Research round eight: learning stability and neural adaptation
How neural systems change without losing dependable function, and the three engineering failure tests that reading produced.
The question this round asks
Earlier rounds studied how specialists coordinate and how sources are verified. This one studies change: temporary learning traces, the regulation that keeps activity stable while a circuit changes, non-neuronal contributions to memory, adaptive myelination, drifting neural representations, and value signals that depend on the comparison set.
Three experiments we would run first
Test whether an irrelevant extra option changes a recommendation between two unchanged options. Test whether a reviewer's correction reaches the operation that caused the error, and only that one, after a delay. Test whether a typed interface preserves the correct decision when a specialist's model version is swapped, while still surfacing a genuine regression.
What we are deferring
A workload controller waits until execution traces show self-amplifying work. Transport and scheduling optimisation waits until timing dominates latency. Astrocyte-inspired promotion rules stay exploratory because they overlap our existing memory program.
The honest limits
Much of the mechanistic evidence comes from rodents, nonhuman primates or cultured tissue, and human cells transplanted into mice remain a mouse experiment. None of the cited work tested Encevra or Noetis, none of the authors have evaluated our systems, and nothing here reproduces a human brain. These are engineering hypotheses to be judged on their own results.