The decision procedure
Three experiments decide it
Each is cheap enough to run on equipment the group already owns, and each has its obligation written down before the result exists. A claim without a stated falsifier is a slogan.
Written in advance
What every possible result obliges us to say
The order matters. E1 gates everything: a negative species result closes the compute programme outright, and no later result can reopen it.
E1
Species
Is the mobile carrier in our layer a proton, and not something else?
Method. Isotope substitution and humidity control on the same films, with electrical response measured across both. A protonic carrier must show the expected mass and hydration dependence; ionic or electronic conduction will not.
If it passes
The carrier is identified as protonic. Only then may any device-level statement about OPALBLACK be made at all.
If it fails
No computing claim is ever made about OPALBLACK, on any sheet, by anyone in the group. The programme stops here and the material returns to being an optical and structural material.
E2
Switching
Can the state be written and read repeatably, quickly, and with low energy?
Method. Pulse-train programming of the layer between conductance levels, measuring write energy, symmetry, linearity and the number of distinguishable levels at room temperature.
If it passes
A quantified device figure: levels, energy per write, and speed — stated with error bars and compared against published protonic and memristive devices, not against a marketing number.
If it fails
OPALBLACK is not an analog-weight device. The proton transport may still be scientifically interesting; it is not a compute story, and the sheets say so.
E3
Retention
Does the state stay put for long enough to be worth writing?
Method. Hold programmed levels and track drift against time, temperature and hydration. Retention is measured, then extrapolated with the extrapolation stated openly rather than folded into the headline.
If it passes
A retention window with a use case attached — inference weights, not archival storage, unless the data says otherwise.
If it fails
The device is volatile. That narrows the target to workloads that tolerate refresh, and the claim narrows with it in the same sentence.
None of these experiments proves a computer. Together they decide whether an oxide grown in a bath can hold analog state beside a processor — which is the only question this ledger asks.