What a Health Inspector Actually Checks (and What the Score Doesn't Tell You)

Most people encounter restaurant inspections as a single artifact: a letter in the window, a number on a placard, a colored sticker by the door. That artifact is the end of a process, and like most summaries, it hides more than it reveals. A health inspector spends anywhere from 45 minutes to several hours inside a kitchen. What they write down follows a structure — and once you know the structure, the score in the window becomes a lot more legible.

This is a walk through what an inspector is actually looking for, how those findings get weighted, and — more usefully — what a single score can and cannot tell you about the place you're about to eat.

Critical vs non-critical violations

Nearly every U.S. jurisdiction sorts violations into at least two tiers. The names differ — critical and non-critical, priority and core, red and blue, high-risk and low-risk — but the underlying logic is the same, and it comes from the FDA Food Code that most state and county codes are modeled on.

A critical (or priority) violation is one with a direct, demonstrated link to foodborne illness. Chicken held at 55°F. A cook handling raw beef and then plating a salad without washing hands. A cooler that can't hold 41°F. These are the conditions epidemiologists trace outbreaks back to.

A non-critical (or core) violation is a maintenance, facility, or documentation issue that doesn't independently make anyone sick. A cracked floor tile. A missing thermometer in a dry-storage room. A dumpster lid propped open. Peeling paint above a prep line.

The distinction matters because scoring systems weight these very differently. In a typical 100-point deduction system, a single critical violation might cost four to nine points while a non-critical costs one or two. That means two restaurants can land on the same number by completely different routes: one accumulated eleven cosmetic issues, the other had two temperature failures. The score treats them as equivalent. Your risk of getting sick does not.

This is the single most useful thing to understand about inspection data. The composition of a score matters more than the score.

The big five risk factors

The FDA groups the conditions most often implicated in foodborne illness into five categories, and inspectors are trained to spend their attention there. Everything else on the form is secondary.

1. Temperature control. This is the biggest single bucket, and it's where most critical violations land. Cold food is supposed to stay at or below 41°F; hot holding at or above 135°F. Between those two numbers sits the range where bacteria multiply fastest. Inspectors carry a probe thermometer and use it constantly — walk-in coolers, steam tables, salad bars, the reach-in under the line, the tub of cooked rice cooling on a counter. They also check cooling procedure, which is a subtler failure mode: a large pot of soup that goes straight from the stove into a sealed container in the walk-in can sit in the danger zone for hours in the middle even though the cooler reads 38°F.

2. Cross-contamination. Raw protein stored above ready-to-eat food. One cutting board doing double duty. A sanitizer bucket that has gone cloudy and stopped sanitizing. Wiping cloths left on a prep surface between uses. Inspectors look at physical storage order in the walk-in almost immediately — raw poultry on the bottom shelf, then ground meat, then whole cuts, then produce and cooked items above. It's a hierarchy based on cook temperature, and a violation of it is visible in about four seconds.

3. Personal hygiene. Handwashing sinks that are accessible, stocked with soap and towels, and actually used. Whether the sink is blocked by a stack of bus tubs is a real and commonly cited finding. Inspectors also look at bare-hand contact with ready-to-eat food, glove use, and — importantly — whether the operation has a written employee illness policy. Norovirus spreads through sick food handlers more than any other route, and "we send people home" is not the same as a documented exclusion policy the staff has been trained on.

4. Sourcing and approved suppliers. Food has to come from a permitted source. That means invoices for shellfish with tags retained for 90 days, no home-canned goods, no meat from an unlicensed processor, no eggs from someone's backyard flock going onto the menu. This category rarely generates violations in chain restaurants and generates a disproportionate share in small independents that source informally.

5. Cleaning and sanitizing. Not the same as "looks clean." Inspectors test sanitizer concentration with strips — chlorine, quat, or iodine each have a required parts-per-million range. Dish machines get checked for final rinse temperature or chemical concentration. A spotless-looking kitchen with a dish machine running at 20 ppm chlorine when it needs 50 is a critical finding; a visually cluttered kitchen with correct sanitizer chemistry may score better.

Beyond the big five, the form covers pest evidence (droppings, gnaw marks, live activity), date marking on ready-to-eat items held more than 24 hours, plumbing and backflow prevention, water temperature, lighting, ventilation, chemical storage, and the presence of a certified food protection manager on site. These matter, but they're not where the illness risk concentrates.

