Paper 7 established that the 31 binary stratum of the PSL(2,7) decomposition — a single rotating triangle with three active gates {R, S, T}, no standing wave — has the truncated Dirichlet series Z(s) = Σ n⁻ˢ as its natural transfer operator and the non-trivial Riemann zeros as its spectrum. The complementary half of the architecture is the 137 ternary stratum (Z₆₂ + T₇₅): two full counter-rotating triangles each carrying the complete five-gate set {S, R, T, F, P}, coupled through a shared substrate. Paper 7 left the form of this ternary spectrum open. The present paper measures it and reports its structural form. The architecture. Two full Paper 33 compute triangles (12 qubits each) share a single 4-qubit database merkabit through cross-chiral memory tunnels u_C₁ → v_D and u_C₂ → v_D at J_mem = 0.5. The total footprint is 28 qubits; the v_D reduced density matrix is the spectral readout. Both triangles run at the same Coxeter period T_CYCLE = 12; the experimental degree of freedom is the relative phase offset offset_T2 ∈ {0, 1, …, 35}, sweeping three full Coxeter periods. The result: cyclotomic Z₃. A 28-qubit cirq.Simulator sweep over 324 configurations (36 phase offsets × 9 Z₃ × Z₃ input-label pairs) produces a mean v_D von Neumann entropy spectrum that is perfectly period-12 with a peak at offset = T_CYCLE / 3 = 4 (collapse event, ⟨S⟩ = 1.1669 — v_D maximally mixed) and a trough at offset = 2 T_CYCLE / 3 = 8 (locking event, ⟨S⟩ = 1.1135 — v_D most coherent). The peak-to-trough contrast Δ⟨S⟩ = 0.0534 sits ~25× above numerical noise. (All numbers from the complex128 Stage C deposit; the earlier complex64 deposit gave 0.0521, ~2.5 % suppressed by single-precision floating-point rounding.) The FFT power spectrum is bichromatic: the fundamental at period 12 (power 0.0966) and the second harmonic at period 6 (power 0.0672); every other Fourier component is three orders of magnitude below at ≈ 10⁻⁴. This is the FFT signature of the Z₃ character table imprinted on a 12-step Coxeter cycle — exactly two non-zero Fourier coefficients (the cyclotomic field's generator and its conjugate), all others forced to zero by Z₃ symmetry. Both the period-12 master clock and the cyclotomic Z₃ symmetry are forced by the framework's forcing chain ℤ[ω] → Z₃ → A₄ → P₂₄ → McKay → E₆, with zero free parameters. Observable 23 — Pentachoric-Lock Spectrum (28 qubits). Four pre-registered falsifiable thresholds operationalise the prediction: (a) FFT dominant period = T_CYCLE ± 1 % (measured: 12.00 exact) (b) Peak-trough offset separation Δ = T_CYCLE / 3 ± 1 step (measured: 4 ± 0) (c) Peak-trough entropy contrast ≥ 0.02 (measured: 0.0534) (d) Harmonic purity ≥ 90 % of total non-DC power concentrated in {T_CYCLE, T_CYCLE / 2} (measured: 98.94 %) All four pass at ideal simulation. Period-12 exactness is bit-identical at 20/24 testable b vs b+12 offset pairs and within one ulp (2.22 × 10⁻¹⁶, double-precision machine epsilon) at the remaining 4. Hardware-noise-calibrated execution on IBM Eagle r3 / Heron r2 and Google Willow is committed code; the sweep awaits scheduling. Observable 24 — Regime 3 forward test (incommensurate coupling). The same double-triangle architecture with Triangle 2's internal Coxeter period set to 13 (or equivalently J_intra_2 = J_intra_1 × √2) breaks rationality between the two triangles' frequencies. Paper 7 Regime 3 predicts that incommensurate coupling destroys cyclotomic structure and produces GUE-distributed collapse offsets — the ternary-scale analogue of Riemann-zero-like behaviour. The script extension build_double_triangle_circuit(..., T_CYCLE_2=13) is committed; the 64-offset sweep is forecast as the multi-stratum continuation of Paper 