NONDUM: Leggett-Garg on the order parameter of an N-qubit cat on a superconducting processor, N = 5 to 11, with an ideal-negative-measurement arm, a decoherent twin and same-job Ramsey companions. A registered experiment
On the IBM Heron processor ibm_kingston, N = 5, 7, 9 and 11 qubits on a line are driven by one tick of the unitary exp(-i theta X^N) (2 theta = pi/3), implemented as a parity ladder into a field qubit; the tick tunnels coherently between |0^N> and |1^N>. Q(t) is the majority sign of the N qubits. Two ticks with a waiting time between them give the three-time Leggett-Garg sum K = C01 + C12 - C02, whose classical bound is 1 and whose quantum value in the cat subspace is 3/4 + 3c/4 with c the surviving cat coherence. The intermediate measurement is taken in three ways: projectively, as an ideal negative measurement through an ancilla coupled to one agent (read at the end, no-click shots counted), and on a decoherent twin of the same state (a Z on one agent in half the shots). One echo pulse on the agents at the midpoint between the ticks refocuses static detunings; the decay of c with the waiting time is compared with the single-qubit Ramsey decay of the same agents, measured in the same jobs with the same delays and the same echo. Lines, thresholds and rules were fixed and hashed before the first registered job (registration SHA-256 5134d1c4015403d08e36511c3db6cedf96304ace05595171d3bc06e87a3bd689). Results, as they fell: at tau = 0, K = 1.274, 1.263, 1.221, 1.137 for N = 5, 7, 9, 11 (projective measurement at t1), 1.264, 1.199, 1.173, 1.112 with the ideal negative measurement, and 0.700, 0.745, 0.751, 0.723 for the decoherent twin with no signalling in time within 0.03; all four violations hold with the ideal negative measurement too (both lines exploratory by the registered Monte-Carlo rule, held). The twin line (binding) held. The INM control line (binding) did not hold: the coupling block on a classically prepared chain moved <Q2> by 0.071 for N = 7 against a registered 0.06 (at most 0.010 for the other N). The scaling line (exploratory) did not hold: the cat's decay coefficient, against the sum of the agents' own Ramsey coefficients, gave an exponent of 2.41 (bootstrap 1.78 to 3.08) instead of the 1 of independent dephasing; the cat decays slower than the sum of its agents for the short chains and almost as fast for the long one, so its rate grows faster with N than the sum, closer to the registered correlated-noise alternative. The coherence read from K and from the twin agree at every point. 20 (N, tau) points, 20 jobs, one device, one day. Closest prior work: Ku et al. 2020 (macrorealism of cat states up to N = 6 on IBM), Zindorf et al. 2025 (parity-based no-disturbance test up to N = 38 on Heron), Ozaeta & McMahon 2019 and Kam et al. 2024 (linear-in-N GHZ decay on IBM), Lambert et al. 2016 (majority-sign Leggett-Garg observable, theory). What is new is the K(N, tau) surface on a collective observable with an INM arm, a decoherent twin and same-job Ramsey companions. Nothing here is a claim of quantum advantage. The deposit holds the registration, the frozen scripts, the registration script with its Monte-Carlo, the smoke tests and the looks of the day on two devices (with and without the echo), the rehearsal on the noisy simulator, the complete run log with the raw bits of every circuit, the verdict, the write-up (nondum_paper_v1.pdf) and the scripts and figures that build it. Licence: CC BY 4.0.
