How we rank

Two kinds of number

Derived from a sourced fact

Computed, not judged. Each of these traces back to something with a URL next to it on the cafe’s page — the price for two, the opening hours, whether the kitchen is vegetarian. Nothing here is anyone’s opinion, so there is no such thing as an unsourced score.

Value

Price for two against the city median. A cafe at half the median scores 5; one at twice the median scores 0. Prices come from Google price bands rather than bills, so this separates a cheap cafe from an expensive one and cannot separate two inside the same band — a visitor-reported spend replaces the band for that cafe.

Closes late

Closing time, on a straight line from 8pm (0) to 1am (5). A cafe closing at 11pm scores 3.

Hours open

Total hours open across the week, out of a 98-hour benchmark (14 hours a day, seven days).

Dietary options

Pure veg is worth 2, Jain options 1.5, vegan options 1.5. Anything we could not verify counts as absent, not as a penalty.

Amenities

Parking 1.5, rooftop 1, outdoor seating 1, reservations 1, cards and UPI 0.5.

Opens early

Opening time, on a straight line from 10am (0) to 6am (5). A cafe opening at 8am scores 2.5.

Google rating

Google’s aggregate rating, damped toward 4.0 until enough people have voted. A 5.0 from three reviewers scores below a 4.6 from eight hundred — otherwise the metric just measures footfall.

Reported by visitors

Wi-Fi, power sockets and how loud a room is are not published in any listing — which is precisely why they are worth knowing, and why the people who have sat there are the only source that exists. These stay blank until at least 4 separate people have reported, and every page shows how many have.

Price for two

Roughly what did two of you pay? Aggregated as a median across reporters.

Wi-Fi

How was the Wi-Fi? Aggregated as a median across reporters.

Power sockets

Could you reach a power socket? Aggregated as a median across reporters.

Quiet

How quiet was it? Aggregated as a median across reporters.

Seating comfort

Could you sit comfortably for a couple of hours? Aggregated as a median across reporters.

Coffee

How was the coffee? Aggregated as a median across reporters.

Food

How was the food? Aggregated as a median across reporters.

The weights

Each ranking uses its own weighting. A position is the weighted mean of the metrics above — nothing else feeds in.

Rules we hold ourselves to

• Every fact carries a link to where it came from, and a page has to say enough to be worth reading. A profile publishes once it holds at least seven verified attributes — roughly what the pages already here carry — and not before. The build enforces both, not a person remembering to check.
• We changed this rule, so here is what it used to be. A profile needed two separate websites to mention the cafe. That sounds stricter and mostly is not: it measures whether a cafe has a marketing budget, not whether our facts can be checked. It held back 76 profiles that carried as much verified detail as the ones already published, because those cafes have no website of their own. Counting what a page can actually tell you is the harder test, and the honest one.
• No fact without a source. Anything asserted has a link beside it. Anything unverified reads “Not stated” rather than being guessed.
• Missing data counts as typical, not as zero — and not as a free pass. An attribute we could not verify is scored at the median for that attribute across the cafes in the list. Zero would punish a cafe for a gap in our research; simply ignoring the attribute was worse, because it quietly rewarded it. A cafe with no published price skipped the value test altogether and so was judged only on the things we did know about it — which is how a national chain rated 4.0 came to sit above an independent rated 4.5 by more than twice as many people. Below 70% coverage a cafe drops out of the ranking entirely: better absent than wrongly placed.
• Nothing older than 90 days. Every page shows when it was last checked, and a stale entry fails the build.
• Median, not average, on visitor reports. One enthusiastic party cannot move a score much. One reporter counts once. Bulk reporters are excluded outright.
• No Google review text or ratings, ever. Storing them is not permitted under the Places API terms. We keep only a place_id, to link out.

Questions

Have you visited these cafes?

No. Every fact on this site is compiled from public sources and carries a link to the record it came from, so you can check any of it. We say so on every page rather than implying first-hand visits we have not made.

Then how can you score Wi-Fi and power sockets?

We cannot, and we do not. Nobody publishes those, which is exactly why they are worth knowing. They come from visitors who have been, and a metric stays blank until at least 4 different people have reported it.

Where do the rankings come from?

Each metric is computed from a fact that has a source URL behind it — price, opening hours, whether the kitchen is vegetarian — or aggregated from visitor reports. Positions are then calculated from the published weights on this page. Nobody picks the order.

Do you use Google reviews or ratings?

No. The Places API terms do not permit storing Google review content or ratings, so we hold none. We store only a place_id, to link out to Maps. Every number on this site is our own or our visitors’.

Can a cafe pay to rank higher?

There is nothing to buy. Positions are computed from published weights, so there is no lever to pull. We take no payment from cafes at all — the site is funded by advertising, and ad slots are labelled as such.

What stops a cafe owner rating their own place five out of five?

Several things. We publish the median rather than the average, so a burst of top marks barely moves an honest score. One person counts once per cafe per metric however often they submit. Anyone rating an implausible number of cafes in a day is excluded entirely. And nothing publishes below 4 separate reporters.

How current is any of this?

Every cafe page shows the date it was last checked, and the build refuses to publish anything older than 90 days. Visitor reports expire after 180 days, so a ranking reflects the room as it is now rather than as it was two years ago.