Texas is the clearest case: our catalogue records that its appraisal roll carries no sale price and no sale date. Across the whole catalogue, 24 of 2,021 connected sources disclose a recorded sale price, spread over 19 states, as of 2026-09-25, the identical 24 sources and 19 states we published on 2026-09-11, while connected sources themselves grew roughly sevenfold. Where price is absent, derive value from the building's own income, never from assessed value.
What the catalogue says, and as of when
Every catalogue figure comes from our open dataCatalog query, run 2026-09-11 (response stamped generatedAt 2026-09-11T12:21:17Z); internal coverage and product figures were measured 2026-09-14. A disclosure count is a snapshot of one day's connections, not a standing property of American property records.
On that date, 2,021 of 2,051 listed sources were connected, and connected rows carried 49 jurisdiction codes (48 states and the District of Columbia). The catalogue's summary field statesCovered returns 45 rather than 49, because it counts the states our source registry can actually route a lookup to: Alabama, Iowa, Kentucky and South Carolina have a connected source but no route yet; both numbers are published here because picking the larger one is how a coverage claim starts to drift. Two states have no connected source at all: Kansas and South Dakota; on 2026-09-11 there were nine.
The number that matters is the smallest in the catalogue, and it is the one that did not move: 24 connected sources disclose a recorded sale price, across 19 states: Arizona, Colorado, Connecticut, DC, Illinois, Massachusetts, Maryland, Maine, Michigan, Minnesota, New Hampshire, New Jersey, Nevada, Ohio, Oklahoma, Oregon, Pennsylvania, Tennessee and Virginia. Recorded price is the thinnest layer in US property data, and two weeks of connecting sources did not thicken it by one. Sale date is thicker but still thin: 34 sources across 24 states. The rest of the same matrix: parcel id 73, address 71, land use 67, area 64, assessed value 59, building area 25, units 20, zoning 18. Assessed value comes from more than twice as many sources as price, which is why it gets misused.
Texas, in the catalogue's own words
The catalogue carries the reason on the Texas rows themselves: "Texas is a non-disclosure state: the appraisal roll carries no sale price and no sale date." That is the appraisal roll. The recorded deed is a different record: our coverage page (measured 2026-09-14) records that Texas deeds give a date and an instrument but never the price. So you can establish that a transfer happened and roughly when, never the amount. A model that needs the event can be built; one that needs the amount cannot be sourced into existence. Texas is 8.4% of our US demand (12,542 runs measured since 01.01.2026), and no vendor we use fills that gap.
Why "what each source discloses" is answerable for only 73 of 2,021
The matrix itself is partial, and it has become more partial, not less. 73 of the 2,021 connected sources carry a field-level disclosure matrix at all. The other 1,948 carry none; for those, "does it give me a sale price?" has no catalogue answer, only a per-source test. On 2026-09-11 the ratio was 66 of 272; the sources we added since are overwhelmingly the kind that publishes no matrix.
That follows from what those sources are. By kind, the 2,021 connected split into 1,910 lookups, 73 enumerations and 38 signals (as of 2026-09-25). A lookup answers one property at a time, on an address or a parcel id you bring it. An enumeration can list every property in an area matching a criterion (what a screen needs) and the enumerations are what carry row counts and disclosure matrices. A lookup-only state is one where you can check a subject but cannot sweep a market.
What non-disclosure removes from a model
It removes the observed right-hand side: the sale comparison approach at its root, the empirical cap rate (price is the denominator), and any price-derived market ranking. What survives is the income approach and the debt tests, which run on rents, expenses and loan terms.
The failure mode is not an empty cell. It is a substitute that looks like a price. Our own 416-unit Tampa multifamily workbook defaulted the purchase price to the assessed value of $35.21M, because that number was already sourced and already in the spreadsheet. The arithmetic downstream was correct, and it returned a 53% levered IRR, a 6.36x equity multiple and a 2.98x DSCR: the error announcing itself. The model was reporting the gap between an administrative tax basis and an income basis as if it were investment return. Priced instead off its own Year-1 NOI at a stated 5.75% going-in cap rate, the same workbook returns 8.66% (8.7% on a second, independently written implementation) and a 1.47x equity multiple, recalculated to the cent by two engines.
What a modeller should do when recorded price is absent
- Keep assessed value out of the price cell, permanently. Demote it to a reference line the sources tab labels never-a-price.
- Derive the price from the asset's own income at a stated going-in cap rate, labelled an assumption, not an observation. An honest assumption beats a sourced number from the wrong basis.
- Check both sides of every ratio for basis. That workbook's exit tab lost its peer-median comparison: the peer median was a median of assessed values, the exit price a market price. Its headline −48.3% "discount to peer median" is a ratio of two assessed numbers, about a tax figure, not the market.
- Treat date and price as two separate availabilities. With 34 sources disclosing sale date against 24 disclosing price, transfer timing and ownership turnover are often reachable where the amount is not. Build the signals that need the event, and do not let them imply it.
- Lean on the loan tests, which need no recorded price. Run both every time and state which binds: DSCR binding means the cash flow, not the collateral value, sets the loan: a hard constraint from the income side alone.
- Classify every cell SOURCED, ASSUMED or COMPUTED, and publish that. In a non-disclosure state most of the price side reads ASSUMED. That is the correct output, and the one a lender can argue with.
- Widen to a range rather than holding a point. Below three comparable sales our rule turns a point value into a range, and where sources disagree confidence is clamped to the weakest.
- Automate the checks that fire in the flattering direction. Four earn their place: going-in yield outside roughly 3.5%–9%, an exit cap tighter than the going-in cap, a DSCR above 2x, and an operating expense ratio under 40% for Florida product. Every one fired on the first draft above. A model that is wrong in the flattering direction does not look wrong; it looks like a good deal.
- Do not rank a market on a field most of it lacks. One live shortlist of ours ranks 25 Tampa buildings by "underpriced %" where only three have any recorded sale; the rest is computed off assessed value: a tax-figure ranking wearing a market label.
- Report missing as missing. If a figure cannot be sourced we report it missing rather than filling it with a plausible number. And the result is an opinion of value: never an appraisal, never a valuation. We hold no appraisal licence.
What we will send you instead
We will send the full source list, including which sources carry a recorded sale price and which do not, before you sign anything.