category00: A Pre-Registered Re-Test of the Simplest Possible Categorical Task, with Status and Context from a Broader Real-Hardware Campaign

We report category00: a hardware-readiness design for a re-test of the simplest possible categorical discrimination task an organoid-simulation closed-loop framework can pose: two electrodes, one driven by each of two labels, a single pulse each, with no continuous magnitude and no shortcut. This exact design, in this project's prior simulation work, returned zero measurable learning signal across every tested drive level, connectivity density, and learning rate. That negative result stands, unrevised. Before writing any code, we wrote down two competing hypotheses and a decision rule to distinguish them: either (H1) Three-Factor STDP genuinely does not drive categorical learning at this scale, in which case a real culture should replicate the simulated null; or (H2) the null result is an artifact of this project's simplified organoid model, in which case real tissue might show a signal simulation could not produce. The decision rule was fixed before any hardware run: a single favourable session does not count, only a result repeated across independent sessions distinguishes a real effect from noise. category00 itself has not yet been run on live tissue. This paper reports that status directly, together with a finding from a broader real-hardware campaign conducted on the same platform since the original design was written, which speaks to exactly the repeatability question this task's decision rule was built around: an isolated single-pair plasticity effect that appeared with a clean, textbook signature at small sample size did not survive when the same measurement was repeated at larger sample size with drift correction, on two independent cultures. The pre-registered caution against trusting a single favourable session is, in that separate test, borne out.