Most edits to Wikipedia enter the page history the moment they are saved, whatever happens to them afterward. A human might revert one an hour later, a bot might flag it, a patroller might leave a note — but the edit existed, visible in Special:History, from the moment the save button was pressed. Some edits never get that far. A script evaluates them in the half-second between the click and the page reloading, and some of them are stopped before a single byte reaches the article.
The mechanism is called the edit filter — technically the AbuseFilter extension, built to catch obvious vandalism and spam faster than any volunteer patrolling recent changes could. A filter is a rule written against the shape of an edit: the words added, the links inserted, the size of the change, the age and edit count of the account making it. Hundreds of filters run on every save across English Wikipedia, most of them looking for page-blanking, inserted phone numbers, or a dozen other patterns common to abuse and almost nothing else.
A filter can take one of three actions once it matches. It can tag the edit — let it through, but flag it for a human to glance at later, invisibly to the editor who made it. It can warn — show a message, ask the editor to look again, and let them resubmit if they choose to. Or it can disallow the edit outright, which means the save simply fails. No revision is created. No diff exists for anyone to read, revert, or discuss, because the edit was never written to the page in the first place.
A disallow filter should be tight enough that a false positive is rare, because an editor who hits one has no normal recourse — there is no edit to appeal, only a log entry naming the filter that stopped it.
That last point matters more than it first sounds. An edit that gets reverted still happened — it sits in the history, a diff someone can point to, defend, or fix. An edit a filter disallows never happened at all, in Wikipedia's own record. The only trace is an entry in Special:AbuseLog: the filter number, a timestamp, the account, and — for the filters that aren't kept private — a one-line description of what it looks for. Filters that would be easy to dodge once their logic is known are deliberately hidden from everyone but the small group of editors who maintain them, so the log entry alone can be the whole explanation an editor gets.
For an agency account, the distinction is worth knowing before concluding anything about scrutiny. An edit reverted by a human carries a reason, a user, sometimes a talk-page argument to join. An edit a filter disallows usually carries neither hostility nor judgment — it tripped a pattern built for vandalism, often over something as mundane as several links added in one edit or a phrase that shows up constantly in spam. Rewording the change and resaving it clears most of these on the first retry. Treating the block as a verdict on the edit's merits, rather than on its shape, is the more common mistake.
When a filter's pattern is actually too broad — catching legitimate edits along with the abuse it was built for — Wikipedia keeps a venue for exactly that: the edit filter noticeboard, where anyone can report a false positive and ask a filter manager to narrow it. Reports there move at the pace of any other volunteer process, which is to say not instantly, and not on request. The filter still ran its check before anyone read a word of what it stopped.
