Experimental roadmap for VELORYN v3
This is a proposed validation program, not a report of fabricated hardware or experimental data. The full model, proofs, numerical checks and claim boundaries belong in research/. Numerical agreement in this release is not laboratory validation. A physicist should verify the apparatus model, units, calibration and statistical plan before collecting confirmatory data.
The target is simultaneous agreement between independently measured gate timing, spectral response and the memory–survival endpoint law. Such agreement would validate this engineered controller model; it would not establish a modification of quantum mechanics or thermodynamics.
Common preparation and records
Freeze the protocols, thresholds, analysis code and held-out validation points before the confirmatory runs. Record raw signals, controller commands, calibration runs and failed runs with timestamps. Separate calibration data from tests. Report register scaling, dynamic range, saturation, transfer-function uncertainty, readout noise, drift and actual write/dwell/blanking times. Obtain uncertainty intervals from independent preparations or a justified dependent-data analysis; do not replace correlated samples by an invented effective sample count.
In three dimensions use the convention $Av=-b\times v$, $U=|b|^2$, $n=b/\sqrt U$, and launch a pre-gated stroke $a=s n$. Initialize a measured nonzero seed $U_0>0$; take $R_0=|x_0|^2-(x_0\cdot n_0)^2>0$ when testing nontrivial writing.
1. Gate order and exact finite-duration endpoints
Measure. Independently calibrate position and area integration, the held launch direction, filter time constants and current tails. For every stroke record $x,b,a=x'-x$, the gate residual and the area residual $b'-b-x\times a$. For the calibrated finite-rate plant, the shaped command uses $y=140\zeta r^3(1-r)^3/T$, $r=t/T$, and $u=(1+\tau_x\partial_t)(1+\tau_b\partial_t)y$. Verify that the observed terminal currents vanish within the instrument bound; otherwise use the measured correction $b_{\rm eff}=b+\tau_bj$ or include the residual explicitly.
Prediction and control. Compare separately verified pre-, midpoint- and post-gates, $(A+\theta x\wedge a)a=0$, at $\theta=0,1/2,1$. Their prediction is
[ \Delta(|A|_F^2/2)=(1-2\theta) [|x|^2|a|^2-(x\cdot a)^2]. ]
Use legal fixtures for each timing rather than reusing a displacement that becomes illegal. One explicit fixture is $x=e_2$, $a=s e_1$, $b=q e_1+\theta s e_3$, $q\ne0$: its tested gate is parallel to $a$, and $\Delta U=(1-2\theta)s^2$. Midpoint timing also predicts unchanged skew spectrum. Include inverse-gate and zero-area controls.
Falsifying observation. A reproducible wrong sign or endpoint residual beyond the independently propagated error budget, while the actual gate residual is verified small, contradicts the specified controller/plant model.
Boundary. The permission direction is held for a stroke. The model does not require $A(t)\dot x(t)=0$ throughout its finite-duration implementation. An unknown delay, unequal axis response, clipping or overlapping readout requires a revised model, not an instantaneous-update assumption.
2. Independent ridge readout and visible spectral reconstruction
Measure. Hold the registers fixed and read several $F_\lambda=x^T[-A^2(\lambda^2I-A^2)^{-1}]x$ through the separately calibrated harmonic probe, rather than only recomputing them from the controller software. For stiffness $K_\lambda=\omega_0^2(I-A^2/\lambda^2)$, measure the equilibrium variance of $Y=x\cdot q$:
[ F_\lambda=|x|^2- \frac{M_p\omega_0^2}{k_BT_{\rm b}}\operatorname{Var}(Y). ]
Calibrate temperature, stiffness, gain and equilibration independently; account for autocorrelation in variance samples. This readout assumes the model's classical canonical equilibrium; a quantum oscillator requires its separate covariance correction. Compare measured one-step gains with the exact ridge formula and its measured approximate-legality bound.
Prediction and control. If at most $m$ distinct positive squared frequencies are visible to $x$, use $2m$ distinct positive values $z=\lambda^2$ to reconstruct $F_{\sqrt z}=\sum_j r_jt_j/(z+t_j)$. Then test extra, held-out resolutions and independent mode-response measurements. The two-frequency fixture $t=(1,4)$, $r=(2,3)$ predicts values $(4,17/5,8/3,19/10)$ at $z=(1/2,1,2,4)$; this is a calibration target, not measured data.
Falsifying observation. Independently measured ridge gains or held-out responses incompatible with the calibrated formula and uncertainty budget reject the specified probe/controller relation.
Boundary. Reconstruction identifies the visible weighted spectrum, not the full matrix, hidden modes, orientation or permission kernel. Weak residues and nearly coincident poles can make the noisy inverse problem ill-conditioned even when exact identifiability holds.
