zedgamer/asset-quality-memory-knn

Asset Quality Memory k-NN

An explainable, zero-cloud k-nearest-neighbor baseline for Asset Factory QA, implemented in Botte Secrète.

What it does

1. Runs deterministic integrity, licence, manifest, SHA-256, and size checks.
2. Applies family-specific checks for images, textures, meshes, animations, or Godot packages.
3. Searches only externally verified neighbors from the same asset family.
4. Returns FAIL, UNCERTAIN, PASS, or PASS_ROBUST, with neighbor IDs.
5. Abstains when fewer than three comparable examples exist.

It is shadow-only and cannot import, activate, or publish an asset. CPU is enough; a GPU can generate upstream features but cannot bypass hard checks.

Quick start

git clone https://github.com/zedarvates/botte-secrete.git
cd botte-secrete
python -m skills.asset_quality.cli evaluate \
  examples/asset-quality/mesh-report.json --json
python -m skills.asset_quality.test_asset_quality

Verified memories remain project-local in .botte/asset-quality.jsonl. Do not upload that file: it may reveal operational fingerprints or evaluation history. This Hub repository distributes the implementation contract and documentation, not a user's private neighbor index.

Intended use and limits

This baseline helps decide whether a specialized micro-NN is justified. Promote a learned model only if it beats k-NN on a representative temporal holdout without weakening deterministic checks or verified quality.

Source and issue tracker: zedarvates/botte-secrete.

Licence

MIT. See the source repository's LICENSE.