Every field traces to a public record

Sources, in order of authority

1. County recorder — in New York City, ACRIS: every mortgage, consolidation, deed, satisfaction, assignment, modification and assumption recorded since 2004. In Texas, each county clerk's real-property records since 2015: who borrowed from whom, on which lot. The Texas index records no amount and no maturity; those come from the document.
2. Assessor — New York's property file and each Texas county's appraisal district: the building, its use, its assessed value and its owner of record. What a district withholds, we do without: Tarrant County publishes no values, so no loan-to-value is shown there. We do not substitute an estimate for a number the county does not publish.
3. The recorded document itself — the scanned pages, read for the stated principal, maturity, term, rate and who signed.
4. Securitization and agency disclosure — SEC filings for CMBS, HUD's insured-multifamily file and Freddie Mac offering circulars. A disclosed loan is matched to the mortgage the county recorded, and the disclosure's maturity, rate and servicing status sit on that loan beside its parties and parcel; it is listed on its own only where the county has recorded nothing for it. A later disclosure showing the loan refinanced retires it, with the disclosure cited.
5. Registers — New York's building registrations (HPD) and permit filings (DOB), and the Texas Comptroller's franchise-tax records: the people behind an LLC and, in Texas, the line of business it registered.

Trade press is never a source of record.

Provenance tiers

A modelled (M) maturity never renders as a fact. In the app it carries an ≈ marker; a stated (X) maturity is bold and cites the document it was read from.

Homes are not commercial loans

The assessor file (PLUTO) has no record for condominium unit lots, so a dollar floor plus a land-use check lets a $4 million mortgage on one apartment through as a commercial loan — 35% of the loans above our floor were homes before we fixed this. We classify from the recorder instead: every ACRIS legal carries a property-type code, and a loan whose parcels are all residential units (single condo or co-op unit, parking space, storage room) is excluded — unless it blankets three or more of them, which is a sponsor's inventory loan or an investor's portfolio and stays. Apartment buildings, five-plus family houses, mixed-use and commercial condos are untouched. Excluded loans remain in the database with their provenance; they are simply not searchable.

One-to-four family houses are handled differently since September 2026. A $1 million-plus loan on a house in the outer boroughs is almost always investor debt, so those loans are in the product under the property type Residential 1–4 Units, off by default: an "all types" search returns commercial collateral only and says above the results how many house loans the same filters would add, picking the type adds them, and an address or keyword that names a house finds it whether or not the type is picked. Because they are 30-year residential paper, we print no modelled maturity on them — only a stated maturity, or the first rate-change date read from the recorded Fixed/Adjustable Rate Rider, or nothing.

Co-ops and rent regulation

Both are New York filters, and neither is a property type. A cooperative borrows as a corporation against the whole building, so its loan is an underlying mortgage rather than a rental building's debt. We flag it from the Department of Finance's building class, which the assessor file carries for every co-op lot, and from the borrower's name where that file has no row — a co-op files as "… Owners Corp" or "… Tenants Corp". The word "co-op" alone is not used; it names streets in Co-op City. The building class also narrows the search for rent regulation — the walk-up and elevator rental classes hold 95% of registered stabilized units — but more than a quarter of the buildings in them register none, so it is a screen, never the flag.

No city roll records rent regulation. An owner who registers stabilized units with the state pays a per-unit fee on the Department of Finance tax bill, and that count, read from the latest statement of account, is what a row shows as registered: the owner's own figure, which misses owners who never registered and lags the bill by about a year. Where nothing is registered, a building completed before 1974 with six or more apartments, in a residential building class, is marked likely stabilized — the statutory presumption, shown as an estimate because many such buildings left the system before 2019. A co-op is never presumed; its unsold units, where registered, show in the count. A building marked neither is not market-rate: condominium lots and buildings with no assessor record cannot be tested. Benefit programmes that bind a building to stabilization for a term (421-a, J-51) are not yet read.

Sizing the loan correctly

New York loans are usually recorded as a set — a new “gap” mortgage plus a consolidation agreement (a CEMA) that folds it into the old debt, sometimes with a deed. Reading one document alone double-counts principal or reports the wrong amount. We read the whole set: the consolidation carries the loan, split notes add up, and the notes it absorbed are marked so they never appear as separate loans. A document re-recorded to fix a defect is the same loan: the latest recording counts.

