Research-Stack/docs/transfold_couch_data_magnetic_domain_comparison.md
2026-05-08 14:50:03 -05:00

2.6 KiB

COUCH Data Magnetic-Domain Comparison

This comparison runs the same byte-signal to magnetic-domain transfold used for enwik8 against the local COUCH JSON data bundle.

COUCH bundle:

source files: shared-data/data/couch_*.json
bundle path: shared-data/corpora/couch_data_bundle.jsonl
bundle bytes: 86878
bundle sha256: 4aab981a761adcb4043dd7e90cd18ea9cbc059d62676c6f72d32948a1b11ace3
receipt: 4-Infrastructure/shim/transfold_couch_data_magnetic_domain_receipt.json
receipt hash: cad08a03b9bce19475676a69e96d9a532511409ba6213fd7a9ba864d0215f64f

Aggregate Readout

Corpus Slice bytes Chunks Mean H/load Mean remanence Mean susceptibility Mean magnetization Mean heat loss
COUCH bundle 86289 22 3.663922436343822 0.8886250883537214 0.5687781936238817 0.7907252340455144 0.23921198455929052
enwik8 64 KiB 65536 16 4.48288790262248 0.8377097662234787 0.7478517983083702 0.6444969211710623 1.0122453515412608
enwik8 1 MiB 1048576 256 4.446752552238698 0.8444054709628067 0.7464948171449458 0.6518732011907801 0.9727067215896326

Stress Sweep

Corpus Threshold Overflow chunks Mean overflow gate Mean heat loss
COUCH bundle 2.50 22 0.22507297397763984 0.9473216166491661
COUCH bundle 3.25 20 0.5737709098179441 0.23921198455929052
COUCH bundle 4.00 2 0.9900600023017341 0.0010042089982083586
COUCH bundle 4.50 0 1.0 0.0
COUCH bundle 5.25 0 1.0 0.0
enwik8 1 MiB 2.50 256 0.06938602211338227 1.815654861325333
enwik8 1 MiB 3.25 256 0.19663556731358448 0.9727067215896326
enwik8 1 MiB 4.00 247 0.5429233805793959 0.22546958198757838
enwik8 1 MiB 4.50 106 0.9653930640465004 0.003749109274715911
enwik8 1 MiB 5.25 0 1.0 0.0

Read

The COUCH bundle returns the same magnetic-domain shape class but in a smaller, cooler regime:

COUCH mean heat loss is about 0.239, versus enwik8 1 MiB at about 0.973.
COUCH mean remanence is higher, about 0.889 versus 0.844.
COUCH mean susceptibility is lower, about 0.569 versus 0.746.

That suggests COUCH data is more memory-like and less transition-agitated under this transform. In routing terms, COUCH looks like a compact high-remanence surface that should prefer dictionary/state reuse, while enwik8 looks like a larger high-transition surface that needs more boundary-aware shaping.

Claim boundary:

This is a byte-statistical magnetic-domain analogue. It supports routing and stress-test comparisons only; it is not a proof of compression improvement.