3. Positive joint-loss endpoint curve without a fitted exponent
Measure. Independently measure the OU correlation time $\tau_h$, stroke integration interval $h$, actual lag covariances, $U_j,n_j$, and bright-mode transmission. Verify stationary pulse preparation and fix $c=h/\tau_h>0$. Use exact writes followed by held loss with exposure $d_h$, calibrated $q_h(u)=e^{-\gamma_hd_hu/2}$, and $\ell_h=\gamma_hd_h\ge\ell_{\min}>0$. Prepare the probe in $n_0$. Report write and blanking time separately from exposure and from the controller clock. Prerecorded stationary OU strokes avoid silently changing the covariance through physical dead times; otherwise verify the actual covariance after those dead times.
Prediction and control. At fixed $c$ and positive target $\ell$, refine $h$ within the verified hardware range while holding $\ell_h\to\ell$. Compute the following curve from independent calibrations and the measured endpoint $U$, with no survival exponent fitted to the test data:
[ S_\ell(U)=\exp\left[-\frac1f\int_{U_0}^{U} F_c(e^{-\ell u/2})\frac{du}{u}\right], \quad f=1-\frac{1-e^{-c}}c, \quad F_c(q)=1-\frac{(1-e^{-c})(1-q)}{c(1-e^{-c}q)}. ]
Compare repeated-probe survival for each recorded controller path, including independently calibrated baseline losses and detector efficiencies. Test several positive $\ell$, the inverse-memory boundary $S_\infty=U_0/U$, and the interval $(U_0/U)^{1/f}\le S_\ell\le U_0/U$. Check dark-line alignment as well as survival. Analyze convergence across resolutions against the localized finite-$h$ bounds; a single finite grid need not equal its limit.
Falsifying observation. A repeatable residual inconsistent with finite-resolution and calibration bounds across refinement rejects this endpoint closure or one of its independently tested assumptions.
Boundary. Finite positive $\ell$ generally gives a curved endpoint relation, not a constant power. The theorem does not cover arbitrary joint paths $\ell_h\to0$, varying $c$ without further estimates, or loss active during the hardware write. Negative-correlation sources need their own measured covariance-filter curve.
4. Classical replay and controller-memory controls
Replay the complete measured $n_j,U_j,q_j$ record without using live geometric feedback. Compare a single-photon probe and a classical-wave probe under the same passive transfer maps. The null prediction is identical normalized conditional attenuation; report raw and efficiency-corrected counts/intensities. Classical agreement is expected and does not invalidate the endpoint theorem.
Reprepare the signal identically while retaining or resetting the area register, using identical future external driver records. Also compare different histories that end at the same complete measured controller state: the stated enlarged-state model predicts identical future statistics. These controls locate memory in the controller. A discrepancy after replay indicates missing apparatus/probe dynamics; it does not by itself establish quantum memory, nonclassicality or an advantage.
5. Correctly prepared thermal reverse experiment and resource floor
Start with a finite, thermodynamically resolved paired-state completion, including controller and finite fuel. Verify detailed balance, temperature and the actual implementation energies; each writing pre-gate must have its post-gated inverse. Independently characterize its Gibbs law $\pi$ and preparation $r_0$. The dimensionless prepared resource is
[ B=\frac{\mathcal F(r_0)-\mathcal F(\pi)}{k_BT_{\rm b}} =D(r_0|\pi), ]
including the full finite controller. Do not estimate $B$ from an area norm. During the autonomous test window exclude unaccounted external work.
Fix a calibrated time-reversal-even certificate and writing threshold. Estimate $p$ in forward runs. For the reverse runs, physically prepare the forward final ensemble $r_\tau$, including fuel and controller state, and use the correctly conjugated dynamics. Estimate the probability $q_R$ of the reversed erasing event. Reversing a video, or restarting from $r_0$, is not this preparation.
Prediction. The prepared finite-resource model requires
[ d(p|q_R)\le B,\qquad q_R\ge e^{-(B+h_2(p))/p},\qquad h_2(p)=-p\ln p-(1-p)\ln(1-p). ]
Propagate uncertainty in both ensembles, reverse preparation and $B$. If no erasures occur in $N$ independent reverse trials, the one-sided upper confidence bound is $1-\alpha^{1/N}$ at confidence $1-\alpha$; zero counts do not establish zero probability. Use joint confidence coverage for the whole comparison.
Falsifying observation and boundary. A reproducible lower bound on $d(p|q_R)$ above an independently bounded $B$ contradicts this complete finite-resource model. First audit omitted work, hidden registers, bath/contact noise and reverse preparation. A same-run erase count can replace $q_R$ only under explicitly verified reverse-law compatibility; then the stationary entropy bound is $\langle\Sigma\rangle\ge k_B(p-q)\ln(p/q)$. None of these relations assigns a universal energy cost per numerical area increment. Ideal constraints coupled to a thermal bath require physical fluctuation–dissipation accounting.
Release and human-verification checkpoint
Publish the apparatus schematic, independently measured transfer functions, units, protocols, calibration uncertainties, raw records, analysis scripts and all exclusion rules alongside any future experimental claim. Have a physicist verify the reverse experiment and a second analyst reproduce the parameter-free curves from raw data. Clearly label simulations, proposed fixtures and laboratory observations separately. This release supplies a testable theory and reproducible numerical work; it contains no verified experimental anomaly.