Methodology

How the numbers on a venue page are computed, and just as importantly, what they don't claim.

This methodology is the research engine behind Platerian: every ranking, comparison, and market page in the publication draws on the calculations described below.

Bayesian shrinkage, in plain words

A venue's displayed rating is not its raw average. It is adjusted using a standard statistical method called Bayesian shrinkage: the raw average is pulled toward the average for its comparison cohort (venues in the same category and market), in proportion to how few reviews back it up. A restaurant with 800 reviews sits close to its own raw average. A cafe with 12 reviews is pulled much closer to the cohort average, because 12 data points aren't enough to trust on their own.

Cohorts and denominators

Every comparison line names the cohort a venue is measured against and discloses how many venues or reviews that cohort contains (the denominator). A confidence read without that context isn't published here.

What every number carries with it

A rating or count on this site always carries the review count it's based on and the date it was last computed. A number shown without that context would be misleading, so it isn't shown that way.

Combined rating

Some venue pages also show a combined rating across Google and Yelp: it takes each platform's own review count as a weight and averages the two star ratings together, so the platform with more reviews counts for more in the result. This combined number is a plain, weighted average for reading the two platforms as a single figure, not a ranking method. It's different from the ranking figure shown elsewhere on this site: rankings use the Bayesian-shrunk Google rating within a comparison cohort; the combined rating doesn't go through that cohort fusion and never sets a venue's rank. Google and Yelp count reviews differently and reflect different reviewer populations, so this combined figure is a reading convenience, not proof that either platform's number is more correct.

What we don't claim

• No trend or momentum figure is shown unless at least two dated snapshots exist to compute one honestly. Until then, that field simply doesn't render, it isn't estimated.
• No public review source is listed as active for a venue until that source's data is actually flowing into this site. A source planned for later is never shown as already live.
• No sentiment, ranking, or comparison figure is invented to fill a gap. Missing data stays missing.

Update cadence

Each page updates when the underlying public data is re-collected. There isn't a fixed public schedule for that yet at this stage; it will be published here once it stabilizes.