Methodology

We grade the operator, with buildings as children. Reputation aggregates up the graph (building → owning LLC → operator → parent), so it survives building rebrands and aggregates across a portfolio. Google rates a building name; nobody rates the operator across its Austin properties. Every edge in that graph is published only after it is verified against primary records (Travis Central Appraisal District, Texas SOS, the operator's own site).

We ingest seven Austin open datasets weekly. One, code complaint cases, is the primary signal. Three more belong to the city's Repeat Offender Program, an escalation flag layered on top of the primary signal, for the small number of buildings serious enough to cross that program's threshold, which most West Campus towers never do. The remaining three corroborate the pattern from independent city processes: issued permits and certificates of occupancy track construction and delivery activity that has nothing to do with tenant complaints, and 311 service requests are a second, independent channel through which the same conditions get reported.

Complaint-based enforcement systematically under-reports, because tenants fear retaliation from a landlord who controls their lease renewal. So we score asymmetrically, and the rule is stated here in full:

Everything on this site belongs to one of three registers, and the register is always visible in the type and color, not just in a label. A reader who never reads a word of the surrounding text still learns which kind of claim they're looking at.

Facts are public records reproduced accurately, with a link to the original: a complaint count, a permit date, a court filing. Anything we derive (a score, a classification, a comparison) is an analysis, marked as one, with the computation behind it shown. We never assert a legal conclusion; where a pattern resembles a regulated category, we say "pattern consistent with…" and link the rule itself and the FTC's reporting channel, reportfraud.ftc.gov.

Every building's case file carries one label for its city complaint record. This is the rule that produces it, stated in full, so you can check any label against the same city rows we link on that building's page. Wherever a label appears, the count it came from, the 36-month window, and the links to the city's own case records appear with it.

Those thresholds are 2, 6, and 12 filings a year across the three-year window. One escalation applies on top: a building registered in the city's Repeat Offender Program moves up one label. No building in the study area is registered today, so that escalation currently changes nothing.

A row the city closed as a duplicate counts as half a filing. Every other row counts as one. This is the only number in the rule we chose rather than read off the record, so here is the reasoning in full.

35% of the rows in the window carry Austin's own "Closed/Duplicate Case" disposition, and the share runs from 0% at one building to 58% at another. Raw rows are therefore not comparable between buildings. Both of the obvious fixes are wrong in a direction we can predict. Counting every duplicate in full overstates, because a single incident can draw a dozen filings. Discarding duplicates understates, because a duplicate is frequently a second household reporting a problem that nobody fixed, which is evidence of how far a problem reached. The record does not tell us which case we are looking at, so the rule takes the midpoint between the two and publishes both bounds beside it. Every case file shows the filed count and the duplicate count, so a reader who disagrees with the weight can work out either bound from the page.

• A case is a complaint filed, not a violation found. Some close with no violation found.
• The count is not adjusted for building size. Verified bed counts don't exist yet for most of the index, so a 900-bed tower and a 200-bed building are compared on the same scale, and larger buildings will tend to sit higher for that reason alone.
• A building we can't join to a city parcel carries no label, rather than a favorable one.

Every dataset in §2 is pulled weekly, and the snapshot (not just the aggregate shown here, but the underlying rows) is committed to a public code repository. Every page on this site states the date its data was captured. Every snapshot is kept; nothing is edited out of the record.