You're assuming a correlation between price and quality. But price has at least two aspects: the base cost of that type of food; and the supply / demand for the quality of preparation.
You could have cheap sushi that's very good; similarly, you could have excellent sushi that's quite expensive compared to other types of food, but is still cheap for that level of quality. And a third place could have the same price as the second, but much lower quality. Which should get $ vs $$$?
Furthermore, not all cheap restaurants are cheap because they are competing on price. Sometimes they are even run at a loss to sell something else. For example, El Bulli, when it was still in operation, was an expensive restaurant on an absolute basis; but relative to the quality of the food, it was very cheap - because it was not even profitable, the restaurant was run at a loss. Instead, it built a brand around Ferran Adrià that sold books and other things.
Similarly, a local restaurant to me is called Corner Room in Bethnal Green, London. It's cheap food selected from a set menu - and rather than controlling demand with prices, it does it with timing and willingness to wait, since bookings are not taken. The same chef runs a much more expensive restaurant, Viajante.
The value of ratings is only as good as the preference function of the raters. If the raters are drawn randomly, then sure, you can probably rely on the rough consensus. But raters are seldom drawn randomly; for one thing, people are much likely to rate bad experiences than good; people with iPhones and who are generally more plugged in are more likely to rate, and there are demographic and geographic patterns to the popularity of various different rating systems.
Using Yelp in an area where all the locals use a different system, you'll end up with ratings from tourists, who may be bowled over by the novelty of sub-par local food. If your own preference function has a primarily foodie basis, but the website you take reviews from has a more balanced function of location, service and price, then it stands to reason that many places will get much higher ratings than you would rate it yourself, because you assign lower weights to those other factors.
If, as you state, you find ratings are a good match to your own preferences, all that means is that you have a preference function somewhere in the middle of the rating sample. But not everyone is like that.