A Ring of Beaches

Six beach neurons on one chart. Each one's axon pours into a sandbar on every other beach, so every beach has a mouth for every neighbour plus one sensory shore. Teach them an order, then wake the first one and watch whether the rest follow.

Who spilled when (last 1,500 steps)

Sandbars: from (row) to (column)

Teach an order

Ready.

Rules of the coast Anti-wave keeperEvery spill sends a flattening wave into the other pools (feedback inhibition, basket cells). Sand budget per beachMouths share a fixed amount of sand: one deepens, the others give a little back (synaptic scaling). Wait for the tideDuring training, footprints only set when the tide comes after each lesson. Speed

Last wake-up

Wokennone yet

What the tests found

Measured headlessly with node tests/experiments.js on 30 seeds the settings were never tuned on (seeds 101–130). Each seed picks a random order of the six beaches, trains it 15 times, then wakes the first one once with all sand frozen.

Condition	Beaches that followed, in order (of 5)	Whole order replayed	Run backwards from the last	Mean sandbar: next / previous / others

Shuffled: the same 15 lessons, but a new random order each time. Anti-wave keeper and sand budget are switched off one at a time. With eight beaches, a fixed sand budget is too tight and the chain breaks after about 5 beaches (7 of 30 replayed all 7 followers); giving each mouth the same share of sand (a change made after seeing that run) brings it to 29 of 30. Details and every number are in the README.