7. Substrate verification — the R-locking test. Paper 34's §5 result rests on h(E₆) = 12 being the architecture's master clock. That premise is independently verified at the single-merkabit scale by R_locking_test.py: of five candidate activation patterns over the gates {S, R, T, F, P}, only the original 5-fold cycling reproduces α⁻¹ = 137.036 to within 10⁻⁴. Every 4-fold variant breaks the Floquet return F = 0.69678 by factors of 6–14× and shifts α⁻¹ by 0.15–0.27 units from CODATA — a deviation falsified by the IBM Eagle r3 hardware confirmation of Paper 24 to > 10³ σ. The 12-step Coxeter period is therefore not a numerical choice; it is the unique value compatible with the measured fine structure constant. Paper 34's cyclotomic Z₃ entropy spectrum and the R-locking test are two phase-shifted projections of the same underlying Coxeter clock — the 5-fold and the 3-fold subgroups of h(E₆) = 12. Position in the Merkabit Research Series. Paper 7 located all 29 known Riemann zeros in the 31 binary architecture's truncated Dirichlet spectrum. Paper 34 measures the complementary 137 ternary architecture's spectrum and reports cyclotomic Z₃ structure as its natural form. Together they exhaust the PSL(2,7) = 31 + 62 + 75 = 168 decomposition: arithmetic spectrum on the binary half, cyclotomic spectrum on the ternary half. Paper 31 establishes the cross-chiral tunnel as the directional ternary write primitive; Paper 32 establishes the 4-square memory register; Paper 33 composes them into the 19-qubit Pentachoric Verification Protocol; Paper 34 extends the architecture to 28 qubits and reports the spectral signature of the ternary stratum. Hardware confirmation of the single-cell substrate (5/5 pre-registered observables on IBM Eagle r3 and Heron r2, sub-Poissonian Fano F = 0.856 ± 0.030 at 131 σ below Poisson) is in Papers 24–26.
Paper 7 established that the 31 binary stratum of the PSL(2,7) decomposition — a single rotating triangle with three active gates {R, S, T}, no standing wave — has the truncated Dirichlet series Z(s) = Σ n⁻ˢ as its natural transfer operator and the non-trivial Riemann zeros as its spectrum. The complementary half of the architecture is the 137 ternary stratum (Z₆₂ + T₇₅): two full counter-rotating triangles each carrying the complete five-gate set {S, R, T, F, P}, coupled through a shared substrate. Paper 7 left the form of this ternary spectrum open. The present paper measures it and reports its structural form. The architecture. Two full Paper 33 compute triangles (12 qubits each) share a single 4-qubit database merkabit through cross-chiral memory tunnels u_C₁ → v_D and u_C₂ → v_D at J_mem = 0.5. The total footprint is 28 qubits; the v_D reduced density matrix is the spectral readout. Both triangles run at the same Coxeter period T_CYCLE = 12; the experimental degree of freedom is the relative phase offset offset_T2 ∈ {0, 1, …, 35}, sweeping three full Coxeter periods. The result: cyclotomic Z₃. A 28-qubit cirq.Simulator sweep over 324 configurations (36 phase offsets × 9 Z₃ × Z₃ input-label pairs) produces a mean v_D von Neumann entropy spectrum that is perfectly period-12 with a peak at offset = T_CYCLE / 3 = 4 (collapse event, ⟨S⟩ = 1.1669 — v_D maximally mixed) and a trough at offset = 2 T_CYCLE / 3 = 8 (locking event, ⟨S⟩ = 1.1135 — v_D most coherent). The peak-to-trough contrast Δ⟨S⟩ = 0.0534 sits ~25× above numerical noise. (All numbers from the complex128 Stage C deposit; the earlier complex64 deposit gave 0.0521, ~2.5 % suppressed by single-precision floating-point rounding.) The FFT power spectrum is bichromatic: the fundamental at period 12 (power 0.0966) and the second harmonic at period 6 (power 0.0672); every