What a score is a snapshot of

Here's the limitation nobody prints on the placard: an inspection score describes a few hours on one specific day, and the day is not random in the way you'd want it to be.

Timing shapes the result. An inspector arriving at 10 a.m. on a Tuesday sees a kitchen that has been cleaned overnight, restocked, and hasn't yet run a service. The same kitchen at 8 p.m. on a Saturday — three hours into a rush, two people short, the line backed up — is a different operation. Most routine inspections happen during business hours but skew toward the calmer end of the week, simply because that's when inspectors work.

Inspector variation is real. Two trained inspectors walking the same kitchen will not produce identical reports. Research on inspection consistency has repeatedly found meaningful inspector-to-inspector variance in both the number of violations recorded and how borderline conditions get classified. Jurisdictions try to control for this with standardization exercises, but it doesn't disappear.

Scores are often re-inspection scores. In many jurisdictions, a failing initial inspection triggers a re-inspection within a set window, and the posted grade reflects the second visit — after the operator has had days or weeks to fix the specific things that were cited. That's a legitimate design choice (it rewards correction), but it means the number in the window can describe a kitchen that was prepared for exactly the inspection it received.

Grade distributions compress. Where letter grades are posted publicly, the overwhelming majority of restaurants earn the top letter. When almost everyone gets an A, the A stops carrying information. The useful signal moves down into the numeric score and the violation detail — which is why systems that publish only a letter are less informative than they appear.

How to read a history instead of a grade

One inspection is an anecdote. Three or four inspections over two years is a pattern, and a pattern is what you actually want. Here's what to look at when the full record is available.

• Repeat violations. The same critical citation appearing across consecutive inspections is the strongest negative signal in the entire dataset. A one-time temperature failure is an event. The same cooler failing three inspections running is a decision.
• Direction of travel. A restaurant that went 88 → 93 → 97 is telling a different story than one that went 97 → 93 → 88, even though both average the same. New management, a new chef, or a staffing crunch usually shows up in the trend before it shows up in the headline.
• Critical count, not total count. Filter to critical/priority violations and count those separately. Two criticals and zero non-criticals is worse than zero criticals and nine non-criticals, regardless of what the arithmetic says.
• Category clustering. If the violations concentrate in one of the big five — say, everything is temperature — that points at equipment or process. If they're scattered across hygiene, sanitizing, and pests, that points at management attention. Different problems, different durability.
• Gaps in the record. A restaurant with no inspection in three years isn't necessarily safe; it may be under-inspected. Many health departments run behind on routine inspection schedules, and inspection frequency is often tied to a risk category assigned to the establishment type rather than to its actual performance.
• Closures and enforcement actions. Separate from scoring, most jurisdictions log emergency suspensions, imminent-health-hazard closures, and hearings. These are rare and unambiguous. One of them in a history outweighs a lot of point deductions.

The practical version

You are not going to run this analysis standing outside a restaurant at 7 p.m. So here's the compressed version of what's worth doing.

Before you go somewhere new — especially somewhere you're taking a group, or somewhere serving higher-risk food like raw shellfish, sushi, or a buffet — spend ninety seconds with the inspection record rather than the placard. Look for critical violations, look for repeats, look at the direction over the last three inspections. That's it. If the record shows a clean critical history and a flat or improving trend, the letter in the window is telling you the truth. If the letter is an A but the last two inspections each cited a temperature failure in the same cooler, the letter is telling you less than you thought.

And when you're standing inside: the handwashing sink is the most honest thing in a restaurant. If it's accessible, stocked, and clearly in use, a lot of other things are probably right too.

Inspection data is public in nearly every jurisdiction, but it's scattered across hundreds of county and state portals in wildly different formats — some searchable, some PDF-only, some with a two-week publishing lag. That fragmentation is the actual barrier, not secrecy. Making the history readable in one place is what turns a score back into information.

About the author — Rick Jenkins is the founder of AI Revenue Forge. ARF builds vertical-specific AI virtual receptionists for service businesses in HVAC, dental, medspa, real estate, home health, credit repair, and pawn shops. Headquartered in Charlotte, NC. Part of Jenkins Worldwide Enterprises.