Verified adaptive teaching with native R cells

Networks of R hybrid cells can carry out finite adaptive teaching programs in which latertrials depend on the state written by earlier trials. Pulses change the input log-gains of freshlearner cells. A read pulse then tests those gains, and its report can trigger further trials or alesson and credit write. Native AND and NOT gates connect the reports.For any positive integer K, a single compiled network identifies a hidden need m ∈ {1, …, K}by binary search. The router, timing and initial pulse queue are fixed across needs. Only theselected leaf receives its declared lesson and credit. A separate nine-cell experiment deliverslessons of 8, 7 and 6 pulses. Cutting one feedback wire prevents the second improvement.Lean proves these results in upward-crossing semantics for independent, fixed fabricationparameters in the ±0.3% or ±10% box. For a fixed setup, complete executions agree onterminal observations and each cell’s ordered firing times.