How we rank restaurants

Better Bet Ranks compares restaurants within the same city and price category. Best Experience and Best Value answer different questions. Version 36 displays 372 restaurants with Experience ranks and 206 with preliminary Value ranks. Each of the 12 city/question/price categories has at least 30 ranked restaurants. These are descriptive results for the restaurants with available evidence, rather than definitive rankings of every restaurant in a city.

One source, one column

Each genuine ranked list or rating snapshot supplies a column in an incomplete restaurant comparison matrix. We compare two restaurants only when a column records both. Missing entries stay missing; they never mean last place. Scores from different publishers are converted to within-source ranks, so incompatible rating scales are not averaged.

One column can support preliminary ranks. With a single numeric source, CPDP reproduces its ordering, including ties. That is a useful starting point; it does not establish agreement across independent sources. A lone rating without another recorded restaurant does not establish relative rank. A verified original publisher rank can still be shown if only one entry is retained: an original #7 stays #7.

CPDP and CTPDP

For restaurants k and i, estimate P(k ≤ i) using the fraction of columns recording both in which k’s score is no greater than i’s. Ties count in both directions. A never-observed pair has no probability estimate.

We use a deterministic largest subset S in which every pair is observed in at least one column. CPDP(i) = 1 + the sum of these pairwise estimates over other members of S. CTPDP(i) = 1 + the count of estimates greater than one-half. Higher scores are better; displayed rank 1 is best. Equal scores receive equal competition ranks. The downloadable results include both scores and ranks.

The table labels these results “Better Bet Ranks.” Restaurants outside the pair-covered subset can retain separately labelled “Source rank” results. Each source list has its own scope. A #1 in a cuisine poll does not become a city-wide #1, and ranks from separate groups are not directly comparable.

Value: a broader, better starting point

This release has 206 explicitly numeric OpenTable aggregate Value ratings: Orlando 40 Budget, 30 Mid-range and 33 High-end; Greater Boston 33 Budget, 38 Mid-range and 32 High-end. The records form one platform column per city, not 70 independent judges. Overall stars, food scores, price symbols and review counts are not substituted for Value scores.

Each imported value has its restaurant identity and source URL in the source records. Collection date is October 27, 2026. Some ratings were visible through indexed versions of the primary pages rather than the current booking view. Index dates vary, and URL variants can display different snapshots. Source variants can report different snapshots. During the v33 import we preferred complete aggregate blocks, then more recent crawl dates and non-photo pages; score outcomes were never a selection criterion. Separate score and review-count URLs retain their provenance. Spanish “Valor” is the same Value dimension, not a separate source. These are recorded snapshots, not a real-time feed. Check the linked restaurant page before making a reservation.

The earlier pilot used three selected Tripadvisor Value subratings per restaurant across 20 restaurants. It remains available in the source audit as supplemental history, but no longer contributes to current Value ranks. Selected review means and platform aggregates are not pooled as interchangeable observations.

OpenTable Best Value recommendations remain a separate, optional list. Their page order has not been verified as Value rank order. GAYOT’s Orlando Best Value guide explicitly uses alphabetical order. Neither is turned into a ranking column. The v32 exclusions remain excluded pending further status checks; this message alone does not establish permanent closure.

Which restaurants we added

Coverage research prioritized familiar restaurants and places with substantial diner participation. The 108 new identities all had at least 200 recorded OpenTable reviews when collected; 99 had at least 300. Existing restaurants selected from published guides remain eligible even when their recorded review count is smaller. Within the collected evidence, we did not select only restaurants with high scores. Review counts guide collection and appear as context; they do not weight CPDP scores, break ties or act as independent observations. This does not establish that these are the most popular restaurants in either city, and the search is not an exhaustive census.

The v33 release imported 139 checked aggregate records, each containing separately recorded Value and overall ratings, and preserves all prior usable Value cells. Source pages marked unavailable, unresolved branches, special-event products, cruises and incomplete aggregate snippets were kept out of the new import. A source being unavailable does not establish permanent closure. The exclusions file documents unresolved candidates.

Experience sources

Experience retains the Time Out Boston numbered guide, Boston Magazine, Orlando Weekly category polls and recorded platform ratings, and adds the separately captured overall ratings from the new source records. It covers Orlando 73 / 59 / 45 and Greater Boston 76 / 67 / 52 ranked restaurants in Budget / Mid-range / High-end order. Original source positions, date and scope are retained. Category-specific poll ranks remain outside the overall CPDP calculation. Source records from earlier releases have not all been reverified for operating status.

