docs/UPLOAD_GUIDE.md

# Upload this release to Hugging Face

Suggested repository: **PureOne/Eidrion**, type **dataset**. The preparation task does not itself publish anything. This guide describes deliberate upload actions.

## Windows: extract and double-click

1. Extract the entire outer release ZIP into one folder. Paths with spaces are supported.
2. Double-click `CHECK_RELEASE.bat` for a local preflight.
3. Double-click `UPLOAD_TO_HUGGINGFACE.bat` to upload to `PureOne/Eidrion`.
4. If requested, enter a Hugging Face write token in the hidden prompt. Create one at [Hugging Face token settings](https://huggingface.co/settings/tokens) with write access to the destination.

The launcher finds Python 3.10 or newer, creates an isolated `.venv-upload`, and installs the upload dependency. No Git, PowerShell execution-policy change, scientific dependencies, or manual change of working directory is needed. The window remains open on success and failure.

The destination is printed before uploading. New repositories are **public** by default. Existing repository visibility is preserved. The uploader does not delete remote files; matching paths receive new versions. It uploads only the sealed file list, so local virtual environments, generated checks, and credentials are excluded.

To choose another destination or make a newly created repository private, run from the extracted folder:

```bat
UPLOAD_TO_HUGGINGFACE.bat --repo-id YOUR_ACCOUNT/Eidrion --private
```

Use an argument override rather than editing the sealed `release_config.json`. Changing a hashed package file causes preflight to fail. `--private` applies when creating a repository; it does not change an existing public repository.

## Python workflow on Windows, Linux, or macOS

Local planning requires only standard-library Python:

```sh
python tools/verify_release.py
python upload_to_huggingface.py --dry-run
```

Install upload dependencies in an environment of your choice, then deliberately execute:

```sh
python -m pip install -r requirements-upload.txt
python upload_to_huggingface.py --repo-id PureOne/Eidrion --execute
```

The uploader reuses an existing Hugging Face login or `HF_TOKEN`. Otherwise it asks for a token using masked input. It does not write a token into the release or print it. To use the current Hugging Face CLI login separately, run `hf auth login`.

Files are verified before network operations and uploaded in batches of at most 64. Transient network errors, rate limits, and server errors receive bounded retries; authorization and malformed-request errors stop immediately. Rerunning after interruption uploads the same allowlisted paths and does not remove files already committed. A partial upload can leave an incomplete repository until remaining batches succeed.

## Browser fallback

Create a dataset repository on Hugging Face using the intended name and visibility. In its **Files and versions** tab, upload the release folder's contents, preserving subfolder paths and placing `README.md` at repository root. Upload `Eidrion_v1.0.0.pdf` and the original research ZIP too. Do not upload the outer delivery ZIP as the sole file: the dataset card and tables need their root paths to render. The Python uploader is the easiest way to preserve the layout.

## If an upload fails

| Message or condition | Action |
|---|---|
| Python not found | Install Python 3.10+ from [python.org](https://www.python.org/downloads/) and enable its PATH option. |
| Dependency installation failed | Check the connection and rerun. Scientific PyTorch installation is not needed for uploading. |
| Hash mismatch or missing release file | Extract a fresh complete copy. Avoid editing sealed files; use CLI target overrides. |
| HTTP 401/403 or permission error | Check the logged-in account and write permission for the destination. Use a suitable write token. |
| HTTP 429/5xx or timeout after retries | Wait until the service recovers, then rerun the launcher. Earlier committed batches remain. |
| Existing `.venv-upload` is broken | Remove that generated upload environment and rerun to recreate it. |
| Root page does not show the card | Verify `README.md` is at root and its relative linked files were uploaded. |

This publishes the supplied **main research package**. It cannot upload the four raw-evidence companions that were not supplied. Their recorded names and checksums are included for later restoration. No dataset licence, DOI, peer-review result, or priority claim is created by upload.

Implementation reference: [official Hub upload guide](https://huggingface.co/docs/huggingface_hub/guides/upload). Dataset metadata reference: [official dataset-card guide](https://huggingface.co/docs/hub/datasets-cards).