RiskyEats Methodology

Where the data comes from

Three layers, in order of how much weight they carry.

• Official state records. This is the backbone. Florida's Department of Business and Professional Regulation (DBPR), Division of Hotels and Restaurants, publishes every restaurant inspection, every emergency closure order and every license record in the state. We also read Sunbiz, the Florida Department of State's business registry, which shows whether the company behind a restaurant is still legally alive. DBPR records are refreshed twice a day. Sunbiz is refreshed quarterly.
• Open map and business-listing data. Foursquare OS Places (Apache 2.0) and Overture Maps (CDLA 2.0) help us work out what kind of food a place serves and match a state license number to the restaurant you would recognise by name. We do not use them to decide that a restaurant has closed. Their "permanently closed" flags were measured to be wrong often enough that we removed them from closure decisions entirely. A frozen, read-only snapshot of historical Google Places results is retained as one evidence source for the small group of licenses we investigated before the Google dependency was retired. We make no live Google API calls; new place-status checks are planned against RapidAPI Local Business Data.
• Community posts and news coverage. Closure and opening posts from public Florida food groups on Facebook (most often Foodies Who Review South Florida) and headlines from local outlets (Sun Sentinel, Miami Herald, South Florida Business Journal, Eater Miami and others). These never stand alone. A community or news signal has to line up with the state records before it earns a mention on a card.

How the page is organised

Cards are grouped by what the state record actually shows.

• Alerts. Something is happening right now: a DBPR emergency order that shut the kitchen, or a confirmed closure.
• Caution. Nothing has been ordered, but the record is worrying. This starts with Near Miss - an open restaurant whose Severity Score reached 300 or more in the last 60 days without an emergency order being issued. Near Miss is our own category. It is not a DBPR term, and it does not mean the state nearly closed the place.
• Trusted. Restaurants with a genuinely clean recent record. See Trusted medals for exactly what has to be true.
• New. Newly licensed restaurants and recent changes of owner.

Alongside those, several panels each answer one specific question: Systemic Violators (who has the worst two-year inspection record), Repeatedly Cited (who keeps getting written up but stays open), Final Orders (who the state has actually fined or disciplined), Chain Activity (what is happening across a brand rather than one address), and Closed / Delinquent and Ghost licenses (see those sections - the difference between them matters).

Why a restaurant can appear twice. Near Miss looks at the last 60 days. The repeat-pattern panels look at years. A restaurant can honestly qualify for both, so the Caution view says so and counts each restaurant only once in its total. One exception: a restaurant currently under an emergency order is pulled out of Near Miss, because the actual closure order is the more important fact.

The RiskyEats Score

Every card shows one number from 0 to 100, where 100 is the worst. Read it like a risk thermometer, not a grade out of 100.

The number is a 50/50 blend of two things: how bad the recent inspections were, and a lifetime public-health risk estimate (PHR) that smooths out restaurants with very few inspections so one bad night at a new place does not look like a decade of neglect. If the lifetime estimate has not been calculated for a license yet, the recent-inspection side is used on its own. The bar under the score fills in a straight line: a score of 92 fills 92% of the bar.

How severity is calculated

Not all violations are equal, and the state says so. DBPR ranks findings as high-priority (directly tied to someone getting sick - bad holding temperatures, raw meat over ready-to-eat food), intermediate (things that lead to that, such as no thermometer or a hand sink with no soap), and basic (housekeeping - a cracked floor tile, an unlabelled container).

We weight them the same way. Each high-priority finding counts 100, each intermediate counts 10, each basic counts 1, and visible pest evidence adds 50. A perfectly clean inspection scores 0. One high-priority citation scores 100.

That raw number is then converted onto the 0-100 scale above, and the conversion is deliberately gentle at the bottom. One high-priority citation lands at 75 - heavy, but well short of severe or emergency - because a single bad finding on a single day should not read as a crisis. Above that the curve flattens, so the genuinely worst restaurants spread out across 75-100 instead of all crowding at 100. For reference: raw 100 becomes 75, raw 115 becomes 80, raw 190 becomes 90, raw 256 becomes 93, raw 1340 becomes 99. The score can never exceed 100.

