BrainCore is my spiking-network simulator: every brain region is a separate service, the services exchange nothing but spikes, and the rule inside a region is “only neurons and synapses”. No vectors, no fact graphs, no language model in the loop. The first report (11 September 2026) showed a retina driving nine cortical columns and a cochlea driving four, with no learning at all. Step 2 is about learning: can a hippocampal region bind what the eye sees to what the ear hears, and can the ear itself be made good enough that a word sounds the same across voices?
This is a status report, not a result announcement. Step 2 is not accepted. Here is exactly what holds, what does not, and what each experiment costs.
What was built in October
A human ear. The old cochlea was a textbook filter bank. The new one has 1,000 cochlear sections and 32,000 individual auditory nerve fibres, takes sound pressure in pascals at the eardrum at 100,000 samples per second, and spikes. Compression, adaptation, phase locking and recovery after 0.1, 0.3 and 1.9 seconds are checked against published windows. Frequency tuning was wrong at first: the maximum error in the equivalent rectangular bandwidth was 52.3 percent. After the fix it is 12.8 percent. Thirty unit and reference tests pass. The verification file says in its own words: human ear complete, false (verification).
A cochlear nucleus cell at native resolution. The bushy cell that sits between the ear and the cortex is now a 357-section, 1,429-segment compartment model stepped at 1.25 microseconds. 300 milliseconds of it is 559,899,000 scalar samples, and a run can be stopped and restored from a checkpoint without storing the input history: the restored suffix matches the uninterrupted run exactly. Restore takes 0.017 seconds; cold initialisation takes 120.6 seconds (verification). That difference is why checkpoints exist.
Paired learning. Three letter and tone pairs, three checks each, so “9 of 9” means nine checks, not nine letters. On 2 October the course gave tone-to-letter 6 of 9 and letter-to-tone 0 of 9 (rollup). On 3 October both directions reached 9 of 9 with plastic synapses (169,925 weight updates), while the frozen control recalled 0 of 9 with 0 weight updates and there were no false recalls on baseline or unknown input (verification). A training run interrupted and restarted from its checkpoint produces the same output as the uninterrupted one.
What does not hold
Unknown voices. The full speech corpus is 54 recordings, 80,941,578 spike events per pass. An external diagnostic recognises all 28 known words, including 8 from other speakers. It also recognises all 28 with learning switched off. So the 28 of 28 is not evidence that the hippocampal region learned anything; it is evidence that the diagnostic is lenient. Worse, of 8 words the model never heard, 5 score at or above the weakest known word with learning on, and 8 of 8 with learning off. The separation margin is negative in both modes (comparison). A model that cannot say “I do not know this word” has not learned words. The verification file says: full service accepted, false.
Candidates rejected on the way. A calcium-and-resource CA1 variant: 7 of 9. A shared AMPA/NMDA presynaptic resource: 0 of 9. A somatic STDP event: 1 of 9. Extra plasticity between cortical layers 4 and 2/3: 27 of 28, identical to no learning. Rank-ordered connectivity: no improvement in the margin. A septal substrate with a resonance at 11.2 Hz against a human median of 2.9 Hz. Each one has its own folder with the run, the checks and the hashes of the source files that produced it, and none of them touched the accepted 9 of 9, whose 37 reference hashes are checked after every experiment.
Why I publish the failures
Because the alternative is the usual one: tune until green, publish the green. The acceptance rule for step 2 was written before the runs: 9 of 9 in both directions, no false recalls, retention over time, transfer to new voices. The first three hold on the fixed course. The fourth does not, and the stage stays open until it does. Every experiment ends with the same three lines: what passed, what was rejected, and that the original hashes are unchanged.
What this costs
All of it runs on one MacBook Pro M2 Max with 64 GB. One pass over the 54 recordings is 80.9 million spike events. The cochlear nucleus cell alone needs two minutes to initialise before it computes anything. The columns run at scale 0.1 because a full Potjans and Diesmann column is 77,169 neurons and about 300 million synapses, which is 6 GB at 20 bytes per synapse, and the architecture document already says the honest thing: on one machine this is three to four orders of magnitude below a human cortex, not one.
The next steps are known and written down: a physiological profile of the cochlear nucleus from independent human data, the missing relays between the cochlear nucleus and the cortex, and the acceptance on new voices. They are compute-bound. When step 2 is accepted, it will be a second report in the same format as the first, with its seed sweeps and its red cells.
Verification files quoted above are copies of the project’s own artifacts: index. The same report lives on the project site, getbraincore.com/progress-2026-10.html, next to Paper 1 (DOI 10.5281/zenodo.23119475). The code will be published with the step 2 report.