Boston coverage update in v35

We researched all 42 Boston candidates that lacked Experience ranks in v34. Of these, 39 now have a checked primary-source overall rating from OpenTable or Tripadvisor. Each rating updates one cell in its existing platform/city column. Search-result or featured-list position was never treated as an ordering. Source variants and indexed snapshots can differ; the curated records retain source URL, crawl label, review count, matching notes and conflicts. Review counts are context only, with no minimum for these previously listed candidates: some ratings have only four recorded reviews and should be read accordingly. Overall stars enter Experience only; Value cells are unchanged.

Roxy’s generic launch entry is resolved explicitly to the Allston location at 485 Cambridge St. The Mad Monkfish matches the Thelonious Monkfish listing by address, telephone and primary-source alias; a separate unrated duplicate listing is not combined. Saus at 33 Union St and Menton are reported closed and appear under Past listings, without new ranking evidence. Maple & Ash Boston remains unranked because a reliable numeric overall aggregate was not captured; eight reviews alone do not determine a score. This targeted check does not verify the operating status of the entire catalog.

39 new Experience records · All 42 research outcomes

Orlando coverage update in v36

We researched all 32 Orlando candidates without Experience ranks in v35. Of these, 29 now have a checked primary-source overall rating from OpenTable or Tripadvisor. Each observation enters the existing city/platform column, with ties and missing cells preserved. List pages were used only for their explicit numeric overall ratings; search, featured and nearby positions were not converted to ranks. Ratings apply only to Experience, and all Value inputs and ranks remain unchanged. Depending on pair coverage, an added restaurant receives a CPDP subset rank or a separately labelled source-local rank.

The source file retains URL, retrieval date, index age, count and identity notes. Where snapshots conflict, identity correctness and an explicit aggregate block are prerequisites; newer snapshots are preferred when comparable. Some newer numeric list snapshots are available when direct page scores are missing. We never select a snapshot because its score is higher. Counts can be small: Meng’s Kitchen has one recorded Tripadvisor review, and Foreigner and Underground Public House each have five recorded OpenTable reviews. These are descriptive preliminary inputs, not independent replicated rankings; review counts neither weight nor replicate source comparisons.

Three restaurants remain unresolved. Bricks & Bowls spans several branches, and the old Millenia delivery listing cannot be reassigned to Park Ave. Cupid’s current Longwood restaurant cannot inherit ratings for its former Lee Road location or California branches. Gabriel’s College Park entry cannot inherit the Curry Ford Road branch’s ratings; directory-mirrored Google scores were not imported as original platform records. Primary Yelp access was blocked, and delivery-platform scores were not treated as dining-experience ratings. These are evidence/identity gaps, not last-place judgments. All 32 outcomes and source notes are downloadable.

29 new Orlando Experience records · All 32 Orlando research outcomes

Price and location

Existing catalog tiers are retained. New entries use the source’s approximate brackets: Budget ($30 and under), Mid-range ($31–50), High-end ($50 and over). The boundary at $50 follows the publisher’s overlapping labels. These are approximate categories, not quotes for a complete meal. Some existing tiers differ from current source brackets; for example, Kres Chophouse, Capa, Bar Mezzana and Brassica retain their inherited categories. Brackets describe the publisher’s page category and do not guarantee that every entree, a complete meal, drinks, tax or gratuity fit that amount. A later price audit should reconcile these differences. Chain branches are separate identities. Greater Boston includes the recorded surrounding communities, identified by location.

Uncertainty and dependence

Publisher lists, platform summaries and repeated occasions can share reviewers and selection mechanisms. More pages do not mean more independent replicates. No confidence intervals or simultaneous 95% rank claims are made. Formal CPDP inference requires sampling and dependence assumptions appropriate to the observations. These source-based comparisons do not use arbitrary weighted composite scores.

Explore and contribute

The website shows only restaurants with ranks in the main lists. Candidates without evidence for the selected question are kept separately. The GitHub Pages version displays the bundled snapshot and saves visited places in your browser. Collecting community comparisons requires the included backend. Experience and Value votes remain separate. “I’ve tried this” saves a visited-place selection, not a vote or rating. It does not add the restaurant to a ranking. In the live app, compare two restaurants you have tried within the same city and price tier; the system offers an available pair. Offline selections remain in the browser only.

Download all results · Coverage by category · Return to restaurants