Letter grade

Some cards carry a letter. Florida does not grade restaurants, and this is not a state grade. It is our shorthand for the single inspection or pattern that put the card on the page.

A clean-record card gets an A. On a Near Miss card, one high-priority citation (or three intermediate ones) is a B, two high-priority is a C, three or four is a D, and five or more is an F. Cards that are on the page because of an emergency order or a long repeat pattern are shown as F, because what put them there is a state action or a sustained history rather than one inspection.

Trusted medals: Platinum, Gold, Silver, Bronze

These are the positive badges. They are earned on a restaurant's most recent routine inspection and the run of inspections behind it.

The first rule is a hard stop. Any high-priority violation in the last two years disqualifies a restaurant from every medal, no matter how tidy everything since has been.

A few things that keep these honest:

• Follow-up visits do not count. A call-back is an inspector returning to check that an earlier problem was fixed. Counting those would let a restaurant build a "flawless run" out of its own clean-up. Pre-opening license checks do not count either.
• Old evidence earns nothing. If the most recent qualifying inspection is more than two years old, we show no medal at all rather than a lower one. A stale record makes a restaurant unknown, not dirty.
• One visit is not a record. A clean restaurant with only one inspection on file is labelled Too new to grade.
• The Trusted section is stricter than the medal. It lists Platinum, Gold and Silver only, and only if the state has actually been in recently - within 365 days for a Level 1 restaurant, within 180 days for Levels 2 and 3. Bronze appears on a restaurant's own page but is not presented as a recommendation.

One honest caveat. The share of inspections recorded as flawless varies enormously by county - from about 54% in Okaloosa to about 8% in Miami-Dade - which almost certainly reflects how different inspectors write things up rather than how clean the kitchens are. A Platinum is genuinely harder to earn in Miami than in the Panhandle.

Clean run

A clean run counts how many inspections in a row a restaurant has had without a high-priority violation. It is a fact about the inspection record, not a certificate anyone awarded. A longer run means fewer recent serious findings on file.

Improving

This section is for restaurants that had real problems and fixed them. It compares two back-to-back periods of one year each.

We add up the high-priority and critical violations in the most recent year, and in the year before that. To appear here a restaurant must clear all of the following:

• At least two inspections in each of the two years, so there is something real to compare.
• The earlier year must have had violations to begin with.
• The recent year's total must be at most half of the earlier year's - a 50% improvement or better.
• The last two inspections must each be below the recent year's own average. This is the consistency check: one good visit after a run of bad ones is not an improvement, it is a good day.
• The license must currently be active.

Restaurants that are currently on the Systemic Violators list, or that carry a Megaviolator badge, are excluded. A restaurant can be improving and still be dangerous, and we would rather not hand out a compliment in that situation.

The percentage on the badge is how much the violation total fell. The count beside it shows the before-and-after numbers that produced it.

Riskiest in the last five years (current operator only)

This is the ranked list at the top of each metro page. It answers a blunt question: over the last five years, which restaurants in this area have collected the most state enforcement?

For each restaurant we look back 1,825 days (five years) and add up:

• every emergency closure episode (order filings dated within seven days of each other count as one closure), worth 10 points each;
• every emergency order filing, worth a further 3 points each;
• every administrative complaint, worth 2 points each;
• every dollar of fines, worth 1 point per $250.

Those weights mean closures, order filings and fines all count heavily: one closure episode counts about the same as $2,500 in fines, and, for the same number of filings, a restaurant that was shut down repeatedly outranks one whose filings all bunch into a single closure.

The important part is what we throw away. If the restaurant changed hands during those five years, everything before the ownership change is dropped. A new owner does not inherit the last owner's record on this list. Publishing somebody else's shutdowns under the current operator's name would be unfair and, on a card built to be shared, lasting.

Each metro shows roughly 20 to 25 restaurants. It is a summary list for the area, not a badge on individual cards.

Systemic Violators score (24 months)

This panel ranks restaurants by how bad their inspections have been over the last 24 months, worst first.

