> I didn't live here when Dark Sky was still good, but Dark Sky was incredibly accurate when I lived in the Bay Area.

I'm kind of skeptical about that, given that the Bay Area has relatively poor radar coverage. The local NEXRAD site is near Mt Umunhum south of San Jose and is quite elevated, so the lowest scanning tilt has limited coverage below ~4,000 feet over much of the SF peninsula and into the Golden Gate. The consequence is that shallow maritime convection can be poorly observed by the radar, and you can frequently have low cloud decks that produce noticeable drizzle or light rain (although possibly not greatly accumulating) across the city and surrounding area wihtout seeing anything on radar. Since Dark Sky wasn't much more than re-packaged NEXRAD data, it has a GIGO problem - if the radars don't see rain, Dark Sky won't predict anything for you.

The exception are the large storm systems that come ashore a few times per year in the Winter and Spring. Those systems behave extremely linearly, so they were "easy" for the algorithms that Dark Sky used to process the radar data. That ease of analysis combined with infrequency probably skews the perception that Dark Sky performed well in the Bay Area.

For what it's worth, I was involved in a study that analyzed the performance of several consumer and enterprise products' performance for reporting and forecasting light rain specifically in the Bay Area, and Dark Sky was indistinguishable from other data products that very obviously used raw, unprocessed NEXRAD data.