What it is
A unified computational framework arguing that memory has two separable dimensions — structural accumulation (how much of a trace has consolidated) and representational fidelity (how faithfully it preserves its original encoding) — which are differentially modified by retrieval. The distinction resolves several phenomena that have resisted unified explanation across consolidation, extinction, and reconsolidation.
The theory paper — Structural Crystallization: A Unified Computational Framework for Memory Formation, Persistence, and Modification — was declined at Psychological Review (June 2026) after review by two experts. The reviews were detailed and constructive, and they’re now driving a substantial revision before the work goes to another outlet.
Where it’s going
This is the trunk the rest of the work grows from. The framing is being carried, deliberately, into running systems rather than left as theory:
- CrystalCache — porting the two-dimensional memory model into KV-cache eviction for long-context LLMs.
- Crystalmem — artificial memory systems built on the same structure (parked for now).
- A crystallization-dynamics engine that aims to make the theory runnable end to end.
The project stays in progress as long as the theory is still being pushed — through revision and resubmission, and through the systems that instantiate it.