tcn-logic-joint

tcn-logic-joint

Exact frozen typed programs from Typed Crystallization Networks. These are not neural networks. Each artifact is a discrete typed program that executes on stdlib Python with no torch, no accelerator, and no framework — program.pyz runs under python3 -I.

Shapes

library/ is the persistent module library: content-addressed modules with recorded conformance fixtures and transitive dependencies. A module loads only if its fixture re-executes exactly.

Reading a card

Every shape carries card.json with its certificate, search size, node count, held-out score and its baseline. A number without its baseline is not a result. certificate: unique means enumeration exhausted the space and exactly one program fits — a guarantee a learned network cannot offer.

Naming follows NOMENCLATURE.md in the source repository: repository is tcn-<domain>-<capability>, one directory per typed shape, everything variable in metadata. Shape matters because a module hardened at one observation width cannot be registered at another — width is part of type identity.

Source: https://github.com/JacobFV/typed-crystallization-networks