How it works, step by step:

1. Take every inspection of that restaurant in the last two years.
2. Score each one: high-priority findings times 100, plus critical or intermediate findings times 10.
3. Take the average, not the total. The question is what a typical visit looks like, so a restaurant is not punished simply for being inspected more often.
4. Discount that average for how long ago the most recent visit was: full weight if it was within 90 days, 70% within six months, 40% within a year, 20% within two years.
5. Scale it by how many visits there were, up to six. A restaurant with one inspection carries less weight than one with six, because one visit is weaker evidence.

Each card also shows how many inspections went into its score, so you can see whether a high number reflects a repeated pattern or a single bad day. The top 200 restaurants per metro are ranked. Cards flagged with an enforcement tier (see below) always sort above the rest.

Separately, the long record badge on some of these cards refers to an 11-year history of closures or complaint inspections. That badge is about the long haul; the score above is strictly about the last two years.

Systemic Violators badges

Three badges summarise what is wrong:

• High violations - high-priority findings are well above normal.
• Repeat offender - the same operator or license keeps producing serious findings across visits or across locations.
• Long record - the 11-year history contains enough closures or complaint inspections to count as a sustained pattern.

Recent evidence counts for more than old evidence in all three.

The warning triangle. A restaurant earns it when it shows up in two or more separate red-flag categories at once (emergency order, lapsed license, repeated violations, ghost record). At least one of the two must be an active problem - an emergency order or a repeated-violation record. A restaurant that is only lapsed and only quiet is classified as likely defunct instead, because that combination usually means the place is gone, not that it is dangerous.

Signature letters

Some cards carry a short letter code showing which categories fired together:

• E - emergency order or active enforcement
• L - lapsed or delinquent license paperwork
• B - a history of serious violations
• G - a ghost record, or over a year with no inspection

A combination such as EB or ELBG simply means all of those letters applied at once.

Elsewhere on a card you may also see C for an externally confirmed closure, D for DBPR disciplinary evidence, V for a high-violation inspection, and R for repeat-offender behaviour.

The enforcement tiers below are named in words, not letters, and are a different system.

Enforcement tiers

Four patterns are serious enough that we lift them above the normal ranking entirely. A restaurant matching one of these sorts to the top of Systemic Violators regardless of its score. (These thresholds were reviewed with our internal review process in May 2026 and have been adjusted since.)

• Inferred foodborne risk (top rank, 200). Three or more disciplinary Final Orders within two years, together with a recent emergency suspension. See the inspection tier section for why we describe this as inferred rather than official.
• Serial renamer (rank 195). Either two or more trading-name changes on the same license plus at least one emergency closure within three years, or three or more different trading names on the same license with no closure required. Changing the name over the door resets a restaurant's online reputation while leaving the state's record attached to the license underneath. Example: REBEL HOUSE became REBEL HOUSE / UNCLE PINKIE'S DELI and then ALLEYCAT, all on license 6020907 in Palm Beach.
• Pest habitual (rank 192). Two or more emergency closures involving rodent, roach or insect evidence within 24 months. The narrow window is the point: two pest closures two years apart is a different story from two in one year, and only the latter qualifies.
• Ghost-kitchen address (rank 185). Three or more different restaurant names licensed at the same address. Shared and delivery-only kitchens can rotate brands faster than enforcement can follow them.

If a restaurant qualifies for an enforcement tier and carries the warning triangle, the higher rank wins. A serial renamer that also has an emergency order plus a violation history keeps its rank of 195, because that is higher than the 70 the combination would otherwise earn.

Repeatedly Cited

These are restaurants that keep getting written up and keep trading. To appear here, a restaurant needs four or more administrative complaints in the last 12 months.

What an administrative complaint is. An inspector finds something serious enough to escalate, and the disposition on the report reads "administrative complaint recommended". The restaurant stays open. Nothing has been decided, nobody has been found liable, and no order has been issued. It is the state starting a formal process, not finishing one.

Why four. We measured what happens in the year after. Restaurants with one complaint are about 2.4 times more likely than average to face an emergency closure the following year; with two or three, about 4 times; with four or more, between 6.7 and 9.9 times. The pattern held across three separate years, so four is where the line is drawn.

