---
language:
- en
tags:
- research
- numpy
- reproducibility
---
# huret-digit-classifiers

Fifteen October digit classifiers: identity, parity and magnitude, five seeds each. The trained arrays are retained from the CP02 replay. Architecture: 64 -> 32 ReLU -> 16 ReLU -> task head. This is the October HuReT audit, not an MRI model and not the missing July model archive.

## Local use
```sh
python -m pip install -r requirements.txt
python validate.py
```
Import `predict` from `inference.py`; see `config.json` for input semantics. Inputs are not automatically normalized except as specified. Loaders reject wrong shapes, nonfinite values and invalid model selectors. Weight files are NumPy archives or JSON; no pickle/custom remote execution is required.

## Evidence and limitations
The validation is loader parity and input-contract testing, not independent external predictive validation. No clinical deployment, universal physics, foundation-model quality or new biological evidence is claimed. These are individual finite model families, not repackaged third-party LLMs.

## Licensing and provenance
Private staging. Author approval and redistribution rights must be confirmed before public upload. No new blanket license, completed HF upload or DOI is asserted. `PROVENANCE.json` records exact source file hashes and transformations.
## Verified packaging revision — 2026-10-20

Preserved model weights reloaded; saved or synthetic example checks passed. No training, held-out evaluation, clinical or hosted verification.

Runtime state: `PASS`. Rights and release approval are separate. See `RELEASE_VERIFICATION.json`. No paper or conference submission is included as a new software result.