Eat or Beat scores restaurants from public health-inspection records across every city we cover. We’ve processed hundreds of thousands of individual inspections and millions of violation records. Here’s what that data looks like at altitude.
Most Restaurants Are Fine
The least clickable finding and the most important one. The majority of restaurants, in every city, have acceptable inspection records. They pass their inspections. Their violations are minor and operational. The kitchen isn’t perfect, but it’s not dangerous.
This matters because the narrative around restaurant safety tends toward fear. Every local news segment leads with the worst-case kitchen. Those kitchens exist. They’re in our data. But they’re a small fraction of the whole.
The Bad Restaurants Are Really Bad
The restaurants in BEAT territory aren’t just a little below average. They have compounding problems: multiple failed inspections, repeated critical violations, and patterns that suggest systemic issues rather than one-off mistakes.
A restaurant doesn’t score below 50 on our scale by having a warm cooler once. It gets there through failed inspections, critical violations across multiple visits, and a history that shows the problems aren’t getting fixed. The gap between YOUR CALL and BEAT isn’t gradual. It’s a cliff.
Recency Matters More Than History
Our scoring model applies exponential decay to older inspections. The most recent inspection counts the most. Three inspections ago, it’s fading. Five inspections ago, it’s barely a whisper.
This reflects a basic reality: kitchens change. Management turns over. Staff gets trained or leaves. Equipment breaks down or gets replaced. A failure from three years ago tells you less about today’s kitchen than a clean pass from last month.
Recovery is possible and visible in the data. A restaurant that failed, fixed its problems, and has passed three consecutive inspections will see its score climb. The failure doesn’t vanish — it’s still in the history — but it stops dominating the verdict. The system should reward improvement, not just punish mistakes.
The Data Is Messier Than You’d Expect
Government data is not clean. We’ve encountered inspection records with no date, violations with no associated inspection, restaurants with duplicate entries under slightly different names, addresses that don’t geocode, and entire violation categories that changed classification mid-dataset when a city updated its code.
Every city we add requires a custom data transformation pipeline — a Python script that takes raw data, cleans it, normalizes it, maps it to our schema, and outputs a database that the site can render. There is no generic government data ingestion tool. Every dataset is its own puzzle.
This is, incidentally, why nobody else has built this. The data isn’t secret. It’s annoying. The barrier to entry isn’t access. It’s patience.
More Inspections Means More Violations
Cities that inspect more frequently generate more violation records per restaurant. This doesn’t mean their restaurants are dirtier. It means they’ve been observed more.
A restaurant with 15 inspections on file will almost certainly have some violations in the record. A restaurant in a city that inspects once a year has fewer chances to get dinged. More data is better data, but it also means you can’t compare raw violation counts across cities without adjusting for how often each city actually sends someone to the kitchen.
Every score on Eat or Beat is computed from public health-department records. We don’t visit restaurants. We don’t accept payments from restaurants. We translate what’s already on file.