On the IBM Heron processor ibm_kingston, N = 5, 7, 9 and 11 qubits on a line are driven by one tick of the unitary exp(-i theta X^N) (2 theta = pi/3), implemented as a parity ladder into a field qubit; the tick tunnels coherently between |0^N> and |1^N>. Q(t) is the majority sign of the N qubits. Two ticks with a waiting time between them give the three-time Leggett-Garg sum K = C01 + C12 - C02, whose classical bound is 1 and whose quantum value in the cat subspace is 3/4 + 3c/4 with c the surviving cat coherence. The intermediate measurement is taken in three ways: projectively, as an ideal negative measurement through an ancilla coupled to one agent (read at the end, no-click shots counted), and on a decoherent twin of the same state (a Z on one agent in half the shots). One echo pulse on the agents at the midpoint between the ticks refocuses static detunings; the decay of c with the waiting time is compared with the single-qubit Ramsey decay of the same agents, measured in the same jobs with the same delays and the same echo. Lines, thresholds and rules were fixed and hashed before the first registered job (registration SHA-256 5134d1c4015403d08e36511c3db6cedf96304ace05595171d3bc06e87a3bd689). Results, as they fell: at tau = 0, K = 1.274, 1.263, 1.221, 1.137 for N = 5, 7, 9, 11 (projective measurement at t1), 1.264, 1.199, 1.173, 1.112 with the ideal negative measurement, and 0.700, 0.745, 0.751, 0.723 for the decoherent twin with no signalling in time within 0.03; all four violations hold with the ideal negative measurement too (both lines exploratory by the registered Monte-Carlo rule, held). The twin line (binding) held. The INM control line (binding) did not hold: the coupling block on a classically prepared chain moved <Q2> by 0.071 for N = 7 against a registered 0.06 (at most 0.010 for the other N). The scaling line (exploratory) did not hold: the cat's decay coefficient, against the sum of the agents' own Ramsey coefficients, gave an exponent of 2.41 (bootstrap 1.78 to 3.08) instead of the 1 of independent dephasing; the cat decays slower than the sum of its agents for the short chains and almost as fast for the long one, so its rate grows faster with N than the sum, closer to the registered correlated-noise alternative. The coherence read from K and from the twin agree at every point. 20 (N, tau) points, 20 jobs, one device, one day. Closest prior work: Ku et al. 2020 (macrorealism of cat states up to N = 6 on IBM), Zindorf et al. 2025 (parity-based no-disturbance test up to N = 38 on Heron), Ozaeta & McMahon 2019 and Kam et al. 2024 (linear-in-N GHZ decay on IBM), Lambert et al. 2016 (majority-sign Leggett-Garg observable, theory). What is new is the K(N, tau) surface on a collective observable with an INM arm, a decoherent twin and same-job Ramsey companions. Nothing here is a claim of quantum advantage. The deposit holds the registration, the frozen scripts, the registration script with its Monte-Carlo, the smoke tests and the looks of the day on two devices (with and without the echo), the rehearsal on the noisy simulator, the complete run log with the raw bits of every circuit, the verdict, the write-up (nondum_paper_v1.pdf) and the scripts and figures that build it. Licence: CC BY 4.0.
Two registered experiments with the QITE equilibrium law on the IBM Heron processor ibm_marrakesh, one day apart in nothing but the place where the law runs: outside the processor on a microcontroller board, and inside one shot in the processor's own feed-forward. Each experiment has its own registration, hashed before its first registered job, its own frozen scripts, complete raw data, verdict and write-up; the two archives in this record are independent of each other. QITE Hold (zenodo_qite_hold_20260923.zip, SHA-256 97853e20d86c146c16d6011191e9c038dddc743dfd5392476993187654f6e1d3; write-up custodia_paper_v1.pdf; working name Custodia). A microcontroller board (ESP32-C3) holds one pair of numbers per qubit for 144 qubits and turns each pair into a rotation angle. Every tick, one circuit of 144 single-qubit rotations is measured 4096 times; the fraction of ones per qubit goes to the board, the board updates its numbers by its fixed law and returns the angles of the next circuit; it never reads the processor's calibration. A registered programme of eight blocks ran over 280 ticks and 290 jobs: square waves, a hold, a staircase, a checkerboard of 18 zones up and 18 down, two state shocks written into the board, two disturbances written into the circuits, and a comparison with an open-loop reference of frozen angles. Of the seven binding lines, six were run and held; the seventh (drift against the open-loop reference) could not be run because the device did not drift by the registered minimum. The square waves came back with a gain of 0.231 of the commanded swing (the law: 0.25), the checkerboard put 98.5 % of (tick, zone) pairs on the commanded side, the disturbances were rejected to 0.557 and 0.539 of their open-loop size (the law: 4/7). A first registration (v1.0) ended at tick 130 through a design flaw of the host and is included in full. Registration SHA-256 1de82ab12a53c1da349196e8ef07851e6b2f176efc6bd1ca6ba14cacd1f7d54b. QITE In-Shot (zenodo_qite_inshot_20260923.zip, SHA-256 a087f88a95f17c6f8023d5ff40aff35901b5e3e87db531dcd76728511f7ae944; write-up qite_inshot_paper_v1.pdf). On