Methodology
A transparent composite score, weighted, with the math published. Review quality carries the largest weight. Community signals, editor visits, trust markers, and neighborhood social tone fill in the rest. Nothing on this page is unbought, and nothing on this site is for sale.
How we rank
Every business in the index gets a composite score from 0 to 100. That number is the weighted blend of four components. We publish the weights, the inputs, and the math so anyone can audit the result.
| Component | Weight |
|---|---|
| Review quality (Bayesian-smoothed) | 75% |
| Community engagement (saves, votes, clicks) | 10% |
| Editor visits + trust signals + freshness | 10% |
| Social sentiment (neighborhood buzz) | 5% |
Review quality (75%)
Review quality is the largest input by a wide margin. It is built by aggregating public review scores across multiple sources, then Bayesian-smoothing the result against a category-mean prior.
Bayesian smoothing matters because it neutralizes thin data. A new bakery with three five-star reviews cannot outrank a thirty-year-old bakery with fifteen hundred reviews averaging 4.5. The prior pulls every score toward what is typical for the category until enough volume accumulates to justify a higher (or lower) rank. The more reviews a place has, the more its score reflects its actual aggregate. With few reviews, the score sits near the category mean, and confidence math caps how far it can climb on volume alone.
When a category has wildly variable review volumes (coffee shops range from a dozen reviews to several thousand), the prior is recomputed per category so a niche category is not penalized for being smaller.
Community engagement (10%)
Engagement is the internal signal from people using this site. Saves, votes, outbound clicks, and return visits. A place locals keep bookmarking and clicking through to climbs. A place no one ever saves fades, regardless of how its review quality looks.
Engagement is also smoothed with confidence math so a new listing with two enthusiastic fans does not jump a long-established listing with steady, broad interest. Vote velocity is tracked but capped to limit brigading by a single neighborhood group.
Editor + trust + freshness (10%)
This component combines three sub-signals.
Editor visits
A local with taste actually shows up, eats the food, sits in the chair, talks to the staff, takes notes. Editor coverage is transparent on each business page (which places have been visited, when, and the editor's bottom-line read). Editor scores are real human ratings, not algorithmic.
Trust signals
Chamber of Commerce membership, BBB listing, verified owner claim, license verification where applicable. A verified, claimed business gets a small but real bump because someone is on the hook for the listing's accuracy.
Freshness
Review velocity matters. A place that hasn't earned a new review in two years loses points until someone re-verifies. Same with last-visited date for editors. Stale data drifts toward the category mean over time.
Social sentiment (5%)
Buzz across neighborhood social channels, AI-classified for tone and topic. We never quote, screenshot, or republish anyone's posts. We extract paraphrased themes (for example, "locals love the patio," "service has slipped lately") and weight them by tone and volume.
Social sentiment is the smallest weight on purpose. It is noisy, vulnerable to viral one-offs, and easy to game. It refines tie-breaks and surfaces emerging stories. It should not, on its own, move a business multiple ranks.
What we don't do
- Sell ranking position. There is no paid tier, no "featured" upgrade that affects rank, no rank boost in exchange for a claim purchase.
- Republish individual reviews or social posts. We aggregate and paraphrase. We never quote.
- Hide score changes. Every adjustment to the formula is logged and dated.
- Auto-rank thin-data businesses high. Confidence math caps how far a low-volume place can climb.
- Pretend the algorithm is neutral. The weights reflect choices we made. They are published so they can be argued with.
Frequently asked
Can businesses pay to rank higher? +
No. There is no paid placement, no tier upgrade, and no rank boost available for purchase. Sponsored placements, when they exist, are clearly labeled and ranked separately from the editorial list. Paying for a claim verification does not affect ranking either.
Why use Bayesian smoothing instead of a simple average? +
A simple average lets a coffee shop with one five-star review beat a coffee shop with four hundred reviews averaging 4.6. Bayesian smoothing pulls every score toward the category mean until enough reviews accumulate to justify a departure from typical. Thin-data places cannot leapfrog established ones on a fluke.
What sources feed the review quality score? +
Public review aggregations from multiple platforms, weighted by reliability and volume. We do not quote, screenshot, or republish individual reviews. We aggregate the underlying scores, smooth them, and surface the rank.
How is social sentiment measured without scraping posts? +
Themes are extracted and AI-classified for tone (positive, neutral, negative) and topic (food, service, ambience, value). We never republish original text. The output is a tone-weighted theme summary, paraphrased.
How often are scores recomputed? +
Review quality and engagement signals refresh on a rolling weekly cadence. Editor visits, trust signals, and review velocity are recomputed whenever new data arrives. Freshness penalties accrue continuously between verifications.
What happens if a business disputes its score? +
Owners can claim their listing and request a re-verification visit. We will not adjust a score in response to a complaint without new data, but we will explain exactly which sub-score is dragging the composite and what would move it.
Published 2026-05-26. Last updated 2026-05-26.