Incident footage gets posted for a reason people rarely state out loud: it is a bid for a verdict. You want strangers to agree that the other driver was wrong. That motive is fine. The problem is that the evidence you are submitting contains a lot of people who did not consent to being part of the argument.
Three things in the frame, only one of which is the point
The other party. Their registration is arguably the one detail with a claim to being relevant, and it is also the one most likely to trigger a pile-on aimed at a specific person. Insurers and police already have it from you directly. The internet does not need it.
Everyone else. Every car in the queue behind, every parked car on the street. They have nothing to do with the incident and their registrations are in your video at readable resolution.
You. This is the one people miss. A dashcam clip usually begins or ends somewhere near where the driver lives, and the route across a few seconds of footage is enough to place it. Street signs, house numbers and shop fronts do the rest.
Why manual editing does not work here
A fifteen-second clip at thirty frames per second is four hundred and fifty images. A plate that stays in shot for four seconds needs covering in a hundred and twenty of them, and it moves in every one. This is the case where the choice is genuinely between an automated pass and not doing it at all - which is why most posted footage is simply not redacted.
What you want from the tool is tracking across frames plus the original length and audio preserved, because a clip that has been re-encoded, cropped or silently shortened is worth less if it later has to serve as evidence.
Keep the original
Redact the copy you publish. Never redact the only copy. If the footage matters to an insurance claim or a police report, they need it untouched, and a redacted version is at best unhelpful and at worst looks like tampering.
The practical arrangement is two files with obvious names: one original, archived and not shared, and one public version that has been through the redaction pass and had its metadata stripped.
Before you upload
Scrub through the exported clip rather than the editor preview, and pay attention to the moments where a car enters or leaves the frame. Those are where tracking is weakest, and a plate that is covered for three seconds and visible for four frames is a plate that is visible.