Layer 1 is starting to work like an object-recognition layer over text.
The system scans text through small sensor patches, then cortical-column-like modules build sparse maps of what they observe:
feature@location -> movement -> feature@location -> movement -> …
Recent progress:
– Pose stays monotonic instead of resetting at boundaries.
– Scanning now has saccade/fixation timing, so columns can settle.
– Repeated paths can be recognized later.
– Ambiguous evidence can keep multiple candidates active.
– Later evidence can narrow the candidate set.
– Neighboring columns provide lateral context without forcing shared codes.
– Frozen familiar input produces lower surprise and higher path fit than unrelated input.
Punctuation used to fragment word identity. “word,” “word.” and “word?” became different Layer 1 objects. Now the alphabetic path closes before attached punctuation, but punctuation still remains in the stream as observed evidence.
So Layer 1 can recognize the shared “word” path without pretending punctuation does not exist.
Next step: Layer 2 observing the Layer 1 SDR surface and learning larger repeated structures without hardcoding phrases, sentences, or topics.