other Fourier component is three orders of magnitude below at ≈ 10⁻⁴. This is the FFT signature of the Z₃ character table imprinted on a 12-step Coxeter cycle — exactly two non-zero Fourier coefficients (the cyclotomic field's generator and its conjugate), all others forced to zero by Z₃ symmetry. Both the period-12 master clock and the cyclotomic Z₃ symmetry are forced by the framework's forcing chain ℤ[ω] → Z₃ → A₄ → P₂₄ → McKay → E₆, with zero free parameters. Observable 23 — Pentachoric-Lock Spectrum (28 qubits). Four pre-registered falsifiable thresholds operationalise the prediction: (a) FFT dominant period = T_CYCLE ± 1 % (measured: 12.00 exact) (b) Peak-trough offset separation Δ = T_CYCLE / 3 ± 1 step (measured: 4 ± 0) (c) Peak-trough entropy contrast ≥ 0.02 (measured: 0.0534) (d) Harmonic purity ≥ 90 % of total non-DC power concentrated in {T_CYCLE, T_CYCLE / 2} (measured: 98.94 %) All four pass at ideal simulation. Period-12 exactness is bit-identical at 20/24 testable b vs b+12 offset pairs and within one ulp (2.22 × 10⁻¹⁶, double-precision machine epsilon) at the remaining 4. Hardware-noise-calibrated execution on IBM Eagle r3 / Heron r2 and Google Willow is committed code; the sweep awaits scheduling. Observable 24 — Regime 3 forward test (incommensurate coupling). The same double-triangle architecture with Triangle 2's internal Coxeter period set to 13 (or equivalently J_intra_2 = J_intra_1 × √2) breaks rationality between the two triangles' frequencies. Paper 7 Regime 3 predicts that incommensurate coupling destroys cyclotomic structure and produces GUE-distributed collapse offsets — the ternary-scale analogue of Riemann-zero-like behaviour. The script extension build_double_triangle_circuit(..., T_CYCLE_2=13) is committed; the 64-offset sweep is forecast as the multi-stratum continuation of Paper 7. Substrate verification — the R-locking test. Paper 34's §5 result rests on h(E₆) = 12 being the architecture's master clock. That premise is independently verified at the single-merkabit scale by R_locking_test.py: of five candidate activation patterns over the gates {S, R, T, F, P}, only the original 5-fold cycling reproduces α⁻¹ = 137.036 to within 10⁻⁴. Every 4-fold variant breaks the Floquet return F = 0.69678 by factors of 6–14× and shifts α⁻¹ by 0.15–0.27 units from CODATA — a deviation falsified by the IBM Eagle r3 hardware confirmation of Paper 24 to > 10³ σ. The 12-step Coxeter period is therefore not a numerical choice; it is the unique value compatible with the measured fine structure constant. Paper 34's cyclotomic Z₃ entropy spectrum and the R-locking test are two phase-shifted projections of the same underlying Coxeter clock — the 5-fold and the 3-fold subgroups of h(E₆) = 12. Position in the Merkabit Research Series. Paper 7 located all 29 known Riemann zeros in the 31 binary architecture's truncated Dirichlet spectrum. Paper 34 measures the complementary 137 ternary architecture's spectrum and reports cyclotomic Z₃ structure as its natural form. Together they exhaust the PSL(2,7) = 31 + 62 + 75 = 168 decomposition: arithmetic spectrum on the binary half, cyclotomic spectrum on the ternary half. Paper 31 establishes the cross-chiral tunnel as the directional ternary write primitive; Paper 32 establishes the 4-square memory register; Paper 33 composes them into the 19-qubit Pentachoric Verification Protocol; Paper 34 extends the architecture to 28 qubits and reports the spectral signature of the ternary stratum. Hardware confirmation of the single-cell substrate (5/5 pre-registered observables on IBM Eagle r3 and Heron r2, sub-Poissonian Fano F = 0.856 ± 0.030 at 131 σ below Poisson) is in Papers 24–26.
