translatestocks-tnt 1.3.0

TNT-ls — Public left-side release of the Thermodynamic Neural Transformer: routing, grokking, and Brouwer isolation layers.

Thermodynamic Neural Transformer (TNT)

translatestocks-tnt

Public release: TNT-ls
Package version target: 1.3.0
GitHub release tag: v1.3-ls

---

translatestocks-tnt is the public left-side release of the Thermodynamic Neural Transformer program. This repository is intended to expose the benchmark-validated public surface already supported by the current evidence trail. It is not intended to claim that the full post-grok TNT stack is public today.

The TNT Architecture Split: Why TNT-ls?

The full TNT architecture is conceptually split into two halves:

• Left Side (TNT-ls) — routing / grokking. This release is responsible for mapping latent discrete symmetries in the data, mathematically routing inputs through combinatorial structure, and enforcing geometric branch isolation (Brouwer isolation) before grokking.
• Right Side (TNT-rs, not included in this release) — thermodynamic crystallization. This is the future complementary side that takes already routed manifolds and applies thermodynamic principles, including H-neurons and quaternionic crystallization, to support efficient and secure continuous post-grok learning.

The name Thermodynamic Neural Transformer should be read as the umbrella name of the full TNT program, not as a claim that the current public package already contains the entire thermodynamic / right-side stack.

In short: TNT-ls prepares the topology; TNT-rs freezes the state.

Release boundary

This repository is the TNT-ls release.

It includes:

• constrained-backbone primitives such as BorweinAttention, RSTFFNBlock, and KTOrderParameter;
• public algebra-aware layers already part of the left-side library surface;
• low-cardinality / routing support used in public evidence, including BitPlaneEmbedding, DualEmbedding, CLogLogGate, and GatedExitNode;
• release-facing documentation;
• the LIB showcase summary as a public example of the released routing boundary.

It does not claim to include:

• the full post-grok / right-side TNT architecture (TNT-rs);
• quaternionic crystallization, H-neurons, continuous learning, or other later/private right-side layers;
• the full frozen ablation corpus for the companion paper as the primary GitHub front door.

The complete paper ablation bundle should be treated as a paper artifact and archived separately in the paper Zenodo bundle.

Why this release matters

The practical thesis of TNT-ls is narrow: structured architectural choices can improve the timing and reliability of useful transition in controlled benchmark regimes.

The strongest public example attached to this release is the LIB showcase:

• matched scale at approximately 93.4M parameters;
• LIB condition: 100% grokking on mult_1009, mult_503, xor, and add;
• matched baseline condition: 0% grokking on all four tasks;
• Brouwer isolation: 3/3 passes, with exact bp_norm = 0.0.

See LIBMODEL_SHOWCASE.md for the clean release-facing summary.

Release documents

• TNT_RELEASE_WHITE_PAPER.md — release-facing white paper for the public TNT-ls boundary
• LIBMODEL_SHOWCASE.md — public LIB showcase summary
• LIB_USAGE.md — usage guide for the TNT-ls public surface
• CONTRIBUTING.md — maintainer-curated upstream policy for the public repository
• NOTICE — patent / dual-license notice
• COMMERCIAL.md — commercial licensing note
• CLA.md — status note for the inactive public CLA path

Contributions

The canonical upstream is maintainer-curated. External code pull requests are not accepted for merge into this repository, and non-collaborator pull requests are closed by policy.

Forks and downstream use remain available under the repository license. If you need to report a bug or propose a change, use the issue tracker or maintain a downstream fork.

Package identifiers

• PyPI package name: translatestocks-tnt
• Python import path: translatestocks_tnt
• Public release name: TNT-ls
• GitHub tag: v1.3-ls
• Package version: 1.3.0

Install target

When the package is published:

pip install translatestocks-tnt==1.3.0

Theory references

Pinto Martins, C. F. (2026). Ramsey Statistics And Infiniteons Theory, Volume I. Zenodo. DOI: https://doi.org/10.5281/zenodo.19329589

Pinto Martins, C. F. (2026). Ramsey Statistics And Infiniteons Theory, Volume II. Zenodo. DOI: https://doi.org/10.5281/zenodo.19330373

Licensing

This repository is distributed under AGPLv3 with a separate commercial licensing path.

• See LICENSE for the full AGPLv3 text.
• See NOTICE for patent and dual-license notice.
• See COMMERCIAL.md for commercial licensing inquiries.

Commercial contact: [e-mail]