Who borrowed and who lent is read from the recorded parties. A mortgage names the roles; a consolidation or modification agreement lists its parties with no role, and the lender comes first on about one in twelve. There we assign the roles from the names — a bank, thrift, credit union or insurer is the lender — and say so on the field. Clerks also type the parties in any order, and a guarantor or signing member often comes first: the borrower is the first company among them, never the person beside it.

Texas: the amount lives inside the document

A Texas clerk's index has no loan amount, so every screened deed of trust is read by a vision model for the stated principal and maturity (tier X, with the page cited). Two honesty rules follow from how Texas paper is written. First, some instruments secure a large facility “but only to the extent of” a smaller local amount — a guidance clause. When one is stated we show the local amount and keep the facility total on the record. Second, national credit facilities record their full amount in every county where collateral sits, with no local cap; those documents say what they say, and we report them as recorded rather than invent an allocation.

Texas: two of the four DFW districts publish no building class

The state class every Texas appraisal district files with the Comptroller has one code for commercial real property, F1, and one for industrial, F2, and the Comptroller's guide reserves F2 for a warehouse that serves a manufacturing operation. A distribution warehouse, a clinic and a strip center are all F1. Dallas CAD publishes a building class in words and Collin CAD a use code, and a loan takes its type from those. Denton CAD and Tarrant Appraisal District publish nothing finer, so for their properties we estimate a type from what they do publish: the district's own description of the account, the businesses on record at the address, Denton's building features, and, only for a self-storage or mobile-home operator, the owner's name. The estimate is stored at the modelled tier, marked ≈ wherever it appears, never written over a district's own classification, and withdrawn on the next run if nothing still supports it. Checked against what is at the address, an estimate is right about four times in five. Denton is nearly all typed; more than half of Tarrant's commercial loans stay “commercial”, because its district publishes only the state class and those parcels have no occupant on file. We do not quote Denton or Tarrant as coverage by property type. Why every warehouse in Denton and Tarrant is filed as commercial.

Retail sub-types: gas stations, convenience stores, car washes

A broker who works fuel retail cannot use “retail”. Since October 22, 2026 a retail property carries a sub-type where the county's own record names one. In New York it is the Department of Finance building class: G3 (gas station with retail store), G4 (with service or auto repair) and G5 (gas station only) are gas stations and GW (“car wash or lubritorium facility”) is a car wash; GU is a car sales lot and is neither. The same reading moved the K store classes to retail, which PLUTO's land-use bucket had filed as office. In Dallas it is DCAD's building description in words; in Collin the CS use code, split into gas station or convenience store by the trade name on the roll, and FSCW / SSCW for car washes. In Tarrant and Denton the sub-type is part of the same estimate as the type, from the businesses on record at the address, and carries the same ≈. Houston, Austin and San Antonio have no sub-type, because their rolls give no class to read it from, and the filter is not offered there. A sub-type only ever sits inside retail; an absent one means the record does not say, never that the parcel is not a gas station.

One loan, many parcels

A blanket deed of trust — a builder's inventory loan, a rental-home portfolio, a multi-site net-lease facility — covers many parcels under one instrument. We keep one loan and link every parcel to it, so the same amount appearing on several properties is one mortgage seen from each of its parcels, never double-counted. The portfolio filter (“2+ parcels”) finds these directly, and each row shows how many parcels ride on the instrument. A Texas facility is the other way round: one financing is recorded as a separate deed of trust per property, each stating the full note. Those are one loan too — the widest recording carries it and the rest are marked as its members — and the amount is never summed across them.

Corporate credit facilities. A deed of trust can also pledge a company's real estate for its revolving credit line or secured notes — a refiner's terminal, a cold-storage operator's plants, a builder's house-inventory line. The lien is real and the maturity is stated, but the amount is the facility's ceiling and the collateral is the company, so no broker refinances one property out of it. We badge these Corporate credit facility, keep them in the list and leave them out of the dollar headlines. The badge is set by a person reading the deed, never by the document reading alone, which nominates candidates: a REIT or a retailer borrowing at company level against its own buildings stays property debt.