This paper composes the primitives established in Papers 31 and 32 into a primitive-complete, self-hosting quantum computer on 19 qubits — a categorically stronger claim than the substrate-of-physics claim of the Merkabit Research Series (Papers 1–28) or the hardware-confirmation claim of Papers 24–26. Every operation required for computation — write, compute, communicate, store, retrieve, and clock — reduces to native merkabit dynamics. No external gate library is required. The architecture does not enable a computer built on top of it; it is the computer. The 19-qubit footprint. Three 4-spinor tesseract merkabits in a closed compute triangle (12 qubits) with cross-chiral tunnels at J_intra = 0.1 on every triangle edge; one database merkabit D coupled to merkabit C through a memory-bus tunnel at J_mem ∈ [0.5, 1.0] (4 qubits); a 2-qubit Z₃-eigenstate reference register; one SWAP-test ancilla. The compute triangle and the memory bus are the same operator (iSWAP^J) at two different coupling strengths — the architecture is uniform under computation, with what distinguishes compute from storage being coupling rate rather than gate type. The native computation is the Z₃ cyclic clock. Composing three triangle-edge tunnels with one memory tunnel produces, as an emergent computational output, the Z₃ Galois rotation α → γ → β → α: writing label X produces a stored state on v_D that recognises cycle(X) = X⁻¹. Three writes return to the original label. The clock was not programmed; it falls out of the architecture's own Z₃ symmetry group acting on its own primitive set. This is the first result in the series in which the architecture's computational output is itself a structural signature of its symmetry group. The Pentachoric Verification Protocol (Observable 22). Five simulation stages map one-to-one onto the five gates of the ouroboros cycle of Paper 24: Substrate (S), Rotation (R), Transfer (T), Frequency / self-sustainment (F), Phase (P). Each gate is a single independent pass/fail test on the 19-qubit circuit; failure of any single gate falsifies one specific named primitive, and the failure pattern across {S, R, T, F, P} localises which compositional step did not survive hardware noise. The decisive gate is R: it asks whether writing label X returns a state that recognises cycle(X) under the Z₃ permutation. Sub-thresholds for all five gates are anchored to public commit SHAs. Cross-architecture simulation results. All five gates pass at three noise profiles: ideal cirq.Simulator (R = +0.159 ± 0.004 at +39.7σ), IBM Heron r2 (p_depol = 0.003, R = +0.110 at +3.56σ), and IBM Eagle r3 (p_depol = 0.005, R = +0.086 at +3.06σ). Willow's published gate fidelities sit between these two operating points; the expected hardware verdict is STRONG PASS or WEAK PASS at the full 19-qubit tier. Cross-architecture pre-registrations for Willow and IBM are deposited in parallel with identical thresholds, anchored to timestamped git SHAs prior to hardware access. The octonion / 8-spinor layer is not required. Earlier drafts placed storage inside an octonionic envelope to leverage Paper 27's interior-volume expansion factor; in simulation, the octonion's Coxeter attractor thermalises content across the envelope boundary and degrades readout contrast. The tesseract-only architecture reported here recovers cleaner readout on strictly fewer qubits. The octonion layer remains relevant for the gravity-sector physics of Papers 20 / 27 / 28 but is not part of the QC-architecture proposal. Position in the Merkabit Research Series. Papers 24–26 confirmed 5/5 pre-registered single-cell hardware predictions on IBM Eagle r3 and Heron r2. Paper 31 establishes the cross-chiral tunnel as the directional ternary write primitive with topology-independence across triangle / 4-square / hexagon rings. Paper 32 establishes the 4-square substrate as a content-addressable memory register holding ~5 bits across 10 Coxeter periods. Paper 33 composes write, communicate, store, retrieve, and clock into a primitive-complete 19-qubit quantum computer with native Z₃ cyclic clock. Paper 34 extends the architecture to two coupled compute triangles on 28 qubits and reports the cyclotomic Z₃ ternary spectrum predicted by Paper 7's Regime 2 framework.
