How to read the scores
The city gives a letter, we give a number out of 100. They both help answer different questions, so you can make a more informed decision.
The short version
The letter in the window describes one inspection. Our number describes the pattern across every inspection we can see, which is usually about three years.
A restaurant can hold an A and score badly here. That is not a contradiction and it is not us disagreeing with the Health Department. It means the most recent visit went well and the history around it did not.
The number out of 100
Every restaurant starts at 100 and loses points for things found in its inspection record. Nothing is added for being new or unremarkable, a place with a clean history and nothing notable stays near the top.
| Score | Verdict | What it means |
|---|---|---|
| 80–100 | Recommended | Nothing in the record gives pause. |
| 60–79 | Acceptable | Minor or old issues. Worth a glance at the detail. |
| 35–59 | Caution | A real pattern, or something recent. Read the red flags. |
| 0–34 | Avoid | Serious, recent, or repeated. We would not eat there. |
Lower is better on the city's score, higher is better on ours. The city counts violation points, so 13 is good and 40 is bad. We count out of 100, so 90 is good and 20 is bad. The two run in opposite directions, but just think of ours as a percentage grade out of 100.
What costs points
Roughly in order of how much they matter:
| Finding | Cost | Why |
|---|---|---|
| A posted B, C or Grade Pending | caps the score | The city's own warning, taken at face value. A posted B caps the score at 34, a C at 14 and Grade Pending at 42. A restaurant that has never been graded is capped at 55, and one with no grade on record at 50. |
| Vermin | up to −48 | Rats, mice, roaches, flies and conditions that attract them. Weighted by how recent the finding is and how much has been done since. The same pest cited three or more times costs another 8. Until an inspector has been back and found nothing, and for a year after a finding either way, the score cannot go above 79. |
| A grade that has gone stale | −18, and caps the score | An inspection scoring 14 or more earns no grade, so the previous A stays in the window. The median wait for the next inspection is 152 days, and 40% of them take longer than six months. The placard you see from the street today can be describing a visit that has since been contradicted, so the score cannot go above 55, or 40 if the latest inspection scored in the C range. |
| A new restaurant's first A | caps the score | A new restaurant gets one inspection after it opens, and the next routine one takes a median of almost 17 months. If that first inspection is all there is, the score cannot go above 85, and above 75 once it is 600 days old. If the restaurant failed that first inspection and got its A on the re-check, the cap is 65 until a routine inspection happens: half of restaurants like that fail their first one. |
| Grade repair | −8 to −24 | Failing an initial inspection and passing the re-inspection is normal and legitimate, but the failed visit still happened, and the A in the window came from the re-check. Costs 8, or 14 if the failed score was in the C range, and another 10 if that inspection found vermin. Counted for each one in the last three years. |
| Health Department closure | −5 to −45 | A closure means an inspector judged the kitchen an immediate health hazard. It costs 45 in the first year, 25 in the second, 12 up to four years and 5 after that, so a recent one outweighs everything else, including a currently posted A. A restaurant listed as closed right now scores 0. |
| Temperature and contamination failures | −8 | Three or more recent findings of the kind that actually cause foodborne illness like cold food held too warm or raw food touching cooked food. |
| Getting worse over time | −7 | Recent inspections averaging 8+ points worse than earlier ones. |
| A borderline A | −5 | 12 or 13 points means "just passed", not "clean". |
| Unusually many critical violations | −6 or −12 | Only the worst 10% are flagged, and the worst 5% cost double. The median restaurant has about 1.7 critical violations per inspection, so an ordinary rate is not worth pointing out. |
| Things that matter for how you eat | up to −10 | Only if you have set eating preferences. Violations that matter more for how you eat count for more. |
We also show when a chain was cited for not posting required information: chains of 15 or more locations must post calorie counts and flag high-sodium items on their menu. This is shown but not scored. It does not say the food is unhealthy, it says they were required to tell you and did not.
What earns points back
Just one thing: how a restaurant compares to its own cuisine, worth at most ±6.
This matters more than it sounds. Initial inspections fail 19% of the time at donut shops and 63% at Indian restaurants. A kitchen doing more raw preparation has more that can be cited, so judging both against a single citywide average flatters one and punishes the other. We compare each restaurant to the range its own cuisine actually occupies.
It is kept deliberately small, and it is applied before every other limit, so being a well-run example of a demanding cuisine can never lift a place past a vermin or closure cap. This helps correct for differing exposure but doesn't excuse an unclean kitchen.
Confidence, and why it is separate
A score built on one inspection is not the same as a score built on twelve, so we say which you are looking at rather than hiding it inside the number.
- High: three or more inspections across at least 18 months.
- Medium: two inspections, or less than 18 months of history.
- Low: one inspection or none. With none at all, the score is capped at 55, because absence of evidence is not evidence of cleanliness.
There is one place where a thin record does move the number itself: a new restaurant's first A, described in the table above, because we measured how often those turn out to be wrong.
Reviews can only lower a score
When you allow third-party content, we look at the Google rating. However, it is kept strictly separate and can only ever subtract. A great rating cannot lift a restaurant with a vermin history, because customers cannot see the kitchen.
The health score is shown on its own alongside the combined one, so you can always see which half moved the number. If you decline third-party content, the health verdict is identical and only the star rating disappears.
Earlier history
The city publishes inspection information in a rolling window: roughly three years back from each restaurant's most recent inspection, active establishments only. Everything older is dropped and cannot be requested again. Katz's has been open since 1888, and the official dataset only goes back to 2022.
We keep what the city drops, recovered from archived copies of the same public dataset. Where a restaurant has that older record it appears under Earlier history, collapsed, and it is not counted in the score. It is there because it provides the historical insight into how the restaurant has fared in the past (and we were honestly curious!). More on where that comes from.
What this is not
- Not the city's opinion. The data is theirs and public. The judgment is ours, and they might weigh it differently.
- Not a food quality rating. A superb restaurant can have mice just as a mediocre one can be spotless.
- Not current. An inspection is a snapshot of one day. Ours is a snapshot of snapshots. Neither knows what the kitchen looks like tonight.
- Not a substitute for the placard. If the grade in the window disagrees with us, both facts are worth having.