What this does not mean. It is not a warning that a restaurant is about to be shut. The typical gap between a restaurant's fourth complaint and an actual emergency order is 587 days, and only about 7% of orders arrive within 90 days. This is a long-term risk signal. A restaurant can sit on this list for two years and never close.

Maximum Enforcement

This badge marks the small group of restaurants where the state has both shut the doors and started paperwork, more than once each. Specifically, in the last 24 months: two or more emergency closure orders and two or more administrative complaints.

It is the narrowest badge on the site - around 121 restaurants statewide - and most of them are open right now, because an administrative complaint is something a business trades through. It says what the state has done about a restaurant, which is different from the Megaviolator badge, which counts what a restaurant was cited for.

It does not claim enforcement is escalating. We checked: across this group, emergency orders in the recent 12 months versus the previous 12 split roughly evenly between rising, falling and flat. What repeats is the enforcement, and that is all the badge asserts.

Final Orders

A Final Order is a signed disciplinary decision filed with DBPR's Agency Clerk - typically a fine, sometimes licence conditions. It is the end of the process that an administrative complaint starts.

This panel lists one card per fined restaurant from roughly the last six months (183 days), newest first, up to 60 per metro. It exists because a restaurant can be fined by the state and still have no card anywhere else on the site: a fine on its own does not create an inspection failure. Seven fined restaurants in one 2026 disciplinary sweep were invisible here before this panel was added.

These cards deliberately carry no score bar and no trend arrow. The order itself is the finding; there is nothing to grade on top of it.

Megaviolator badge

A restaurant earns this when at least one of its inspections produced a total violation count in the worst 10% of all Florida restaurant inspections. The threshold is not a fixed number - it is recalculated from the real spread of every inspection in our 2016-2026 history. The label on the card ("Top 1% statewide", for example) says where that restaurant's worst inspection landed.

This counts all violations, not just the serious ones, so a restaurant can earn it through sheer volume of moderate problems. That is the point: it measures how wide the failure was, while the RiskyEats Score measures how severe it was.

About the date shown. The date on the card is the most recent of that restaurant's top-10% inspections. It is not necessarily the day of the single highest-count inspection. If a restaurant had two qualifying inspections, the card shows the newer date and the higher count, and those can come from different visits.

Emergency orders and overdue callbacks

An emergency order (DBPR calls these EOR and EOS records) means a state inspector ordered the kitchen shut. These are often temporary. The restaurant fixes the problem, an inspector returns, and a callback inspection posted to the record reopens it.

A callback is overdue when the state has not posted that follow-up when it would normally be expected. We mark it a warning after more than 14 days, and severe once 30 days have passed with nothing recorded. An overdue callback does not prove a restaurant is still shut - paperwork lags - but a long silence after a closure order is worth knowing about.

Inspection tier (how often the state visits)

DBPR schedules inspections by risk: broadly, Level 1 restaurants get about one routine visit a year, Level 2 about two, Level 3 about three, and Level 4 about four. This matters when you read a violation count. Five closures at a restaurant the state visits once a year is a very different story from five at one it visits four times a year.

This number is our estimate, not a published state assignment. DBPR does not publish a per-restaurant risk level in the data we can access, so we work it out from two public signals: how often that restaurant has actually had routine inspections (we count first-visit routine food inspections and divide by the years it has been licensed), and a rule-based floor of Level 3 when a restaurant has three or more disciplinary Final Orders in two years. Where the observed schedule is higher than the rule would predict, the observed schedule wins - the state's own scheduling reflects things we cannot see.

Level 4 deserves its own caveat. The state's rule makes Level 4 depend on a confirmed foodborne illness reported by the Florida Department of Health within the previous year, and Florida law (statutes 381.0031 and 119.071(5)(g)) keeps the source of public-health investigations out of public records. So no per-restaurant outbreak attribution is published, and no Level 4 exists anywhere in the licence data we receive - the highest value that appears is Level 3. Where we describe a restaurant as carrying Level 4 risk, that is explicitly our inference from three or more Final Orders plus a recent emergency suspension, not a state designation.

