88,384,522 fingerprints,
in just 2.74 GB

The first full-table challenge was its many segmentation fragments, not only neurons.

Build one partition at a time.

Keeping a 50-dimensional counter for every body is expensive. Synapse contributions are partitioned on disk by ID range, then accumulated and fingerprinted one partition at a time.

All inputs are preserved. Only storage and construction order change.

Store repeated results once.

Many fragments have the same quantized sketch or winner set. A dictionary stores unique sets; each body uses an 8-byte ID and a 4-byte reference.

Compression preserves verifiable results.

Every leaf still binds the data root, seed, body ID and all 100 winners. Tests compare compressed and in-memory implementations and obtain identical fingerprints and artifact roots.

The saved index is rechecked on restart. Previously generated proofs still verify after reload.

Share repeated representations; preserve individual identity.

This does not prove RAM residency.

Data is mapped with mmap and pages are managed by the OS. Compression improves practicality; it does not prove continuous ownership of a certain RAM capacity.