Dogtrousers
Lefty tighty. Get it righty.
"twitter scraping"? FFS. It's the sort of technique that might work if you're trying to crowdsource on a massive level, but everyone's watching the same streams, so you might as well just watch pro cycling.
The obvious point to start would be estimating underreporting - do some surveys of riders about car incidents and see how much it turns up you didn't know about. Use PCT/CT riders to make sure you've got small races covered. Cross reference with which races they've ridden. That's just your scoping.
Not a statistician or survey expert or sociologist or anything, but this really shouldn't be too difficult
I may be doing them a disservice. Maybe there are robust procedures around the bit where they say "Race commisaires or other UCI stakeholders can then add more incidents" Maybe things have moved on enormously since the link below (which is all I could find) was written.
https://incidentdb.sportsdatascience.be/#/