What you get
- A spreadsheet and a data file with one row per request and per thing to remove: whole account or one repository, owner, repository, date filed, current state, link to the public issue and a fingerprint of the request text.
- A filter script that reads your local copy (Parquet if you have pyarrow, or JSONL), drops the matching repositories, writes a filtered copy and a removal report with counts per request. It never changes your original files.
- A list of requests we could not read, so nothing is hidden. In this edition that is 129 of 8,259.
This edition, in numbers
Counted from the 2026-10-01 edition of our capture of the public repository.
| Item | Count |
|---|---|
| Issues read (open and closed) | 8,259 |
| Rows (request × thing to remove) | 62,584 |
| Distinct owners named | 8,210 |
| Distinct repositories named | 56,487 |
| Whole-account requests (distinct owners) | 5,266 |
| Requests we could not read | 129 |
This edition contains 5,864 readable opt-out requests filed from January to September 2026, including 650 in August and 558 in September. These counts exclude requests we could not read. The current Stack release was downloaded 177,226 times last month. Downloads are not unique buyers. Sources: filing dates counted from this edition’s optouts.csv; downloads from the Hugging Face API for HuggingFaceCode/stack-v3-train, read on 2026-10-01.
Why not just read the issues yourself
The requests are free-text issues in several formats: whole accounts, repository lists, links, odd spacing. Turning thousands of them into a clean list takes hours each month, and a later edit can overwrite the earlier text. We keep a dated capture, so edits are recorded from the day we started capturing.
What this is not
It does not say any request is valid, that a licence is revoked, or that a model forgot anything. It is not legal advice. It covers requests filed in that one repository only, and it reads each issue’s comments only to catch the person who filed it cancelling; none was found as of 2026-10-01.
Look before you buy
The excerpt has the same columns and the same filter script as the full edition; the sample README adds a sample banner. After checkout the full file set is delivered in your browser; no account needed.
Questions or a refund: ustechautomations@gmail.com