[Research] A fixed-footprint, self-organizing memory substrate — a different angle on brain-inspired long-term memory

[igs0707]

Hi EverMind team,

Your framing of EverMemOS around brain-inspired memory (hippocampal
indexing, cortical consolidation) resonates with something I've been
building independently — a small (N=12) architecture with a similar
inspiration but a different mechanism, sharing it in case it's a useful
reference point rather than a feature request:

• A fixed-size (N×N) substrate that individuates per user through lived
  experience alone — no per-user fine-tuning, no growing store. Two copies
  of the same base model, given different experience, become measurably
  better at their own user's context (+0.16 own-context advantage, 20/20
  seeds positive, replicated at 5x scale).
• The long-term substrate uses pure multiplicative accumulation, which
  self-organizes its own threshold and ceiling — no external tuning
  required, verified across a 1000x range of the core parameter.
• Recall from a partial cue exceeds the cue itself (pattern completion),
  and the memory footprint stays fixed regardless of how long the agent
  has been running (verified across a 256x range of lifetime length).

Preprint + code: https://doi.org/10.5281/zenodo.21122080

Sharing in case it's a useful data point for EverMemOS or related
benchmarks — not asking for anything.

— Kimiyasu Igarashi, independent researcher