How Crave ratings work
Crave rates a particular dish at a particular restaurant branch. The displayed score is the same score used for rating order and Crave Rave awards.
Initial estimates
Before a dish has an eligible Crave review, its estimate is the source restaurant rating plus 0.2, capped at 5. It is labeled “Initial estimate,” has zero Crave ratings, and has no model confidence interval. Without a usable source, the dish is “Not yet rated.”
Crave reviews
Verified accounts contribute one current rating per dish, in half-star increments. A genuine new visit can replace a rating. Editing text does not refresh the visit date.
Recent visits carry more weight, with a 180-day half-life. A current, regionally applicable Tastemaker review has 1.5 times the relative influence. It still represents one account, not extra people.
Model details
Relative weights are recency × Tastemaker influence. Effective evidence is the smaller of Kish effective size and total recency weight. A positive nine-bin Dirichlet prior has a target mean clamped to 1.2–4.8 and strength 3, aging from the original source snapshot. Without a source, the prior mean is 3.5 and strength 1.
The score is the analytic expected star value. A 90% model interval uses reproducible Dirichlet samples; it describes uncertainty about the mean, not the spread of individual experiences or guaranteed real-world coverage.
A fresh 5.0 seed followed by one fresh five-star review becomes about 4.85. A 4.7 seed becomes 4.775. The transition preserves uncertainty.
These are explicit product assumptions, not scientifically proven optimal defaults. Version: crave-1.1.0.
Crave Rave
A Rave goes automatically to the highest-scoring offering for a canonical dish in a named region, with at least five qualifying restaurant locations. Every winner must score strictly above 4.5 before rounding. Verified nominations break exact top-score ties only; if nomination counts also tie, leaders share “Joint top rated.”
Search radius, budget, dietary filters, advertising, and follower counts never change the award competition. If any competing dish uses an initial estimate, the award discloses that seeded evidence.
You can nominate one offering per canonical dish and region after an accepted, non-conflicted rating. Nominations do not add stars or outweigh a higher rating. Tastemaker nominations have no extra weight.
Tastemakers
Admins assess sustained participation, independent restaurant coverage, useful contributions, and moderation history. Initial screening signals are 60 account days, 30 accepted reviews, and 10 restaurant locations. Quality and conflicts matter; positive ratings and agreement with the majority do not. Tastemakers have no special nomination or winner-selection powers.
Moderation & appeals
Conflicted, removed, or confirmed fraudulent ratings are excluded. Duplicate prose and coordinated bursts may be held for review. A low rating, shared household, new account, or missing photo alone is not evidence of fraud. Use “Report” on a review or the appeal action in your account. Decisions are audited.
Delivery
The delivery filter includes only restaurants with a verified accepting-orders check that has not expired. Checks last at most five minutes. Provider links do not guarantee that a particular dish can be delivered to your address; confirm with the service before ordering. Unknown or stale status is excluded.
Launch coverage
The launch target is 40 verified Novato restaurants selected by Yelp review count. No licensed launch records have been imported yet. Any development catalog is labeled and cannot produce real awards. The initial award boundary must be confirmed before activation.
Privacy
Your email, lists, reports, and security data stay private. Your chosen display name and published reviews are public. Location is requested only when you choose current location and is used for discovery. Call, delivery, and directions taps add private rating reminders; they are not proof of a visit and never create ratings automatically. Swipe to dismiss a reminder.