This paper establishes the 4-square merkabit substrate as a content-addressable memory register on present-day superconducting hardware. Where Paper 31 specifies how trit information is written into the lattice via the cross-chiral tunnel, Paper 32 specifies how it is held and read out — the memory-side counterpart to the cross-chiral write primitive in the Merkabit Research Series. The substrate. Four 4-spinor tesseract merkabits arranged on a closed perimeter ring (N = 4, 17 qubits including SWAP-test ancilla), with cross-chiral tunnels iSWAP^J(u_i, v_{i+1 mod N}) on every edge at J = 0.1, executing Protocol 4S internal chiral dynamics on each cell. The geometry is forced rather than chosen: the 4-square is the smallest closed register on the Eisenstein lattice ℤ[ω] that admits both bond-level locality (inherited from Paper 31's universal 9-entry lookup table) and a complete Z₃ address space of 3⁴ = 81 ternary patterns. Two complementary measurements establish the register. Persistence. Six deterministic memory configurations M1–M6, spanning mixed labels and two of three uniform Z₃ classes, produce bond-pattern fingerprints that remain mutually distinguishable across horizons n ∈ {1, 2, 3, 5, 10} Coxeter periods at p_depol = 0.005. Mean pairwise distances grow from 0.70 at n = 1 to 0.84 at n = 3 — an early-period amplification feature structurally analogous to the first layers of a trained neural network, but achieved without training — before slow decay to 0.51 at n = 10. All six memories remain mutually distinguishable across every horizon tested. Capacity. Full enumeration of the 3⁴ = 81-pattern address space at the peak-distinguishability horizon n = 3 and the hardware-viable horizon n = 1 yields 51 mutually distinguishable patterns at the 3σ shot-noise floor for 4096-shot SWAP-test readout — approximately 5 classical bits of ternary content held on the substrate without external scaffolding. Section 5 calibrates the 4-square damping factor at p_depol = 0.005 (Willow- and Eagle-realistic) on three representative patterns to fix the operational margin. The register meets the three substrate-level requirements of an associative memory: content-specific distinguishability, temporal persistence of distinct content over multiple Coxeter periods, and a complete addressable alphabet. Distinguishability between memories actually grows for the first three Coxeter periods before noise damping dominates — a Hopfield-style early-amplification regime that is architectural, not engineered. Observable 18 pre-registers the persistence + capacity result on IBM Eagle r3 / Heron r2 (17 qubits, ≈ 25 QPU-min) with explicit pass/fail thresholds anchored to a public commit SHA. Each threshold names the specific structural primitive — content-specific distinguishability, temporal persistence, or ~5-bit capacity at moderate discriminability — that is falsified if the threshold fails. Position in the Merkabit Research Series. Papers 24–26 confirmed 5/5 pre-registered single-cell hardware predictions on IBM Eagle r3 and Heron r2, including the sub-Poissonian Fano signature F = 0.856 ± 0.030 at 131σ below Poisson. Paper 31 establishes the cross-chiral tunnel as the directional ternary write primitive and demonstrates topology-independence of its 9-entry lookup table across triangle, 4-square, and hexagon rings. Paper 32 closes the memory-side of the substrate at the 4-square scale, establishing persistence and capacity as the empirical content of the merkabit's associative-memory register. Paper 33 composes the write primitive and the memory register into a primitive-complete 19-qubit self-hosting quantum computer with a native Z₃ cyclic clock α → γ → β → α (Pentachoric Verification Protocol, Observable 22). Paper 34 extends the architecture to two coupled compute triangles on 28 qubits and reports the cyclotomic Z₃ ternary spectrum predicted by Paper 7's complementary Regime 2 framework.
Two 4-spinor tesseract merkabits coupled through a partial iSWAP^J between the forward spinor of A and the inverse spinor of B produce an ordered-pair Z₃ × Z₃ correlation that is inaccessible to any local observable on either merkabit alone. Under Protocol 4S internal dynamics — three isoclinic rotation planes plus the chiral P gate — at J = 0.1 and n = 2 Coxeter periods, the input pair (β, γ) gives tunnel coherence 0.000 ± 0.000 via destructive interference while the reversed pair (γ, β) gives 0.762 ± 0.005, a 166σ directional gap on Cirq state-vector simulation (125σ on Qiskit Aer). The cross-chiral tunnel is a directional ternary gate: its rule lives in the channel between merkabits, not in either merkabit individually. Replicating the two-merkabit protocol on three ring topologies — triangle (N = 3, 13 qubits), 4-square (N = 4, 17 qubits), and hexagon (N = 6, 25 qubits) — with cross-chiral tunnels on every bond produces the same 9-entry ordered-pair lookup table at every bond, agreeing within 0.02–0.03 across topologies. The β → γ destructive zero is identically 0.000 on every bond of every topology. The architecture is therefore a Z₃-symmetric quantum cellular automaton with a derived local rule, not a gauge theory with loop-level invariants: within-phase-class variance exceeds between-phase-class variance on every observable tested, falsifying the Z₃ plaquette-holonomy interpretation. Register capacity scales