Four ways to stop
| How | What it stops | When | Content already collected |
|---|---|---|---|
| Email the operator | Everything, for the URL or the whole origin | On receipt, exclusion recorded first | Text purged, excluded from every later export |
| A robots.txt Disallow for CrawlCharter | Fetching | On the next policy read | Quarantined on the next re-check if it covers the collected paths |
| A well-known AI policy denying training | Training use only | On the next policy read | Quarantined on the next re-check |
| A TDM reservation | Training use only | On the next policy read | Quarantined on the next re-check |
What happens to content already collected
A quarantine keeps the text and stops using it. A takedown destroys the text. Ask for removal and you get a takedown.
| Aspect | Takedown | Quarantine |
|---|---|---|
| Trigger | a verified removal request, or a decision by DEVAesia | the policy the origin published became more restrictive |
| Stored text | purged | retained |
| Stored decision | tombstoned | left intact as the record of collection time |
| Reversal | never | manual, and re-applied if the evidence still refuses |
Not built
Limits
A removal reaches every later export and stops there.
Nothing maps a URL back to a file already handed on, and nothing propagates a removal into it. That work is unbuilt.
Release from quarantine is one decision.
There is no per-document verdict, no queue, and no two-person approval. The next re-check re-applies the quarantine if the evidence still refuses.
Request verification is undecided.
How to confirm a request comes from the site owner has not been settled.
Explicit permission
If you would rather it collected something
The permissive direction has to be as easy as the restrictive one. An explicit permission with a recognised licence at the well-known AI policy path is what reaches training allowed automatically. This site publishes its own: /.well-known/ai.json.