Datasets:

PureOne
/

AELITH-Sigma-Geometric-Cascades

A proposed proof of the off-critical forced-droplet propagation theorem for Poisson random geometric graphs in every fixed dimension d >= 2, with the original 15-page manuscript, analytic proof source, separated-defect geometry, propagation-time bounds and reproducible supporting checks.

This kit prepares the release files. It does not create a repository or upload to your account.

Guided Windows uploader

Extract the entire ZIP to a folder, keeping its subfolders together.
Double-click UPLOAD_HUGGING_FACE.cmd in that extracted folder.
The launcher finds an installed Python 3.10 or newer, creates an isolated environment under uploader/.venv, and installs the official Hugging Face library if needed. The uploader does not require Python 3.12 specifically or the numerical verification dependencies.
If you already have Hugging Face authentication on that computer, the uploader uses it. Otherwise, paste a token with write access at the hidden local prompt. Create one at Hugging Face token settings. Do not paste the token into this chat.
Check the authenticated account name, enter a repository name or accept the suggested one, and select public or private visibility for a new dataset.
Review the destination, visibility and file count. Type UPLOAD to create or update the dataset, or press Enter to cancel.

The authenticated username is read directly from Hugging Face, preserving its exact spelling and case. This guided uploader targets that personal account. An existing repository retains its visibility. The release files are added or updated; other remote files are retained.

The uploader verifies every manifest entry before authentication and uploads an in-memory snapshot of exactly those files plus the manifest. Extra local files, virtual environments and generated run folders are excluded. The SDK handles normal and large-file upload modes. After committing, the uploader downloads each release file at that exact commit and compares its SHA256 hash before reporting success. This also detects omissions or stale files caused by repository filtering.

The token is not placed in command-line arguments or saved by this uploader. Existing cached authentication is left in place. Downloaded verification copies may be stored in the normal Hugging Face file cache.

If no suitable Python is found, the launcher gives the official Windows download link. Install a current Python 3 with its launcher enabled, then double-click the uploader again. No administrator access is requested by these scripts. Their bootstrap uses Windows PowerShell 5.1 and a process-local execution-policy option; it does not change the computer's persistent policy.

For macOS/Linux, or to run directly with an existing Python environment:

python -m pip install "huggingface_hub>=1.0,<3.0"
python uploader/upload_hugging_face.py

To ignore a cached or environment token and enter a different one, add --fresh-token. To check the package without authentication or network calls:

python uploader/upload_hugging_face.py --check-only

If a network error occurs after a commit, the uploader prints the commit link when available. Inspect that revision before retrying. It does not automatically repeat a failed or uncertain commit. A concurrent change to the destination branch is rejected through the SDK's parent-commit guard.

Uploader validation

Ten offline tests cover manifest exclusion, tampering, path traversal, exact account naming, new private repositories, cancellation, existing visibility, invalid owner input, token redaction, concurrent changes and remote-content mismatch. Run:

python -m unittest discover -s uploader -p "test_uploader.py" -v

These tests use a simulated SDK and no credentials. The SDK interfaces and Windows launcher were reviewed; the Windows launcher and live authenticated upload were not executed in the release environment.

Browser upload

Extract the ZIP to a folder on your computer.
In Hugging Face, create a New Dataset under your intended account or organization. Use the suggested slug or another valid repository name.
Open Files and versions and choose Add file / Upload files.
Upload the extracted contents, including the subfolders. README.md must be at repository root, alongside CITATION.cff, claims.json, RIGHTS.md and SHA256SUMS.txt.
Commit the upload. Open the dataset card and manuscript link to confirm the root layout.

Upload the extracted files, rather than only the outer ZIP. Uploading only a ZIP does not expose the enclosed README as the repository card. If the browser does not preserve subfolders, use the CLI folder upload below.

CLI upload

The current official CLI is hf, supplied by huggingface_hub. Install it and authenticate on your own computer:

python -m pip install -U huggingface_hub
hf auth login
hf auth whoami

Use the exact namespace shown for your authenticated account, including its spelling and case. YOUR_NAMESPACE below is a placeholder to replace, not an account name inferred by this kit.

Open a terminal in the extracted release folder, where README.md is visible, and run:

hf upload YOUR_NAMESPACE/AELITH-Sigma-Geometric-Cascades . . --repo-type dataset --commit-message "Release AELITH Sigma 1.0.0 research archive"

The documented CLI creates a missing repository automatically. If you want a private initial repository, create it as private through the web interface first. The example adds or updates files; it does not delete old remote files. For this first release, a new empty repository gives the clearest layout.

Run the upload from a clean extracted release, before creating a virtual environment or local reproduction outputs inside it. Do not place access tokens or other private files in the folder you upload.

Dataset card behavior

The supplied front matter uses viewer: false. Hugging Face documents this setting for disabling the Dataset Viewer. This suits a document/code archive and avoids presenting the manuscript as a table to browse.

The card intentionally does not declare dataset configs, data_files, ML task categories or training splits. The research does not provide an ML training dataset. Readers should use the Files tab or snapshot_download, rather than assuming datasets.load_dataset can load the archive.

No standalone license was present in the supplied source package. The card omits the license field and preserves that state in RIGHTS.md. If the rights holder later chooses a reuse license, update the license text and matching card metadata together, then regenerate the release checksums for that revision.

Check the published layout

After upload, inspect these relative paths:

Item	Expected path
Research card	`README.md` at repository root
Complete manuscript	`manuscript/AELITH_Cascade_Theorem.pdf`
Proof source	`source/AELITH_Cascade_Theorem.tex`
Illustration	`figures/cascade_geometry.png`
Verification source	`verification/verify_checks.py`
Original results	`results/original/geometry_checks.json`
Fresh rerun record	`results/rerun/rerun_record.json`

No DOI, public URL, download count or external endorsement has been invented in the card or citation files. Once a repository is published, its actual URL may be added to a later release's citation metadata.

Official documentation

Documentation checked during release preparation on 27 October 2026. Interface labels can change; the root-file layout and repository type are the essential choices.