The user-reported vs source-reported thing is bigger than people think. I work with car listing prices and the gap between what people say something costs and what the source actually shows is basically the whole reason my product works. Pulling from the warehouse numbers is the right call even if the coverage is smaller.

It's an extensive dataset (Gasbuddy is one of the most used apps in the US) and they have sophisticated techniques to prevent fake/incorrect prices. I've used it for years and have only come across incorrect prices twice.

Costco prices are great but it should not be used to calculate things like "average national gas prices". You will inevitably over-represent the types of places where Costcos are more likely to be built.