gouvêa & Murayama (neutrino anarchy); Renou et al. (real-valued quantum theory excluded); Boddy & Carroll (vacuum decay and Boltzmann brains); PDG for all measured inputs.