How this works

All property data is public record, published by the local authorities that hold it — assessors, licensing bodies and building departments. The datasets differ by market, because every city publishes different things under different names.

What we hold, market by market:

• Washington, DC — sources, coverage, and what the data cannot tell you
• Arlington County, VA — sources, coverage, and what the data cannot tell you
• Philadelphia, PA — sources, coverage, and what the data cannot tell you
• New York City, NY — sources, coverage, and what the data cannot tell you

Every record we show carries the source it came from and the date we retrieved it. We never assert that a person or company owns a property; we say what the records indicate, and link to them.

A landlord profile shows a Public Record Score and a Tenant Score side by side. They are never combined into one number, because they answer different questions and one is opinion.

The Public Record Score is computed only from government records. It cannot be moved by reviews, and it exists for every landlord in the database whether or not anyone has reviewed them.

Violations are weighted by severity — a hazardous condition counts for more than a cosmetic one — then divided by the number of units in the portfolio. Raw counts would simply rank landlords by size, which would be a unit census wearing a score’s clothes.

Small portfolios are pulled toward the city average in proportion to how little evidence there is. A duplex with one violation is one inspection away from zero, and should not be judged as loudly as a 500-unit portfolio with the same rate.

The result is a percentile, where higher is better, and it is computed only among landlords who have records on file. Landlords with identical records receive identical percentiles.

Most landlords aren’t ranked at all. Around 92% of the owners we track have no housing-code violations recorded against them, so there is no distribution to place them in. Their page says “no violations on record” and shows no percentile. Ranking them anyway would mean inventing a position: we tried it both ways and each produced a claim we could not defend — a small owner with a spotless record reading as below average in one arrangement, and the owner with the lowest violation rate in a market reading as near-worst in the other. The honest output is the fact itself.

This also means a percentile here is a narrower statement than it looks. It compares a landlord against others who also have records — not against every owner in the city.

Buying and selling is shown, not scored. A landlord’s page lists how many properties they bought and when, because it helps explain a short record. It deliberately does not feed the score: we can tell you a property changed hands, but not what that means about a landlord. Frequent buying can be someone who neglects buildings, or a growing business, or a family settling an estate — and a number that quietly treated all three the same would be a judgement we cannot support.

A sale resets the record. We count only violations opened while the current owner held the building. Someone who bought a troubled property last year inherits the building, not its record — judging them on their predecessor’s citations would be straightforwardly unfair, and it is the single easiest way for a scoring system like this one to be wrong about a real person. Where a jurisdiction has no recorded transfer for a property, we cannot draw the line, so the records still count and the landlord’s page says so.

This cuts both ways, and the profile page states the window for that reason: a landlord with few records may have a good record, or may simply have bought recently. A near-empty record is not an endorsement.

We withhold the score rather than guess when there are too few comparable landlords, when we hold no unit count to divide by, or when so few records have been ingested that a percentile would measure our own coverage rather than anyone’s conduct.

Most jurisdictions publish violations without an abatement date, so the data cannot say whether a violation was ever fixed. Where that is true — and it is true in every market we currently cover — we report violations as on record, never as open or outstanding. Describing a 2018 citation as unresolved today would be a statement of fact we cannot support, and we weight those records conservatively in the score for the same reason.

Where a source does track resolution, we say “confirmed still open” and mean it.

Tenants rate eight specific areas from 1 to 5. There is no separate overall star — the headline number is computed from those eight, so a reviewer’s detail is what counts rather than their mood.

Reviews are weighted, never required to prove anything. Anyone with a confirmed email address can publish, and no review is held back for lack of paperwork. What proof changes is how much a review counts:

• Verified — a moderator has seen a document showing the person lived at that address. Counts triple. We delete the document as soon as the decision is made.
• Confirmed account — a confirmed email address on an established account. Counts once. This is most reviews.
• New account — brand new, or the address is not yet confirmed. Counts a third as much, until the account has been around a day.

That last rung is about brigading, not suspicion of any individual: a burst of accounts created to move one landlord’s rating should cost something, and a real tenant loses nothing by waiting a day.

A review’s weight also halves roughly every two years after the tenancy ends, because a complaint about 2016 management says little about today’s. An ongoing tenancy never decays.

A landlord with one angry review does not outrank one with forty calm ones: thin evidence is pulled toward the average of other reviewed landlords until there is enough of it to stand alone. We show the distribution of ratings alongside the average, because a 3.0 from all-3s means something completely different from a 3.0 from half-1s and half-5s.

Below one effective review we show no tenant score at all, rather than showing you the average of other landlords and calling it this one’s.

Voting a review helpful changes the order reviews appear in and nothing else. This is deliberate and structural — it is how brigading a landlord’s rating is made impossible rather than merely discouraged. Ordering uses a statistical lower bound, so a review with one helpful vote does not outrank one with 180 of 200.

Landlords commonly hold buildings through many single-purpose companies. We link them using shared mailing addresses, shared registered agents and name similarity, and we never silently merge two owners. Every inferred link carries a confidence level and the specific evidence that produced it, and uncertain links affecting significant numbers of units are queued for a person to decide.

Being wrong here attributes one landlord’s record to another, so the system is tuned to refuse rather than to guess. In our most recent run it rejected roughly 97% of candidate matches.

Owner names come from the public record and are shown as recorded, including where the owner is an individual rather than a company. Many rentals are owned by people rather than corporations, and operating a rental is a business activity whichever it is.

Two limits on that. We never publish an owner’s mailing address — the assessor’s record is frequently a home address, and naming an owner is not the same as saying where they live. And owner-occupied homes are excluded entirely: in DC that is about 61% of parcels, and someone living in their own home is not a landlord.

If a record is wrong about you — you sold the property, you live there, or the owner shown is not you — request a correction.

Reviews are the personal experiences and opinions of individual tenants. They are not statements of fact by RateMyLandlord, and we present them as what a tenant reported. Our review composer actively declines to publish unqualified criminal allegations and coaches people toward describing what happened instead.

Summaries of reviews are counted, never paraphrased. The tags on a profile are options a tenant chose from a fixed list. We do not use AI to characterise a landlord.

Information on this site may not be used to make tenant screening, credit, insurance or employment decisions. We hold no tenant-identifying eviction data of any kind, by design — the eviction data published in the markets we cover is aggregate — ward- or district-level counts — and cannot be attributed to a landlord.

• Violation records carry no abatement date, so we cannot tell you whether something was fixed.
• We hold unit counts for only some properties. Where we do not, the Public Record Score is withheld rather than estimated.
• Ownership records lag reality. A building sold last month may still show its previous owner.
• We have a recorded transfer for about three quarters of the properties we track. For the rest we cannot tell when the current owner bought, so violations that may belong to a predecessor are still counted — and the profile page says where that applies.
• Where several lots share one street address, we divide the building’s unit count between them. The portfolio total is right for an owner who holds the whole building and an estimate for one who holds part of it.
• Company-to-company links are inferences, not findings. Some are wrong, which is why each shows its evidence.
• Our lease review covers a limited set of legal topics. A clause we do not flag is a clause we have no rule for — not a clause we have approved.

Tell us and we will investigate. Request a correction. We correct records on evidence, and we log what we changed and when.