How we rank
Most restaurant guides sort by star rating. That rewards whoever games reviews hardest, and tells you nothing about whether the kitchen is clean. We score five independent signals instead — four of which a restaurant cannot buy, solicit or astroturf — and we publish the full working on every venue page.
The five signals
Weights are fixed and published. They were chosen so that no single gameable signal can carry a venue to the top on its own: reputation is 30 of 100, so a venue with flawless reviews and nothing else cannot outrank one with strong evidence across the board.
Food hygiene
30 ptsThe Food Standards Agency rating from the venue's last local-authority inspection, 0–5.
Not gameable. A statutory inspection by an environmental health officer. It cannot be bought, solicited or astroturfed, which makes it the single most trustworthy signal available.
Reputation
30 ptsGoogle rating and review count, with the rating shrunk toward the city mean in proportion to how thin the evidence is.
Gameable. Public sentiment at scale is genuinely informative, but a raw average treats three reviews the same as three thousand. The shrinkage fixes that.
Verification
15 ptsA claimed Google listing, a live website, a published phone number, and an operational trading status.
Not gameable. Establishes the place exists, is trading, and is who it says it is. Cheap to satisfy honestly, awkward to fake convincingly.
Accolades
15 ptsRecognition by an independent awarding body — the Michelin Guide and The Good Food Guide, weighted by the strength of the award.
Not gameable. Anonymous professional assessment against a published standard, independent of the venue's own marketing. Awards from different bodies add together, so two independent judgements outrank one.
Inspection freshness
10 ptsHow long ago the hygiene inspection actually happened.
Not gameable. A rating of 5 from six years ago is a weaker claim than a 5 from last month. Evidence decays.
Why a 5.0 does not beat a 4.6
A perfect score from a handful of reviews is a weaker claim than a strong score from thousands. So before scoring, every rating is pulled toward the city-wide mean of 4.2 by an amount that depends on how many reviews back it — a standard Bayesian shrinkage, with a prior weight of 60 reviews:
adjusted = (n × rating + 60 × 4.2) / (n + 60)
| Venue | Raw | Reviews | Weighted |
|---|---|---|---|
| DAA Centre Manchester | 1.0★ | 1 | 4.15★ |
| Rolled | 4.3★ | 43,927 | 4.30★ |
Live examples from the data — the thinly-reviewed venue moves toward the mean; the heavily-reviewed one barely shifts, because its evidence is strong enough to stand.
Scored on what we know, not what we lack
Building this dataset is incremental — hygiene data covers every venue, Google data is added district by district. A venue scored against the full model before we have looked it up would appear to fail for reasons that are a property of our pipeline, not of the restaurant.
So each score is divided by the weight of the signals we actually hold, and every venue page shows an evidence coverage figure alongside it. Signals we have not gathered are greyed out and excluded, never counted as zero. Below 60% coverage the score is marked provisional rather than presented as settled.
Hygiene is a gate, not a weighting
A hygiene rating of 2 or below sits under the FSA's “generally satisfactory” threshold. Any venue at that level is placed in Flagged regardless of everything else — no volume of good reviews and no award can lift it out. Those venues stay visible rather than being hidden, because concealing them would make the guide less useful, not safer.
The top tier has gates in the other direction: Verified & Acclaimed requires a hygiene rating of 4 or 5, at least one independent award, and 75% evidence coverage. A venue cannot reach it on reviews alone.
The tiers
Top hygiene, strong verified reputation, and recognised by at least one independent awarding body.
Hygiene rating of 4 or 5 and a well-evidenced reputation across a meaningful number of reviews.
Reasonable evidence across hygiene, reputation and verification. No red flags.
Thin or mixed evidence. May be new, small, or simply under-documented.
Not enough independent signal to rank fairly yet.
Hygiene rating of 2 or below, or a status warning. Shown for transparency.
Where the data comes from
- Food Standards Agency
- Hygiene ratings, inspection dates, addresses and coordinates for every registered food business. Open Government Licence. This is the base layer and we hold it permanently.
- Google Places
- Ratings, review counts, websites, opening hours and price levels. Cached for no more than 30 days and refreshed, in line with the Google Maps Platform terms. Reviews themselves are not stored — venue pages link out to Google to read them.
- Michelin Guide & The Good Food Guide
- Award data only — venue, location, designation and year. Both guides' written assessments are their own intellectual property and are not reproduced here. Michelin entries are matched on postcode; Good Food Guide entries on coordinates, since the Guide records only a town.
What this does not yet do
- Cuisine tagging is incomplete. Cuisine is inferred from the venue name and Google's category types. Distinctive names classify well; a place called simply “Aori” does not. We leave those blank rather than guess, so an untagged venue means unknown, not uncategorised.
- Some award holders cannot be matched. Seven Good Food Guide entries have no counterpart in the hygiene register under a recognisable name — usually because the venue is registered to a company name, or opened after the last inspection round. They are recorded as unmatched rather than attached to a plausible-looking neighbour. AA Rosettes and the Manchester Food & Drink Festival awards are supported by the model but not yet loaded.
- Google coverage is uneven across the region. 6,407 of 15,378 venues have been looked up so far. Manchester, Salford and Trafford are largely complete; the other seven boroughs currently carry hygiene-only provisional scores, marked as such on every card and venue page. They are being enriched borough by borough.
- Nothing here is a personal recommendation. This measures the weight of independent evidence behind a venue. It cannot tell you whether you will enjoy the food.
Spotted something wrong? Scores are derived, not hand-set — if a venue looks misjudged it usually means a mismatched listing or a stale inspection, both of which are fixable.