A memory readout test that can fail
A prospective protocol draft for distinguishing specified material-memory models, with independent specimens and explicit negative outcomes.
The question
Can two differently trained material states produce distinct readouts after a predeclared scalar-relaxation model predicts matched retained state? The aim is to test a bounded memory model, not to prove new physics or disprove every fading-memory system.
Proposed carrier and competing explanations
Use cyclically conditioned elastomer specimens as a candidate material, with force–displacement readout. Conventional Mullins softening and viscoelastic relaxation are both competitors. The current β ≥ 0 oscillator stiffens; it is not a direct model of stress-softening. A softened k = k₀(1 − γm), 0 ≤ γ < 1, would be a separately calibrated law. Crumpled sheets remain another candidate, not an interchangeable replicate.
Matched-history construction
r(A) = αA² + 1/τ m∞(A) = αA²/r(A) T(A) = −ln[1 − mtarget/m∞(A)]/r(A)
Choose a target below both plateaus and train initially reset states at different amplitudes for these durations. The scalar model predicts the same terminal m. For the oscillator it also requires matched q and p and identical future forcing. Resetting mechanical state is an explicit intervention; its energy and disturbance must be recorded. In a material test, define and verify the actual matching variables rather than treating simulated m as measured.
Design to freeze before collecting decisive data
| Item | Protocol draft |
|---|---|
| Experimental unit | Independent specimen; repeated time samples and repeated sweeps are not independent specimens. |
| Arms | Two training amplitudes; sham; matched-exposure control; calibration-only specimens. Randomize specimen assignment and blind history labels to the readout analyst. |
| Readout | One prespecified force–displacement or tangent-stiffness curve per independent specimen. Fix sweep rate, temperature, dwell time, geometry, calibration and normalization after pilot work and before registration. |
| Rivals | Calibrated one-state relaxation; a two-state or stretched-exponential relaxation alternative where identifiable; conventional hysteresis/softening model. Fit calibration specimens only. |
| Primary statistic | Between-history difference in normalized stiffness integrated over a fixed probe-amplitude interval around the lower training amplitude. Fix the interval in the registered protocol. |
| Decision rule draft | A familywise 95% simultaneous interval must exclude the one-state null over the prespecified statistic and exceed a 5% reference-stiffness difference. The 5% is a proposed practical threshold, not an evidence-derived constant; finalize it from instrument capability before registration. |
| Sample size | Use pilot specimen-to-specimen variance and covariance to simulate power for the frozen statistic and 5% effect. Target 90% power at familywise α = 0.05; no sample size is claimed without those data. |
| Failure and inconclusive outcomes | If the uncertainty interval is wholly within ±5%, the proposed practically meaningful effect is not supported under the tested conditions. If it overlaps the threshold, report inconclusive. A rival-model explanation limits the claim to rejection of the simpler model. |
| Registration | This is a protocol draft. No OSF registration, recruitment, data collection or external approval has occurred. Timestamp the finalized plan before the decisive experiment. |
What success would mean
A reproducible residual beyond the declared null and nuisance models would motivate a better material model. An amplitude-specific curve alone does not establish universal memory, learning, or a new force. Archive negative results, exclusions, calibration files, units, raw time series and specimen identifiers. Material-memory background · Why the null must be explicit