Who borrowed, and what for

A loan carries a borrower type where the record can say one: home builder, developer, contractor, hotel operator, health care, religious or nonprofit, bank, or public body, and in Texas also investor, real-estate services and operating business. In Texas the type is the line of business the borrowing entity itself registered with the state Comptroller. Elsewhere, and for a Texas entity that registered nothing telling, it is read from a word in the name (“… Custom Homes LLC”) and marked ≈; words every LLC uses — Development, Holdings, Properties — are not read. A registered type always beats one read from the name. Most single-property LLCs carry no type, and that means the record does not say, never that the borrower is not a builder. A bank in the borrower's seat is most often property the lender took back, not a customer.

The borrower box takes several names or words at once and matches the owner of record as well as the borrower on the loan; a word matches a whole word, so “Shell” finds the oil company and not every Shelley. You can also upload your own list of firms and people — an association directory, a CRM export — and use it as a filter: a loan matches when the borrowing firm is on your list, or when a person on your list is one of the people on record behind it. Your list is yours alone.

Purchase or refinance. In New York, a loan recorded together with the deed that bought the property is a purchase loan, and one recorded on its own is not. A Texas deed states no price, so there the appraisal district's record of how the owner took title decides it: a loan recorded with that deed, or within weeks of it, financed the purchase; a loan recorded months or years later did not. Where the district cannot say — the property has sold since, or its roll is older than the loan — the field is blank rather than guessed.

Previous owners. When a building sells, the seller's loan is paid off at closing or assumed by the buyer, and often nothing is recorded that says which — Texas sets no deadline for filing a release. The loan stays listed, because the record has not closed it, and is labelled Previous owner wherever the assessor's owner of record is a later borrower on the same property. Its people are listed last on the contact card under that label and never as the decision-maker.

Estimating a maturity when none is recorded

No county index records a maturity. Where the recorded document states one, or states a term (“thirty-six months”), we read it and cite the page. Where it states nothing — most New York consolidation agreements leave the note unrecorded — we do not pick one date and present it as fact. We keep every plausible term for that kind of lender:

• Banks, credit unions, unknown lenders — 5, 7 or 10 years.
• Debt funds, and bridge or construction agreements — 3 or 5 years.
• Agency — 10 years. Life companies — 10 years, some 5.
• CMBS — 10 years for loans originated before 2023; 5 or 10 from 2023, when conduit lending moved to five-year terms.

These are the common loan terms for each kind of lender — the convention a broker would quote — not statistics drawn from our own data, and a change to them is an editorial decision dated in the data notes. If you want only what the record states, Stated dates only in the filter bar keeps loans whose maturity, term or rate-reset date was read from the document or a securitization disclosure, and with a date window set, the stated date itself must fall inside it.

A search window that lies entirely in the past — “matured, no payoff recorded” — matches only maturities read from the recorded document. A lapsed estimate means our model guessed a shorter term than reality, not that the borrower is in default, so estimates never appear as past due. The label says what the record says and no more: the stated date has passed and no satisfaction is matched to the loan. Most often that is an extension or a workout nothing was recorded for, or a refinance whose satisfaction was filed late or never — not a loan in default. Lists follow the same rule: sorted soonest first, they open on what is coming up, then on stated maturities that have passed with no payoff recorded, and put lapsed estimates last. The same window also matches a first rate-change date read from the rider that has passed: the loan is floating today, which is the event on 30-year paper, and the row says “rate reset”, not past due.

Long-dated bank notes: the reset, not the maturity

When the recorded document does state a maturity but it is 25–30 years out — the standard form for New York bank multifamily lending — that date is not the event a broker works. The note carries a fixed rate for an initial period, then floats. We read the Fixed/Adjustable Rate Rider attached to the recorded instrument and record the stated first rate-change date, the fixed period and the cadence after it — and, where the rider prints them, the initial rate, the index and margin the rate re-sets to, and the caps that bound it. Each is quoted as the rider states it; we never compute a current rate from the index, because that figure would be ours, not the record's. Across the 5,405 riders with a stated reset in the 1–4 family segment (read on every instrument recorded since 2015) the fixed period is 10 years for 34% of loans, 7 years for 32% and 5 years for 18%; after it the rate re-sets annually on 43% and every six months on 39%. Where a commercial long-dated note's rider was not found the loan keeps 5/7/10-year candidates weighted by the measured mix, and the row says so (“reset ≈ 10-yr”); a house loan gets no model at all. A stated reset date is shown as a reset, with the page it was read from — in the app and through the connectors — never as an estimated maturity.

