| --- |
| license: agpl-3.0 |
| pretty_name: Cubed read-trust v1 |
| tags: |
| - rubiks-cube |
| - computer-vision |
| - tabular-classification |
| --- |
| |
| # Cubed read-trust v1 |
|
|
| A dependency-free, NumPy-only classifier used by |
| [Cubed Core](https://github.com/KingBobJoeIV/cubed-core) to assign a trust |
| weight to each per-cell color read. It is not a tracker, cube-state estimator, |
| or camera-to-moves decoder. |
|
|
| ## Files |
|
|
| | File | Purpose | SHA-256 | |
| | --- | --- | --- | |
| | `trust_v1_numpy.npz` | Distilled read-trust model | `3f40e5d5f5f385f8b96547de11b0d92f17d5aa5575302646338e273fb0c5bfd7` | |
| | `calibration_gan12.json` | Companion six-color gtD1 calibration | `c259404e7257fa83f4dabf33286dd49cb20696e3216804aa0aaf9010529c48e1` | |
|
|
| Use Cubed Core's |
| [Decode guide](https://github.com/KingBobJoeIV/cubed-core/blob/main/docs/tutorials/DECODE.md) |
| for the supported setup. Pin an exact repository revision for reproducible use. |
|
|
| ## Limits |
|
|
| The model and calibration are tied to the maintainer's cube, camera, and |
| lighting. They are unmeasured on unrelated setups and make no standalone |
| accuracy, decode, or generalization claim. The unavailable source model and |
| training corpus prevent a public byte-for-byte rebuild. |
|
|
| The owner-authored model and calibration are `AGPL-3.0-only`. |
|
|