each of 54 qubits, twenty rounds of a measurement followed by a rotation conditioned on that measurement, with angles that set the target level and the memory g; the records of a qubit form a two-state Markov chain whose second eigenvalue is 1 - g. The lines test that eigenvalue, not the level: the autocorrelations of the records for g = 0.3, 0.6 and 1.0 were +0.607, +0.366 and +0.046, the two memories in the ratio 1.657 (the law: 1.75); the relaxation from |0> came back at 0.690 and 0.446 per round (the law: 0.70 and 0.40, lowered by T1 during a round); a step of the target proceeded at 0.658 per round and, with g = 1, completed in one round; a kick of the state was followed by a return to within 0.012 of the unkicked hold; over the same twenty rounds an open-loop state lost 0.451 while the closed loop held within 0.006; 98 % of the qubits with a clean read-out showed their own rate within the registered band. All seven binding lines held; 17 jobs. The size is what the service's compiler accepted on the day (144, 72 and 64 qubits failed with an internal compiler error; every attempt is recorded with its job id). Registration SHA-256 51933de4196bfdd7998e6a5305f7576bb7d35ac137e8b6ce0109965372b4fe5c. Nothing here is a claim of quantum advantage: the observables are classical bits and the circuits contain no two-qubit gate. The claim of both experiments is one of control: a fixed law, registered in advance, holding a public quantum processor at commanded levels with the numbers the law says, once from a board over minutes and once inside the coherence time of the qubits. The board's firmware and the serial link layer are not part of the record; both are identified by their SHA-256 in the registration of QITE Hold. Licence: CC BY 4.0.
If quantum mechanics had a state-dependent, nonlinear term of the kind proposed by Kaplan and Rajendran (Phys. Rev. D 105, 055002), the two outcomes of a measurement would not be closed to each other: what is done in one outcome would appear in the other as a weak background field, a possibility first noted by Polchinski for Weinberg's nonlinear theory (the "Everett phone"). In linear quantum mechanics, in any interpretation, the effect is exactly zero. Laboratory experiments have bounded the electromagnetic parameter at the level of 10^-11 to 10^-12. This deposit holds a test of the same effect in which the measurement, the sender and the receiver are two qubits and the control electronics of one publicly accessible processor. On the IBM Heron processor ibm_fez a switch qubit is measured inside the circuit. If it reads 1, the controller fires N signed pi pulses (N = 200 to 1600) on a sensor qubit 25 edges away; if it reads 0, the sensor rests for exactly the same time on the y axis of its Bloch sphere, where a small rotation about x appears linearly in <Z>. A leak would rotate the resting sensor by epsilon p N pi with the sign of the drive, p being the Born weight of the outcome in which the drive was fired (1/2 and 2/3). The session has four arms: the measurement arm; a classical twin in which the switch is prepared instead of being in superposition (the same pulses and feed-forward, no other outcome); a sender check; and the response of the sensor to a real rotation injected over the window, measured in the same session. The sensor and the switch were chosen at submission by a registered rule from a scan job. Which of the two sign labels is the + sign was sealed before the first job and opened after the lines were formed. Lines, thresholds, the layout rule and the analysis were fixed and hashed before the first job (registration SHA-256 fbdec5d40b76af84b1fc0d98f71ee0d534f454a1ea96052ba5059077ce16d134; a first version was replaced before any job ran under it and is part of the deposit). Results, as they fell: 1060 circuits, 844,000 shots, 11 jobs run back to back in five minutes. The slope is epsilon = (+0.8 +- 4.5) x 10^-6, 0.18 standard errors from zero; |epsilon| < 8.2 x 10^-6 (90 %). The silent sender, the classical twin and the split by the preceding shot are flat at all eight settings (at most 1.9, 1.7 and 1.3 standard errors against a registered limit of 4), the sender check separates the two signs (0.82 and 0.65 against a floor of 0.5), and the response of the sensor was 0.90, 0.86, 0.79, 0.62 for the four windows. All nine binding lines and the four attestation lines held. The bound is five to six orders of magnitude weaker than the laboratory bounds on the same parameter and competes with none of them. What the experiment adds is the place: the test runs on a public device, the sensor listens within microseconds of the measurement, and anyone with access to such a processor can run it again. Nothing here is a statement about the existence of other worlds. The deposit holds the registration and its superseded first version, the shared module with the circuits, the design and the estimators, the model and registration script, the runner, the verdict script, the logs of two preliminary looks on the device, the rehearsal of the whole session on the noiseless simulator under the hashed registration, the session log with the bits of every shot of every circuit in the order of execution, the seal file, the verdict (formed once), the ledger, the write-up (nuntius_paper_v1.pdf) and the scripts and figures that build it. Licence: CC BY 4.0.