A maturity filter such as “next 12 months” matches a loan ifany plausible term lands in the window, and the row says which one did (“≈ 5-yr”). The headline estimate shown elsewhere is the most common term for that lender type (5 years for banks, 3 for debt funds, 10 for agency, life companies and CMBS originated before 2023, 5 for CMBS from 2023), marked ≈. A maturity read from the recorded document replaces the whole set and renders in bold with its citation.

Prospect score

A 0–100 convenience rank, computed from fields you can see — not a black box — and re-stamped every night against today's date. It is four groups, and every loan shows its own: hover a score in the results, or read it under the score on the property page.

• Timing, up to 40 — the broker event (a stated next rate reset, a stated maturity, or the weighted average across the plausible terms when the maturity is modelled) scores 40 at 3–18 months out, 30 under 3 months, 22 at 18–36 months or matured within the last two years, 8 at 3–5 years.
• Certainty, up to 20 — 20 when the event date is read from the recorded instrument, 8 when the maturity is read but the reset is modelled.
• Size, up to 15 — unit count (20/50/100+) and blanket breadth (3/10+ parcels) up to 10, then loan amount, log-scaled, up to 5.
• Signals, up to 10 — lis pendens or foreclosure signals; minus 10 for recently modified paper, and minus 10 for one lot on a builder's revolving line. Minus 20 when the record says the loan is probably gone: on a loan secured by one parcel, a later mortgage on that lot at least half its size, or a later sale for at least half its amount — a refinance or payoff whose satisfaction was never recorded or never matched.

The score is the four added up as a share of 85. Property type is not scored — which types you work is a filter, not something we rank for you — and loan size is a tiebreak rather than a driver. Brokers weight these differently, so the score is a default ordering, not a verdict.

Choosing the score. The Score picker in the filter bar swaps the default for one of two others, and the column and its hover follow: rescue & distress scores a loan past its stated term with no payoff as full timing and adds distress, modified paper and leverage above 65% — the default score's penalties are a rescue desk's leads — while a loan the record says is gone loses 30; size-neutral is the default without its size group, for small-balance desks. A saved search keeps its score, so its Monday email ranks the same way. Templates — bank refi, rescue and preferred equity, small balance with a non-bank lender, bridge-to-perm takeout, lender intake — fill in the filters and score one kind of desk starts from; everything they set stays editable.

Keeping it current

Every week, per county, we re-read the recorder and check each loan against what was filed since (in Texas the same events arrive as releases, trustee's deeds, modifications and foreclosure-sale notices):

• Satisfied — a recorded satisfaction or release, matched to the loan it names, takes the loan off the active list. A New York satisfaction names the notes a consolidation folded in rather than the consolidation itself, so a satisfaction of any of them retires the loan we list. A partial release is recorded but does not retire the loan. In Texas the opposite holds: a facility recorded once per property is released one property at a time, and a release frees that property only.
• Superseded — a later consolidation that cites the loan folds it into the new one, and the old note stops being a separate maturity. This is how most New York loans actually end. A condominium sponsor's loan is often recorded again with each unit sale, every copy citing the original mortgage: when the copies add up to more than that mortgage, they are one loan stated several times, and only the latest is listed.
• Modified / assumed — a later modification or assumption flags the loan: its stated terms are superseded and must be re-verified before citing. A Texas clerk does not cite the loan a modification changes, so it is matched through the property and the borrower's name, the way a release is; a loan past its stated maturity that carries the flag was most likely extended.
• Foreclosed — a referee's deed conveying the collateral marks the loan foreclosed.
• Distressed — assignment to a special servicer, a recorded judgment, a deed-in-lieu, or a tax-lien sale flags the loan as distressed or on watch.

