I would recommend against building a score and always stick to booleans whenever possible. Having a numeric score means you're trying to extract nice-to-have data points, and in my experience this always leads messy codebases and false assumptions.

If you allow "slightly unhealthy" scrapers in production, almost inevitably the state of your scrapers will always be "slightly unhealthy". Save yourself the trouble and always treat it as either "it works" or "it doesn't work", no in between. Your first iterations will probably break every day, but eventually you'll get to a happy place.