organoid-oi v4: Reflexive Word-Length Sentence Assembly — Design, Verification, and Real-Hardware Test Results

We report a hardware-ready simulation design in which a simulated organoid's own firing intensity, which scales directly with input drive, is used by a reward/penalty-calibrated readout layer to correctly order a set of words by length and insert new words into an existing ordering, producing sentence assembly from a simple, honestly-named mechanism. We term this a reflexive demonstration, following the precedent set by Kagan et al.'s closed-loop Pong system: the organoid does not construct novel categorical structure, and we do not claim it does. This system follows, and is motivated directly by, a negative result. Three prior task designs (fruit-category images, Morse-coded anagrams, and Braille-cell discrimination) produced no replicated categorical-learning effect under Three-Factor STDP, after six independently identified and corrected measurement defects. That result is not revised here. In simulation, the dose-response relationship the demonstration depends on held with a correlation of r = +0.999 across eight test words, and a trained readout correctly assembled and updated a target sentence. This paper now reports the result of the next step described in the original design: testing the same untrained dose-response relationship directly on real tissue. Across three independent sessions on two human neurosphere cultures accessed through FinalSpark's Neuroplatform, no relationship between word length and firing intensity was observed, under two separate stimulation encodings. The mechanism that produces the simulated result is intact and correctly described; the physical assumption beneath it did not transfer to either culture tested.