Default and distress

A missed payment is never recorded. The record first hears of a default when the lender files: in Texas a notice of sale posted at least 21 days before a first-Tuesday auction (Dallas and Denton publish them, Tarrant and Collin do not, and every county takes them down once the date passes), in New York a notice of pendency filed with the County Clerk, which the register never carries. So a loan with no signal is a loan the recorder has not been told about, never one we know to be performing, and the Distressed and Watch labels name the document that put them there. Payment status itself is public only for securitised and HUD-insured loans, whose servicer reports we attach where they match a recorded loan. The default guide walks each state's paper trail and what to check for one loan.

Check one building

Everything above can be seen for a single address on the free maturity lookup: every active recorded mortgage on the property, the lender, the amount, and the maturity as read from the document or marked ≈ where it states none. If you own the building and would like the loan refinanced, the same page has one line for that; a broker who works your market will call.

One county at a time

The same counters, per county: how many active loans the record holds, how many come due in each window and how many of those dates are stated on the instrument, the collateral mix, and the ten institutions with the most loans maturing in the next year. Each page is recomputed daily from the record and says when. Manhattan · Bronx · Brooklyn · Queens · Dallas County · Tarrant County · Collin County · Denton County · Harris County · Travis County · Bexar County.

Owner contacts

A borrower is usually an LLC. We resolve the people behind it from the public record first, and every name says where it came from: the officers and owners an LLC must list on its HPD property registration, the owner named on a DOB permit filing, the assessor's owner of record, the mortgagor of record, and the person who signed the recorded instrument itself (read from the signature block and notary acknowledgment; tier X, cited by page). A property shows who is there by role and source. One credit unlocks every name, firm and address on the record and looks for a work email: the decision-maker's first, then the next principal's only until someone can be reached. Any other person's work email is a free ask from the card. Every unlock also reads the last pages of the leading instrument, and who signed for the borrower lands on the card within about a minute, with the title they signed under.

The card leads with the decision-maker on record — people are ranked by their role, the record that names them and their title, whether or not a contact has been found yet. The signer of the instrument outranks everyone; on an HPD registration the person it designates as head officer counts as holding a top title whatever was typed in the title box, because two “Members” of one LLC are otherwise a tie. When the decision-maker cannot be reached today, the best-ranked person who can is shown beside them, labelled as exactly that. A managing agent or a contractor's permit contact never leads, and nor does anyone behind a loan in a previous owner's name: they are listed last, under that label. In Texas the clerk types a person surname first with no comma, and the name is read that way round before any source is asked.

A direct phone number is the one add-on, per person and charged only when a number is found; it is never looked up unasked. Nothing is charged when the record names nobody. If you doubt what a card shows, Re-check contacts asks the sources again about the top people and replaces your view with what they say now, one credit; it can drop a stale answer, and it can also lose one a source no longer returns, so it is your call, not ours.

The public record names the LLC, not the firm the person runs, and contact sources are indexed by firm. Where the record names only the property's LLC we identify the operating company behind the person, confirm it against a page that names them, and show it beside their name (“runs …”). The same distinction drives the phone lookup: a person registered at a home can be traced there; a person registered at an office is found through their firm, never by tracing the office, which would return its other tenants.

When the record names nobody. Some loans reach us without a borrower at all: a securitization or agency disclosure reports the building and the debt but leaves the borrower blank, and a Texas deed of trust names a single-purpose LLC the clerk indexes no person against. A disclosed loan is first tied to the instrument the county recorded at its address, and that instrument's parties, HPD registration and signature page name people the disclosure does not — so this case is rarer than the disclosure alone suggests. Where no instrument is recorded, we ask the open web two questions — who bought this building, and who operates it — and accept an answer only when a page we fetched prints the property and the firm together. The page is linked on the card and the sale is dated, so you can judge it yourself. The principals of that firm then come from a business directory or the firm's own leadership page, and they are labelled at the owner firm, never “on record”: these are people the open web places at the company, not people a public record ties to the property. An operator is not an owner, and the card says which it found — a management company or a leasing broker is shown as the operator, because that is what the leasing page proves. A property whose owner the web cannot name with a citation stays blank and is not charged.