License status badges

These four badges describe very different things and are easy to confuse, so here they are side by side.

• Closed - we have real evidence the business has shut: an owner's own statement, news coverage, or verified on-the-ground reports.
• Delinquent license - DBPR has marked the licence inactive or delinquent (status code 45 or 46). This is a fact about the paperwork, not proof the restaurant is gone. In Florida, a licence is attached to the premises and is usually taken over by the next operator, so an inactive licence often just means nobody has renewed it yet. We show this in a neutral colour and never call it a closure.
• Suspected silent closure - suspected, not confirmed. The licence is still active, but there has been no inspection for over a year (or none on record at all), and there is some evidence the business is gone, from the retained legacy places snapshot, open map data or verified community reports.
• Ghost - an active licence with no inspection history at all and no sign either way of whether anyone is operating. A question, not an answer.

Uninspected over 1 year

This badge means the opposite of the ones above: the restaurant appears to be open and trading, and the state has not inspected it in over a year. It is about a gap in regulatory coverage, not about the restaurant being in trouble.

For this badge specifically, a restaurant must have an active DBPR licence, no inspection for more than 365 days, not be a stadium or arena concession stand, and show positive proof of life: either the retained legacy places snapshot listing it as operational with a strong address match (0.8 or better), or a Google or Yelp review within the last 90 days. A live Sunbiz corporate registration does not count as proof of life here - a company can stay registered long after the kitchen stops.

This is an annotation on cards that are already on the page for some other reason. It does not produce a browsable list of its own. Restaurants the retained snapshot reports as closed or cannot find are routed to the closure surfaces as suspected closures instead, and ones we have never been able to verify go into a manual check queue.

This is not the same rule as the "Regulatory blind spot" group inside Ghost licenses. The two share a name and use different evidence. See the next section.

Ghost licenses

A ghost licence is an active state licence that is not behaving like a working restaurant. These are follow-up cues for us, not claims that anything has closed. The panel holds two distinct groups, and each card says which one it is in.

• Paper records. An active DBPR licence with no inspection history at all, at least 180 days old, and no sign of life or death from Sunbiz or the retained legacy places snapshot. Somebody holds a restaurant licence that no inspector has ever visited.
• Regulatory blind spots. An active DBPR licence where the company behind it is also active in Sunbiz, the last inspection was more than 365 days ago, and the licence is at least 180 days old, with no evidence the business is dead. Here the retained legacy snapshot listing the restaurant as operational is allowed and in fact strengthens the case - it confirms the gap in inspection coverage is real.

Note the difference from the Uninspected over 1 year badge above: that badge requires legacy-places or Yelp proof of life and explicitly rejects Sunbiz, while this group is built on Sunbiz being active. They can flag different restaurants, and that is intentional.

If a licence would qualify for both groups it is listed only under Paper records.

Zombie licences are a stricter subset: the DBPR licence is active but the Florida company behind it has been administratively dissolved or revoked in Sunbiz. That usually means the operating company wound down while the restaurant licence stayed open. When any are present, they are called out above the cards.

Closure detection

Deciding a restaurant has permanently closed is the most damaging thing this site can get wrong, so the bar is high and the language is careful.

A quiet restaurant is not a closed restaurant. No recent inspection, on its own, is never enough. To be labelled confirmed closed, a restaurant needs external verification of closed status, an owner's own statement, or verified news or community evidence. Anything weaker stays labelled as suspected, and says so on the card.

An inactive or delinquent DBPR licence (status 45 or 46) is reported separately as a fact about the licence. It is one piece of evidence, never a conclusion. In practice most Florida licences outlive the business that held them - they are tied to the premises and get picked up by the next operator - so treating "licence inactive" as "restaurant closed" would be wrong far more often than it would be right.

Closure dates are approximate by nature. The last inspection is always earlier than the real closing day, and a licence expiry is usually much later. We do not claim to know the exact day a restaurant closed.

Evidence count

Some closure cards show a small count such as 3/4. It is a legacy summary of how many independent kinds of evidence were found: inspection silence, an inactive or cancelled DBPR licence, a dissolved company in Sunbiz, and confirmed outside evidence (owner statement, news or verified reports).

