---
language:
- en
task_categories:
- other
tags:
- product-launch
- metadata
- video
license: other
size_categories:
- 1K<n<10K
---

# Mayflower

Mayflower is a metadata-only index of product-launch films catalogued by [What Ships](https://whatships.com/). It preserves the directory entry and original X post for each launch; it does **not** contain videos, audio, frames, posters, transcripts, or model-generated media annotations.

## Why metadata-only

What Ships states that its catalog is not a download CDN and that the videos remain the property of their respective owners. Its public catalog may be indexed, but that does not license copying or redistributing media. This project therefore only collects public descriptive metadata with attribution. See the [What Ships terms](https://whatships.com/terms/) and [developer guide](https://whatships.com/llms.txt).

## Build the manifest

Requires Python 3.13 and [uv](https://docs.astral.sh/uv/):

```bash
uv run mayflower --output data/mayflower.jsonl
```

The command reads `search-index.json`, requests each video page as Markdown, and writes JSON Lines. It intentionally has no media downloader.

For a small smoke run:

```bash
uv run mayflower --limit 3 --output data/sample.jsonl
```

## Schema

Each row has a stable source ID, What Ships and original-X URLs, title/product/company/category, publication date, duration, author handle, tags, description, poster URL, source snapshot date, and these controls:

- `license_status`: always `metadata-attribution-only` in this collector.
- `media_included`: always `false`.
- `analysis_status`: `not_run` unless a separately permissioned workflow is added.

## Planned enrichment boundary

ASR, music classification, OCR, frame extraction, and video-model analysis (including Marlin) are out of scope for these public-source records. They may be run only for media supplied with explicit rights-holder permission covering the specific processing and redistribution. Keep a consent ledger and remove derivatives when consent is revoked.

## Hugging Face publication

The intended dataset is `guychuk/Mayflower`. Upload `README.md` and `data/mayflower.jsonl` after review:

```bash
hf upload guychuk/Mayflower . --repo-type dataset
```

The dataset card must retain this provenance and restriction language.