with bond count E rather than vertex count N, giving native edge-level parallelism on the Eisenstein coordination lattice. A five-channel stress test (depolarising, amplitude damping, phase damping, combined amplitude + phase, and IBM Eagle r3 via FakeSherbrooke) across J ∈ [0, 1.0] and n ∈ [1, 10] establishes the hardware-viable operating point at n = 1, J ∈ [0.05, 0.13] or [0.2, 0.5]. At this point the predicted directional gap sits in [0.09, 0.30] with significance 6.4σ (FakeSherbrooke) to 37σ (uniform depolarising) at 6 repeats × 4,096 shots. Phase damping is nearly transparent to the signal — the (β, γ) tunnel zero persists at exactly 0.000 under pure T₂ dephasing — identifying the cross-chiral tunnel as a natural decoherence-free subspace for ordered-pair ternary information. Three pre-registered hardware observables operationalise the protocol on present-day processors. Observable 14 (2-merkabit tunnel, 9 qubits, ≤ 45 QPU-min) targets the directional gap with four sub-tests: 14a gap magnitude, 14b sign persistence, 14c information retention versus local observables, 14d phase-damping enhancement via dynamical decoupling. Observable 16 (triangle lookup table, 13 qubits, ~25 QPU-min) validates the 9-entry table on the Eisenstein-native minimum ring. Observable 17 (4-square parallel-bond pattern, 17 qubits, ~10 QPU-min) validates multi-bond locality on a 4-coordinated ring. Each observable carries explicit pass/fail thresholds anchored to public commit SHAs, and each names a specific structural primitive that is falsified if its threshold fails. Agreement across Observables 14 + 16 + 17 within 0.05 confirms the cellular-automaton interpretation. Position in the Merkabit Research Series. Papers 24–26 confirmed 5/5 pre-registered single-cell hardware predictions on IBM Eagle r3 and Heron r2, including the sub-Poissonian Fano signature F = 0.856 ± 0.030 at 131σ below Poisson. Paper 31 extends that pre-registration discipline from the single-cell pentachoric ouroboros to the multi-cell tunnel and its lattice-cellular-automaton composition. Paper 32 is a focused companion that extracts the universal 9-entry lookup table and topology-independence result as a standalone deposit. Paper 33 composes the primitive into a primitive-complete 19-qubit self-hosting quantum computer with a native Z₃ cyclic clock α → γ → β → α (Pentachoric Verification Protocol, Observable 22). Paper 34 extends the architecture to two coupled compute triangles on 28 qubits and reports the cyclotomic Z₃ ternary spectrum predicted by Paper 7's complementary Regime 2 framework. Code, raw simulation data, and timestamped pre-registration files are committed at https://github.com/SelinaAliens/tesseract_quantum_implementation. Hardware targets: IBM Eagle r3 / Heron r2 and Google Willow in cross-architecture parallel.
Paper 15 of the Merkabit Research Series We introduce the binary-ternary hybrid architecture and report three results from its simulation on the Eisenstein lattice ℤ[ω]. The flat rotation gap. The detection rate difference between the T₇₅ ternary architecture and the B₃₁ binary architecture is constant across four orders of magnitude of error rate (3×10⁻¹ down to 10⁻⁴), with coefficient of variation of the detection gap below 0.05 within each cell size. The gap varies with cell size — 24.0 pp (7-node), 15.5 pp (19-node), 13.3 pp (37-node) — but is strictly flat within each, across 42 independent measurements (14 error rates × 3 cell sizes, 50,000–200,000 Monte Carlo trials per point, seed 42). The counter-rotation advantage is a structural constant of the architecture, not a dynamical or error-rate-dependent effect. Two-scale independence. The merkabit's error correction operates at two independent scales: an intra-unit toroidal mechanism (Scale 1, S_torus = 35×, directly measured on the Eisenstein torus with periodic boundaries) and an inter-unit torsion channel mechanism (Scale 2, 35× on the toroidal T₇₅ lattice). The T₇₅/B₃₁ suppression ratio has CV 0.06–0.09 across all error rates, confirming genuine independence. Combined suppression with Level 1 π-lock cancellation (f_sym = 0.5, 2×) gives S_total = 35 × 35 × 2 = 2,450× — three independently measured quantities, zero tunable parameters, zero qubit overhead. At ε = 10⁻³ the 19-node and 37-node pure ternary cells achieve zero logical errors in the Monte Carlo sample. Asymmetric hybrid coupling. The optimal coupling is κ_intra = 0, κ_inter = 1.0, confirming the two scales are physically distinct. The hybrid achieves 8.6–22.6× suppression across cell sizes (including Level 1 π-lock cancellation) as a migration path: B₃₁ qubit algorithms run unmodified with fidelity > 0.999 under T₇₅ protection. Three falsifiable experimental predictions follow directly from the flat gap, testable on existing hardware today.