Two caveats. A higher number is not a health grade and not a prediction - it just means more independent sources pointed the same way. And this count does not make the decision. Whether a card is published as a confirmed closure is settled by the closure classification described above, which reads a full event record for each licence. The count is a summary shown alongside it. In particular, open map data such as Foursquare is never counted as closure confirmation.

Recency bucket

Closure cards are split at 90 days. last_90_days means the evidence is recent enough to read as current news. historical means it is an older record we are keeping on file.

Requires revalidation

A staleness flag. If the newest inspection or verification touch on a closure card is more than 16 months old, the card is marked for a manual re-check before anyone should treat it as a current statement about the restaurant.

Confidence label

A short plain-English note explaining why a piece of evidence was strong enough to publish. Examples: "Operator-attested closure", "DBPR-confirmed plus multi-source corroborated", "Google-confirmed opening", "Chain rollup signatures". It tells you what kind of proof sits behind the card.

Operator overrides

Restaurant owners and operators can tell us something the records do not show, and we can act on it. An override is an editorial decision recorded by hand, always with a documented reason. There are three kinds:

• Closure attested. The operator confirms the restaurant has permanently closed. We publish a closure card, labelled as operator-attested.
• Do not publish. The licence is removed from the site entirely, everywhere. This is what we use when something in the automated pipeline is wrong about a real, operating business - for example a community post naming the wrong restaurant.
• Operating confirmed. The operator confirms the restaurant is open. It is removed from every closure-shaped section - closures, ghost licences, lost licences and emergency orders. Its inspection and violation record stays exactly as it is. An owner can correct a false closure; nobody can remove a real inspection finding.

New and reopened restaurants

The New section covers roughly the last 180 days and separates three genuinely different situations, because "new" can mean very different things:

• New business signal - a new-business or licensing signal at an address; this does not by itself confirm that a storefront opened or is operating. A brand-new state licence at an address that has never held a restaurant licence is evidence of a new licensing event, not proof that the doors are open. A new place in a new space still needs an independent operating signal before we call it confirmed open.
• New owner, new licence - a brand-new licence at an address that used to hold a different restaurant licence. Something closed here and something else opened.
• New owner, same licence - the state approved a change-of-owner on an existing licence. The paperwork survives; the people running it changed. Note that the inspection history stays attached to that licence, so a brand-new owner may inherit a card with an older record on it.

Restaurants that are simply an established operator opening another branch, or reopening after an emergency closure, are classified separately and kept out of the New list - they are real events, but they are not a new restaurant. Administrative re-licensing of a long-standing business is also excluded; if the first inspection on record is old, the business is not new no matter how recent the paperwork is.

Chain kind

Where we can tell, each card says what kind of business it belongs to. Single means no chain pattern was found. Regional means the operator appears to run several Florida locations. Franchisee-owned means a local franchisee controls some of a bigger brand's Florida locations. National or national-owned means the location belongs to a large brand-owned footprint.

Chain Activity

Chains distort counts. One franchisee's company restructuring can look like a dozen separate closures, and a national brand's licence churn can crowd out independent restaurants on a page. So chain activity gets its own panel, organised in three layers: brand, then the companies that hold the licences under that brand, then the individual locations.

Each brand shows a summary of what is happening across its Florida locations: how many operating companies have been dissolved in Sunbiz, how many licences are lapsed, how many locations have recent high-violation inspections, repeat-offender records or a long history. A small chart shows closure events by quarter across an 11-year footprint.

The routing rule is deliberate. Closure-type events for chain locations move here rather than crowding the main closure lists. Red-flag events stay on the main page - if a chain location is under an emergency order or has a serious violation record, that is exactly the kind of thing a reader is looking for, and hiding it in a brand rollup would be the wrong call.

Chain closure exception

Sometimes one specific chain location really has closed, and the normal chain routing would bury it. When the evidence is about that address rather than the brand, we flag the card as an exception so it stays visible on the closure list even though the wider brand is still trading.

