Skip to content

How we rank spots

A plain average is a bad way to rank restaurants. One five-star review from the owner’s cousin would beat a place with 1,500 reviews averaging 4.7. So we don’t use one.

The weighted rating

Every spot’s score starts at the average for all HMC-certified spots and moves towards its own rating as it collects reviews. The more reviews, the more its own rating counts:

score = (v / (v + m)) × R  +  (m / (v + m)) × C

R = this spot's Google rating
v = how many Google reviews it has
C = the average rating across every certified spot   (right now: 4.30)
m = how many reviews it takes to count fully         (right now: 100)

C and m aren’t guesses. They’re recomputed every night from the whole catalogue: C is the mean rating, and m is the 40th percentile of review counts among places to eat (kept between 25 and 250, so it can’t drift somewhere silly while the catalogue is small).

What that does in practice

With C = 4.3 and m = 100:

SpotRatingReviewsScore
Popular staple4.71,5004.675
Solid local4.62204.506
Small gem4.9404.471
One review5.014.307

The single 5.0★ review lands almost exactly on the average — one review carries about 1% of the weight — while the 1,500-review 4.7★ keeps nearly all of its own signal. That’s the ordering a person would give you if you asked them.

Hidden gems

A gem is highly rated, under-discovered, and not so small that it’s noise: at least 12 reviews, at most the catalogue median, and 4.6★ or better. For that tab we use a much smaller m (15), which lets 40 genuine reviews at 4.9★ shine without letting three reviews at 5.0★ through.

Most popular

Lifetime review count blended with the last 30 days, so a place with 4,000 old reviews doesn’t permanently pin the top over a 900-review place gaining 80 a month.

Where the numbers come from

Spotted something wrong? Every spot page has a “report a problem” link, and it goes straight to a human.