The merkabit is a balanced ternary computational unit based on dual spinors (u, v) ∈ S³ × S³ with counter-rotating phase evolution at π-lock. The Eisenstein lattice ℤ[ω], balanced ternary logic {−1, 0, +1}, and the gate set {Rₓ, Rz, P, F, C-SWAP} are derived — not assumed — from this definition. The governing algebra E₆ enters via the McKay correspondence from the binary tetrahedral group P₂₄ with zero free parameters; E₆ is uniquely isolated from its Langlands companions B₆ and C₆ by the McKay constraint and the self-duality condition h = h∨ = 12. Three algebraically independent routes yield α⁻¹ = 137: Route A (phase-space counting: 7 × 24 − 31), Route B (Casimir–Eisenstein: N(12 + 5ω) + dim(so(8)) = 109 + 28), and Route C (zero-point Berry phase: −ln F/10, where the return fidelity F is transcendental with two exact rational layers capturing 99.999966% of −ln F). A three-order correction from the E₆ Coxeter spectrum gives α⁻¹ = 137.035999083, matching experiment to 0.005 ppb. Error correction is intrinsic and geometric at three nested levels (π-lock, pentachoric complementarity, E₆ root syndromes), with composite suppression of 3,500–6,200× at ε = 10⁻³. The fault-tolerance threshold is ε ≈ 22%; the Peierls bound on the intrinsic torus gives ε_th ≈ 43 — both far above the surface code's ~1%. The ouroboros cycle (period 12) is a discrete time quasi-crystal with Z₂ topological order and Z₃ subharmonic character. The F gate reduces long-range coupling from O(d) to O(1). Appendices A–P (46 parameter-free simulations, Python 3 + NumPy only) verify every testable claim.
This paper reports three hardware measurements on IBM ibm_strasbourg (Eagle r3, 127 qubits) on 9 April 2026, confirming two further predictions from the merkabit architecture's Appendix N protocol and completing the rotation gap sweep. Combined with Paper 24, four of the five pre-registered, parameter-free predictions are now retired on quantum hardware. P1 — Berry phase (confirmed): Ramsey interferometry on the dual-spinor pair (zero CX gates, transpiled depth 6) traces Berry phase accumulation across n = 1–12 ouroboros steps. The forward–reversed chirality difference shows the predicted sign flip at n = 6 → 8, peaks at n = 10 (96.4% fidelity vs ideal), and returns to near-zero at n = 12 (cycle closure). Z₂ antisymmetry holds at every step count (max residual 0.048). At n = 4 the hardware matches the ideal to 99.8%. P2 — Quasi-period (confirmed): Stroboscopic return probability P(|00⟩) across n = 1–59 ouroboros steps (30 circuits, single batched IBM job, zero CX gates, depth 6) oscillates at 96–99% fidelity with no visible decay envelope. The predicted quasi-period peak at 3.3T appears at n = 39 = 3.25T (97.9% fidelity), within the ±0.3T tolerance. A secondary peak at n = 57 = 4.75T confirms quasi-periodicity. Tau=5 Fano gap: The 3-merkabit triangle rotation gap (9 qubits, transpiled depth 279, 108 CX/ECR gates) shows the first directional Fano gap: paired F = 0.506 vs control F = 0.564 (gap 0.058 favouring the merkabit). Per-round Fano is stable at 0.487 ± 0.006 across all five syndrome extraction rounds. Appendix N scorecard: P1 Berry phase: sign flip, peak, cycle closure — Confirmed (this paper) P2 Quasi-period 3.3T ± 0.3T — Confirmed at 3.25T (this paper) P3 Z₂ chirality symmetry < 5% — Confirmed at 0.016 (Paper 24) P4 Centre-node detection rate 91% ± 5% — Confirmed at 99.3% (Paper 24) P5 DTC robustness under 10% gate noise — Pending All P1 and P2 circuits are pure single-qubit rotations that transpile to depth 6 regardless of the number of ouroboros steps. They test the E₆ Coxeter angle table, not hardware coherence. No error mitigation was applied. All results are derived from raw hardware counts. This is Paper 25 of the Merkabit research series. Four of five parameter-free predictions were retired on IBM hardware within a single week of quantum processor access (6–9 April 2026), with no prediction missed or revised after the fact. Repository: github.com/SelinaAliens/merkabit_hardware_test — all scripts, raw JSON counts, and IBM job identifiers. Keywords: merkabit, Berry phase, Ramsey interferometry, quasi-period, discrete time quasi-crystal, Fano factor, sub-Poissonian statistics, IBM Eagle r3, pentachoric code, dual spinor, E6 geometry, ouroboros cycle, Coxeter number, zero free parameters, quantum error correction, ibm_strasbourg, Standing wave