---
license: other
pretty_name: Scientific Post-Training Raman Evaluation Resolver v1
tags:
  - raman-spectroscopy
  - bioprocess
  - post-training
  - evaluation
---

# Scientific Post-Training Raman Evaluation Resolver v1

This repository is the canonical provenance manifest for the
`raman-bioprocess` Harbor task. It intentionally contains **no Raman spectra or
labels** because the upstream RamanBench mirror requires users to respect each
original dataset's terms and prohibits unapproved redistribution.

`manifest.json` pins the public Hugging Face source, exact commit, four Parquet
file hashes, deterministic split, and target definitions. During the Harbor
image build, the task downloads those files directly from
`HTW-KI-Werkstatt/RamanBench`, verifies every hash, and materializes the local
read-only evaluation bundle used by `/app/evaluate.py`.

The selected test rows are evaluation-only. No training examples are supplied
by this repository or the Harbor task.