Phone numbers and emails are never guessed. Each one either comes from a contact-data source that has it for that exact person — the name must match, or the answer is discarded — or is read from a page the firm itself publishes, which is linked beside the number so you can see it for yourself. A number we cannot verify is not shown. Every number is labelled for what it is — mobile, home line, or the firm's switchboard — with the kind of source that supplied it, and mobiles are listed first. Some numbers take a few minutes to arrive; the card says so while it waits. Every number carries the Do-Not-Call and litigator flags we have for it: a number marked DNC is shown greyed and is not to be called, and nothing here is for automated dialing. A contact source that is unavailable closes that lane for the lookup rather than failing it; the person is asked again once it is back.

Lenders, and the people at them

The lender on a recorded mortgage is a name as the clerk typed it, not an institution. We count one row per mortgage however many parcels it covers, and only the party that actually lent: MERS, which sits in the lender's seat as a registry for the real lender, loan servicers and securitization trustees are left out — MERS alone is the second-largest name in that seat in New York and has never lent anyone anything. Every spelling of one institution is counted together (JPMorgan Chase is recorded under 64). A trade name, a successor bank and a sponsor's lending vehicles are folded onto the institution a broker would call by review, checked against the sponsor's own filings: Derby Copeland Capital records its loans under more than 30 entity names, including a joint venture with Fortress. Where one parent runs separate lending desks on their own terms, each desk keeps its own page and the others are listed beside it as related desks. A lender whose every loan goes to one borrower is a related party, not a market lender, and is not counted as lending.

Lender rankings — “who lends on deals like this” on a property, the lender search and each lender's page — count what a lender recorded in the last three years, list it only from three closings, and date it to its last one. That is appetite revealed by the record, not a credit box: no lender has told us what it wants to lend on today. The lender named is the one that originated the loan, which is not always the one that holds it now; loans are sold, and the record keeps the original name.

Named people at a lender come from the lender's own website — its team, leadership and originations pages — linked, with the date we read the page. A person from such a page is shown as covering a market for that lender, never as the person behind a particular loan. A name is tied to a specific loan only when the lender's own announcement of that loan names them. Every name must appear on the page it cites; the pages are re-read about once a quarter and a person who has left drops off. Many lenders publish nobody, and for those the page says so rather than guess. A work email is shown in full, with the page that publishes it, where one is published. Where none is, the address may be shown in grey with a question mark: built from the format the firm's other people use, likely right but not certain, and labelled as exactly that.

Interest rates

The recorder files the mortgage, not the note, so a rate appears on the public record only when the lender’s form restates it. Where it does, we read it from the document; securitized loans carry their rate in the deal’s disclosure. A hybrid note’s rider restates the initial fixed rate and names the index and margin it re-sets to, so a loan with a stated reset usually carries its rate as well. Every other loan shows a typical rate instead: the middle half of the stated fixed rates we hold for loans of the same lender type and origination half-year, with the count behind it. It is a market range for that cohort, never the loan’s own rate, and it is withheld when the cohort is too thin to be meaningful. It recomputes daily as more rates are read.

What public records cannot tell you

Honesty about limits is part of the method. No county index records a maturity, so many maturities are estimates until read from the document, and New York consolidation agreements often leave the note unrecorded. A paid-off loan stays listed until its release is filed, and Texas sets no deadline for filing one. Most lawsuits live with the courts, not the recorder, and a Texas foreclosure notice is posted and taken down, so distress there is undercounted. A loan-to-value needs the value of every parcel behind the loan, which we have for fewer than half of active loans: it is shown where it exists, but it is not a filter, because a filter would quietly drop every loan without one. Current balances, debt coverage and occupancy exist only in securitization disclosure. We show these gaps rather than fill them with guesses.

Frequently asked questions

Where does Circlemark's commercial mortgage data come from?

From the county recorder first: NYC ACRIS and the Texas county clerks' real-property indexes. Then the assessor (NYC PLUTO, the Texas appraisal districts) for parcel attributes and values, the scanned instrument itself for principal, maturity and rate, and SEC, HUD and Freddie Mac disclosure for securitized loans. No loan fact — amount, lender, maturity, status — ever comes from trade press.

Can Circlemark tell me who owns a building when the loan names only an LLC?

