metadata/reconstruction.md

# Reconstruction

## What this release contains

- Public QA tables for `eval` and `sft`.
- Split manifests and split counts.
- Public release-relative references for each item.

## What this release does not contain

- Full raw videos.
- Full extracted frame caches.
- Local absolute paths used during internal development.
- Separate structured task-meta fields.

## How to use the public references

Each row exposes three reconstruction-oriented fields:

- `visual_ref`
- `source_episode_ref`
- `reconstruction_key_json`

These are designed to be stable, path-free references. They are sufficient to map a released item back to the corresponding source episode or recording once you have obtained the upstream dataset under its original terms.

## Source-specific units

- `GM-100`: grouped by `(task_id, episode_id)`.
- `RH20T`: grouped by `recording_id`.
- `REASSEMBLE`: grouped by `recording_id`.
- `AIST-Bimanual`: grouped by `recording_id`.

## Expected workflow

1. Obtain upstream datasets under their original terms.
2. Use `source`, `source_task_id`, `source_unit_id`, `camera`, and `frame_indices_json` from the public release.
3. Reconstruct the required visual inputs with the source-specific evaluation scripts in the main codebase.

## Release policy

This Hugging Face release is a reviewable minimal closed loop for metadata, QA, splits, statistics, and baseline results. Full visual reconstruction remains upstream-data dependent.