A three-qubit GHZ decoder proposed in the earlier Q1 spin-star preprint transfers a global phase sign to one central qubit. This follow-up traces its implementation on Rigetti Cepheus-1-108Q (centre q107, neighbours q106 and q98) and asks whether its readout can distinguish Theory Φ from standard quantum mechanics. It connects the theoretical preprint, the book The Language of Φ and the programme’s methodological evidence review to the original measurement records. Five completed acquisition stages account for 101,860 physical shots: a 10-shot native timing witness, a 10,240-shot Ramsey measurement of individual and concurrent couplers, 28,082 intermediate calibration and decoding controls, 42,024 shots in the primary GHZ cycle (including 33,792 primary shots in 12 trajectory blocks), and a 21,504-shot frequency pilot. The original Ramsey report figure, full phase-and-contrast scan, intermediate control figures and Part I hardware figures are reproduced. With full-amplitude joint drive the first scan could not identify either phase rate. Reducing the drive to 75% restored measurable responses. In the primary experiment the serial/joint sign-error rates were 12.66%/16.20% for transverse disturbances and 19.59%/21.64% for commuting disturbances. None of the three prespecified criteria for the predicted hardware advantage was met. Carrier retuning improved the worst-case joint inversion from 97/384 to 199/384; the development threshold was 346/384. The published ideal Q1 mapping and a separate standard quantum calculation predict the same 3.522800135 percentage-point transverse advantage at mean jump count π/20. Their difference for this observable is exactly zero. The broader equations reviewed supply no independently specified correction on this register. The release provides original shot records, numerical estimates, program-timing evidence, uncertainty analyses, editable figures and manuscripts, and offline reproduction tools.
On the IBM Heron processor ibm_fez, 144 qubits form 36 domains of four. A field law running on a microcontroller (an ESP32-S3, firmware v4.12) turns the measured populations of every cycle into the rotation angles of the next, and a Hebbian coupling matrix of two images, a unicorn and a heart, makes the two images attractors of the collective: cued with part of either, the field completes it. The experiment writes a blend of the two: of the 16 pixels on which the images differ, k are taken from the unicorn and the rest from the heart (which pixels: fixed in the registration, per trial), the 20 shared pixels as they are. The blend is held for two cycles and released, and the field runs six free cycles on its own. Resolution is the first free cycle in which the field's majority picture agrees with one image on at least 90 percent of the pixels. Lines, thresholds and rules were fixed and hashed before the first registered job (registration SHA-256 81398e6660c6538ebcf1566b3a2ebb6ecbaaefe29da074894848516b5749acda; a first registration was refused by its own budget guard after four burn jobs and is part of the deposit). Results, as they fell: of 13 field-on trials 11 resolved into one of the two images; the fraction resolved to the unicorn was 0.00, 0.00, 0.33, 0.50, 1.00, 1.00 for k = 5, 7, 8, 9, 11, 13, a monotone curve with a single flip whose interpolated crossing lies at k* = 9.0; the censored decision time was 4.4 cycles at the two levels nearest the flip against 1.0 at the levels far from it; the two field-off controls, the same blend with the inter-domain coupling off, never resolved. The device's lean, read from the cycles in which the field was off, pointed to the heart and moved the fair point up by about one pixel, as the registered exploratory line asked. All five binding lines and the exploratory line held. The decision dynamics is the one a textbook predicts, a mean-field quench of 16 domains from the blend's magnetisation; the content of the experiment is that it runs, pre-registered to the pixel, on a quantum collective with the controller in the loop. Two chips were in the loop: the controller (ESP32-S3, with the field, the coupling matrix and the agents drawn on its display during the run) and a witness (ESP32-C3) running the byte-identical kernel source, compiled with the same arctangent and without fused multiply-add, so that the RISC-V chip in software arithmetic and the Xtensa chip with its floating-point unit produce the same bits: a seeded programme of 527 exchanges over every kernel path gave the same vector on both before the run, and on every one of the 148 ticks of the run the witness returned the same bytes as the controller. 148 jobs of 2048 shots, one device, one night. The deposit holds the registration (and the refused first registration with its burn log), the model and registration script, the runner, the verdict script, the kernel replica, the rehearsal on the noiseless simulator with both chips in the loop under the hashed registration, the run log with the measured populations of every qubit in every cycle and the witness flag of every tick, the job ids, the verdict (formed once), the bit-check vectors of the two chips, the write-up (buridan_paper_v1.pdf) and the scripts and figures that build it. The firmware sources and binaries are not deposited; their SHA-256 are in the registration.
