Detection
Community reports and the verdict cache
A shared, anonymous database of verdicts keyed by a hash of the content. It holds classifier results and report tallies, never model rechecks and never who produced them.
Updated
What the cache is
Every piece of content the extension checks is identified by a one-way hash. Against that hash we keep the kind of content, the score, the label, each detector's signal and notes, the source URL of an image (so the same image on two sites resolves to one verdict), and a hit count. Nothing in an entry identifies a user: no account id, no session, no IP address, and never the text itself.
What the Community row shows
When your extension checks a post, our API looks the hash up. If an earlier Pro user triggered the classifier on it, that row appears in your card. Model rechecks are not shared: one you did not run yourself never appears. If people have reported the post, the row shows the tally as an unverified count, and "No reports yet" means nobody has reported it.
Reports do not change scores
They used to. They no longer do, anywhere, for anyone.
A report is recorded and shown as a count of unverified user opinions beside the estimate, in its own row: for example "4 unverified user reports: 3 say AI, 1 says human." That row carries no score. It does not change the estimate you see, it does not change the estimate anyone else sees, and there is no threshold at which it starts to count, because there is nothing for a threshold to gate. The scoring function is no longer given the report tally at all.
What decides the estimate is, in order: a provenance detector that affirmatively found AI generation, then a model recheck you ran, then the writing-pattern analysis, then a neutral baseline when nothing had an opinion.
This changed because the old behaviour was unsafe. Reports used to be blended into the score, and for images — which have no writing to analyse — they could become the score outright. Three accounts were enough to push a photograph to an AI verdict, including a photograph whose own Content Credentials said a camera took it. Making reports advisory removes that entirely rather than making it harder.
Why it is shared
Paid checks are budgeted. Caching the classifier means one Pro check of a viral image helps everyone who scrolls past it, and nobody pays twice for the same image. Model rechecks are private to the account that ran them and are never cached for others.
Removal
Entries are anonymous, so deleting your account does not remove them. If an image source URL in an entry identifies you, email privacy@slopsquash.com and we will purge that entry. Copyright complaints about a URL go to dmca@slopsquash.com. See DMCA and removal requests.