OPH record-law emulator preflight v1
Status: digital logical emulator executed; backend hardware not run Date: 26 October 2026 Evidence: 0 Verifier: node verification/admissibility/verify_oph_emulator_preflight.mjs
Build
Encode the six one-seam occupation keys into a three-qubit logical subspace. Four reusable record ancillas accommodate the largest twelve-label instrument, giving a seven-qubit one-seam build. The adjacent-seam test uses the ten two-load configurations on three active ports plus one spectator, which fit into four system qubits; mid-circuit measurement, logging and reset keep the full design at eight qubits.
The compiler must expose both views of the same operation:
- retain the ancilla outcome to reconstruct the instrument;
- discard the ancilla outcome to reconstruct the unread channel.
The distinction is part of the target, not a software convenience.
Power boundary
At phase zero the named one-step projector probabilities are
| law | probability |
|---|---|
C0 | 0.7285533906 |
C | 0.625 |
E | 0.375 |
T | 0.375 |
C_signed | 0.125 |
E and T are intentionally indistinguishable through g; the separate empty--full k probe distinguishes them. The closest distinct one-step pair is C0 versus C, separated by 0.1035533906. A conservative two-arm Hoeffding design at familywise alpha=0.01 requires 4,470 shots per arm, so the frozen first-stage plan uses 5,000 per phase and law.
Depth is the cost wall. For C, the phase-zero separation from E is 2^(-r-1). At depth four it is 1/32, requiring 49,083 shots per arm under the same conservative bound. At depth eight it is 1/512, requiring 12,565,012. Depths one through four are therefore primary. Depths five through eight proceed only if the fitted visibility, reset and memory model predicts adequate held-out power; exact source arithmetic is not a reason to collect an underpowered physical record.
Stages
| stage | admission gate |
|---|---|
| A | one common classical table and one-step g/k tomography pass |
| B | depths 1--4 plus adjacent reversal and disjoint null pass |
| C | depths 5--8 clear a frozen measured-visibility power calculation |
| D | a second platform is attempted only after one complete conventional instrument/process-tensor model predicts held-out data |
The ideal adjacent-order total-variation distance is 0.25; the disjoint ideal is zero. Five thousand shots per order make the adjacent effect easy in the ideal model. The hard requirement is the null: after self-consistent preparation/readout correction, the inferred underlying disjoint-order distance must remain below total variation 0.02 with a predeclared confidence or equivalence bound. A raw empirical distance is not that bound: the independent reference run obtains 0.0204 from two 5,000-shot samples of exactly identical distributions. The digital Gate-5 execution fixes a conservative two-sample bound and shows that even zero empirical distance needs 39,687 shots per order; its 200,000-shot qualification run passes. If both adjacent and disjoint pairs retain statistically resolved order after correction, the device has uncontrolled memory and the OPH-specific comparison fails.
Interpretation
A pass certifies an engineered realization of the finite record-law benchmark. It does not select OPH as nature's record law or identify the device carrier with spacetime. A failure accounted for by the frozen noise model is a useful compiler or detector result. Only a residual reproduced on an independent platform after instrument and process-tensor alternatives are exhausted becomes eligible for anomaly review.