Band II of the learning-memory programme: three pre-registered experiments conceived by the author, performed on the Band I platform (an ESP32-C3 controller that learns images by looking and steers 156-qubit IBM Heron processors). Discipulus — a learner with a certifiably empty coupling matrix, given only k weak environment fragments of the world's pointer state, acquires exactly what the world left redundantly (recall 0.361 / 0.722 / 0.917 / 0.861 for k = 0/1/4/8 on ibm_fez) and nothing when the access basis or the pointer value is missing (0.556, 0.500). Arbitrium — measurement independence as a registered dial: a separable source whose hidden variable is one machine's learned model of another fakes CHSH from 0.978 to 3.971, crossing the Bell bound for 0.274 bit of measured settings-source dependence, while a singlet reads |S| = 2.656 and settings correlated only with each other never move it. Praeteritum — Leggett-Garg on a learned record with an Elitzur-Vaidman probe as a per-shot non-invasiveness certificate: K = 1.147 at θ = π/3 on ibm_fez against the macrorealist bound 1, with the pinned-history control at 0.538, the certified record intact at 0.941 and the idle apparatus moving nothing (0.013). Five hardware sessions, 39 binding verdict applications generated from the deposited ledger, 36 passed, 3 missed and reported as they fell; each miss answered by a re-registered protocol, not a rewritten verdict. The deposit contains the 19-page paper, registrations with complete Monte-Carlo distributions, test laboratories, runners, dress rehearsals, raw per-job records, manifests, figure and animation scripts.
WITHDRAWN (v1.1, November 2026). The main claim of this preprint — that πνZZτ* ≈ 0.873 rad is an observed invariant across IBM and IQM superconducting devices — is withdrawn by the author. 0.873 was an input, not an observation. In every IBM run (2026-03-28 to 04-19) the delay was set to τ = 0.873/(πνZZ) from the measured νZZ before execution; πνZZτ recomputed from the run logs is 0.8730 for every run, so the constancy in Table 1 holds by construction. Table 1 mixes values from different sessions and experiments. Only the Q13–Q14 row matches a logged run at the listed τ*; other rows match a calibration-file plan or Bell-clock results at other delays. The reported GPS is not the quantity defined in the paper. The logged value for Q13–Q14 (+0.199) is Fff − Fstatic from a dynamic mid-circuit-measurement protocol, not Pff − Pref for the static circuit described. Pair selection. Eight IBM pairs were run; four are reported. IQM. The raw IQM scans (2026-04-14/15) give GPS at τ = 7 µs of −0.006 ± 0.007 (all 18 values); the tabulated +0.026 ± 0.011 (n = 5) could not be reproduced. No IQM optimum is supported. The measurements remain available as records of GPS at the set operating point πνZZτ = 0.873; they do not test whether this point is optimal. The derivation φ*ff = arg⟨eiφ⟩ (Sec. 2) is unaffected. The PDF contains the withdrawal notice followed by the original text unchanged.