axiter

Excitable-neuron path retrieval: beam search over a graph where node score is similarity plus excitation plus novelty minus inhibition minus fatigue, with a refractory period on revisit.

Part of the Neuroca-Inc research tooling.

---

Install

pip install -e .

Layout

  └─ demo.py
    └─ events.py
    └─ metrics.py
    └─ neuron.py
      └─ default.neuron.json

Notes

• Scoring is alphasim + betaexcitation + gammanovelty - deltainhibition - epsilon*fatigue, with refractory_decay applied to per-path fatigue and a hub diversity penalty.
• events.py is carried in because neuron.py imports VTTouchEvent / SpikeEvent / BaseEvent from it. In the source tree that import sat behind a try/except that silently substituted stand-in classes.
• memchunk_cli shipped a byte-identical copy of the policy file (sha256 fbe28f8f356b...). Only one copy is carried in.
• metrics.py is carried in because events.py does 'from metrics import StreamingZEMA'. That is an implicit same-directory import that only resolved inside the original tree. It is the reason numpy is a dependency of an otherwise pure-stdlib package.

Provenance

Every carved tree ships with a PROVENANCE_manifest_*.json (vdm.provenance.manifest.v1): SHA-256 per file, byte sizes, and a deterministic tree hash over the whole set. To confirm nothing has drifted since the carve:

neuroca-carve verify --repo .

License

Apache-2.0. See LICENSE.