PTCNA — Prime Tensor Circled Neural Architecture

One architecture, four layers. Not four repos — four layers of one thing. PTCNA consolidates the former The-Interdependency repos pcna, pcta, and pcsa into a single package and is the single upstream that feeds interdependent-lib.

The four layers

Each layer's tensors divide into the next; every circle, seed, and core is itself a tensor.

• Back-propagation lives only in the neural layer. Circle, seed, and core tensors are auditing and timing tensors — they do not differentiate.
• fiqs gate when cores propagate internally, per Fick's first law J = −D ∇φ (structure diffusing down its field gradient). Timing, not gradient descent. The fiq substrate lives in ptcna.core.prime_core.
• PCEA (Prime Circular Encryption Algorithm) is not a layer — it stays a separate, orthogonal repo (the guardian: "last state as key" at every layer).

Provenance

Install & test

pip install -e ".[dev]"      # neural layer needs numpy; seed/core are stdlib-only
pytest                       # testpaths = ptcna

Status

Alpha (0.1.1). All four layers import and the repository test suite passes. The layer boundary is now executable rather than only descriptive:

• ptcna.circle.CircleTensor is the one circle type used by circle, seed, and core composition.
• ptcna.neural.NeuralScalar is the only PTCNA type that owns reverse-mode gradients; structural hosts carry it opaquely.
• UCNS integration reports a typed suspended state and fails closed until a reviewed PTCNA-specific higher-gonol producer profile exists.
• EDCM remains an external authority. ZetaEngine accepts an explicitly injected measurement provider; PTCNA contains no shadow EDCM module.

The core layer still intentionally exposes PTCA-named public objects such as PTCATensor and PTCAInstance; those names live in the correct layer. History and unresolved evidence boundaries live in docs/architecture.md.

License: MPL-2.0.