Paid content copy detection

Copy detection for paid content

Goal: a buyer can always copy files they paid for, so make the obvious abuse (re-uploading paid files as free content on your own platform) easy to catch and costly. Hold matches for a person; never auto-delete.

How it works

1. Fingerprint paid items. When an item becomes paid, and whenever a new version of a paid item publishes, store one fingerprint per file: sha256 of the file's normalized content, with the item id and path.

2. Normalize before hashing, so trivial edits do not escape:

   • text files: drop a leading YAML front matter block (it usually names the item, the first thing a copier renames), collapse all whitespace to single spaces, and trim;
   • binary files: hash as they are;
   • skip manifest files (they always differ) and anything under a minimum size, for example 200 bytes after normalizing, since tiny boilerplate files match by chance.
3. Check public uploads. Before a public version is reviewed or published, fingerprint its files the same way and look up matches against paid items owned by someone else. Never match an owner against their own items.
4. Hold, do not reject. A match puts the version in a held state with the matching item names, alerts staff, and publishes nothing. Staff compare and either publish (fair use, the owner, a licensed copy) or reject.
5. Private copies are fine. Buyers may keep private clones of what they bought. Only public uploads are checked.

Steps

1. Find where uploads are validated and published, and where "held for review" already exists (reuse it if so).
2. Add a fingerprint table (item id, sha256, path) with an index on sha256, and the normalize-and-hash function.
3. Index on pricing and on each new paid version; clear the index when an item becomes free.
4. Call the check before publishing public versions; record the reason on the held version and send one alert.
5. Add a report reason such as "a copy of paid content", so buyers and creators can flag copies found later.
6. Write tests: an exact copy is held, a copy with a renamed header and changed spacing is held, the owner's own next version is not, a private copy is not, and small files never match.

Rules

• Never block a person's private copies or downloads of what they bought.
• Do not store file contents for this, only hashes.
• Off the platform (other sites), the creator's copyright and a takedown notice are the tools; say so in your terms.

paid-content-copy-detection

A house kit harvested from building SlashClone. See SKILL.md for what it does and how Claude uses it.