Name history and reputation resets

A Florida restaurant licence stays with the premises when the name over the door changes. The inspections, complaints and disciplinary actions all remain attached to that licence number - but the name a customer sees no longer reveals the earlier identity. Florida does not require a name change to be disclosed to customers, so a rebrand can wipe a restaurant's public reputation while the official record underneath stays exactly the same.

When a card shows a Renamed badge, that licence has traded under more than one name in the full inspection history we hold - not a fixed window. Previous names are listed with the years they were in use.

This is a transparency note, not an accusation. Most name changes are ordinary: a sale, a refit, a franchise rebrand. It is here so that if you see a name you do not recognise, you can weigh the record that comes with it.

Where a licence number is ambiguous - Florida issues different kinds of permits from the same numeric range, and a few hundred numbers match more than one food-service permit - we show no name history rather than risk attaching one restaurant's past to another. Showing less is the safe failure here.

A real example: licence 6020907 in Boca Raton traded as REBEL HOUSE from 2016 to 2021, briefly as REBEL HOUSE / UNCLE PINKIE'S DELI in 2022, and as ALLEYCAT from 2023. Today's ALLEYCAT card carries the whole ledger, including the April 2026 emergency order that closed the kitchen for a day.

Local 10 / Dirty Dining coverage

This badge marks restaurants featured in WPLG Local 10's Dirty Dining segment, South Florida's main broadcast investigation series on restaurant hygiene. Coverage may follow an emergency closure, a bad inspection, or later follow-up reporting. Where we have the article link, the badge goes straight to it. The date shown is when the article was published, which is not the same as when the inspection happened.

Location and context badges

A few badges describe the restaurant rather than its record. None of them affects any score.

• Waterfront - the restaurant sits close to water. The badge names the kind of water (canal, bay, ocean) and how close.
• High tourism - the restaurant is near a major tourist anchor such as a theme park, stadium or cruise terminal. Useful context: these places serve a lot of one-time visitors.
• Kosher and Halal - the restaurant appears in a certification directory. We distinguish certified (listed by a real certifying body, which we name and link) from self-declared (the registered business name says kosher or halal but we found no certification record). A restaurant can hold both certifications.
• Multi-tenant - several active restaurant licences share this exact address. Typically a food hall, a shared or delivery-only kitchen, or a mall food court. The badge counts the licences and, where DBPR records it, names the neighbouring registered businesses. It never guesses a neighbour's consumer-facing brand name.

Trend labels

Some cards are labelled improving, stable, declining or volatile. This compares a restaurant's three most recent inspections against the three before those. It is a count-based comparison, not a calendar one - whenever those six visits happened, subject to a five-year maximum span, because comparing today against a decade ago says nothing useful.

• Improving - the average number of violations fell by at least 3, and high-priority findings did not rise.
• Declining - the average rose by at least 3, and high-priority findings rose by at least 1. A new serious finding is a different kind of fact from three more minor ones, so it is required here.
• Volatile - the direction reversed three or more times across those six inspections. The restaurant is inconsistent rather than heading one way.
• Stable - nothing moved far enough to say. A change of one or two violations is normal variation between inspectors and does not earn a label.

A restaurant needs six qualifying inspections to get a label at all. Some cards also show 60-day figures; those are reported for reference and play no part in the trend label.

This is a description of the past, not a forecast.

How we avoid making things up

Every fact on a card - the restaurant name, the counts, the dates, the violation types, the closure and licence claims - comes from records already loaded into the build. Nothing is invented to fill a gap. When a piece of evidence is missing, the sentence or the section is simply left out. Silence is always preferred to a guess.

To be precise about how the wording is produced: some of the descriptive prose on cards is generated by a language model. It is constrained to the verified facts for that specific restaurant, explicitly instructed not to introduce anything else, and falls back to fixed templates when generation is unavailable or the facts are thin. So the guarantee we make is about grounding and omission - every claim traces to a record, and a missing record produces no claim - rather than a promise that every sentence was assembled mechanically.

If you believe something on this site is wrong about your restaurant, use the Report a correction action with the page or record URL, venue or license, disputed claim, and public source. Corrections are reviewed, documented, and applied by hand, as described in operator overrides; a message is a review request, not an automatic change.