Metadata-Version: 2.5
Name: trainration
Version: 0.0.1
Summary: Decide what to train on, and never move the yardstick — data rationing and immutable evaluation yardsticks for robot learning.
Project-URL: Homepage, https://github.com/trainration
Project-URL: Source, https://github.com/trainration/trainration
Author: TrainRation
License-Expression: MIT
License-File: LICENSE
Keywords: data-selection,dataset-curation,embodied-ai,evaluation,lerobot,robot-learning
Classifier: Development Status :: 1 - Planning
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.10
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Requires-Python: >=3.10
Description-Content-Type: text/markdown

# TrainRation

> *Decide what to train on, and never move the yardstick.*

Data rationing and immutable evaluation yardsticks for robot learning.

## Status: early, pre-alpha

This release reserves the package name for an in-development project. **There is no
working functionality yet.** The version number (`0.0.1`) and the
`Development Status :: 1 - Planning` classifier are accurate, not decoration —
please do not depend on this package in anything real.

If you are looking for something usable today, this is not it. Watch
[github.com/trainration](https://github.com/trainration) instead.

## What it will do

Robot learning data is expensive to collect and much of it is redundant. TrainRation
works on two problems:

1. **Deciding what to train on** — dataset selection, mixture ratios, curriculum
   ordering, and failure attribution over [LeRobot](https://github.com/huggingface/lerobot)
   v3.0 datasets.
2. **Keeping the yardstick fixed** — the evaluation criteria that decide whether a
   change helped must not be reachable by the process that proposes changes. A
   self-improving system that can edit its own success criteria will optimize the
   criteria instead of the policy.

The second point is the design constraint that shapes the first. Research direction
evolves; the measuring stick does not.

## Planned components

| Package | Scope |
|---|---|
| `trainration-probe` | Cross-dataset quality and consistency checks for LeRobot v3.0 |
| `trainration-yardstick` | Evaluation criteria and admission constraints as structural, non-bypassable boundaries |
| `trainration-core` | Recursive decision kernel: mixture ratios, failure attribution, curriculum |

## License

MIT