Usually. The public record comes first: the officers on an LLC's HPD registration, the assessor's owner, the mortgagor, the person who signed the instrument. Where the record names nobody — a securitized loan with a blank borrower, a Texas single-purpose LLC — we identify the owning or operating firm from a page that names both the building and the firm, link that page, and show its principals as being at the firm rather than on the record.

What is a provenance tier?

A letter on every field saying how it was known. R is read off a recorded instrument, P from a public dataset, X extracted from the scanned document by a vision model, C computed from those, S securitization disclosure and M modelled. An M value is always shown as an estimate with an ≈ marker, never as a fact.

How does Circlemark know a loan's maturity date when the record has no maturity field?

It reads the scanned instrument. ACRIS has no maturity field, but the document or its rider usually states one, and that date is recorded as stated with the page cited. When nothing is stated, the loan keeps every plausible term for its lender type, each marked ≈, and a past-due filter never matches an estimate.

Why does one loan appear on several properties?

A blanket mortgage or deed of trust covers many parcels under one instrument. Circlemark keeps one loan and links every parcel to it, so the same amount on several properties is one mortgage seen from each of its parcels, never double-counted. Each row says how many parcels ride on the instrument.

How is a Texas loan amount found when the clerk's index records none?

Every screened deed of trust is read by a vision model for the stated principal and maturity, with the page cited. Where an instrument secures a larger facility only to the extent of a local amount, the local amount is shown and the facility total kept; a facility recorded in full in every county is reported as recorded.

How does Circlemark count who lends in a market?

One row per recorded mortgage, only the party that actually lent — MERS, servicers and securitization trustees left out — with every spelling and lending vehicle of an institution counted together. Lenders are ranked by recorded closings in the last three years, listed from three, and dated to their last one: what they funded, not what they promise.

Where do Circlemark's lender contacts come from?

From each lender's own website — its team, leadership and originations pages — and, for a lender that publishes no team page, a page that lists the person at the firm; every name is linked to its page and dated. Those people are shown as covering a market for the lender, not as the person behind a particular loan; a loan-specific name appears only when the lender's own announcement of that loan names them. A work email is shown in full where a page publishes it; otherwise it may appear in grey with a question mark, built from the format the firm's other people use and labelled as exactly that.

Can I filter rent-stabilized buildings and co-ops in New York?

Yes, as two New York filters. A co-op is flagged from the Department of Finance building class or a borrower filing as an owners' or tenants' corporation; its loan is the corporation's underlying mortgage. Rent stabilization is read from the per-unit fee a registering owner pays on the Department of Finance tax bill, with the unit count shown; a pre-1974 residential building of six or more apartments with none registered is marked likely, as an estimate. A building marked neither is untested, not market-rate.

Can I find loans to home builders or developers?

Yes. Each loan carries a borrower type where the record can say one. In Texas it is the line of business the borrowing entity registered with the state Comptroller; elsewhere it is read from a word in the name, such as “Custom Homes LLC”, and marked ≈. A loan with no type is one the record cannot place, not a loan to someone outside the type.

How often is the data re-verified?

Every week, per county, the recorder is re-read and each loan checked against what was filed since: a satisfaction or release retires it, a later consolidation supersedes it, a modification or assumption updates it. In New York a satisfaction of any note a consolidation folded in retires the agreement; in Texas a release frees one property only.

Data notes

Every change that moved a count, widened coverage or altered a definition is dated when it reached production, on the data notes page. The latest:

• Oct 23, 2026 — Loans on single houses: found by address again, and counted above the results. A New York loan on one house is found again when Residential 1–4 Units is picked. From the evening of October 22 to October 23, picking the type returned no loan on fewer than three parcels, on the web and through the connector, so a search for the segment came back empty. Two things are new with the fix. An address or a keyword now finds a house loan whether or not the type is picked — whoever types the address of a house wants that house. And an "All types" search still returns commercial collateral only, but a line above the results now says how many loans on 1–4 family houses the same filters would add, with one click to include them. The coverage counters are unchanged by it.
• Oct 23, 2026 — Securitized-loan disclosures had not been read since mid-September. The nightly read of securitization disclosures — the monthly servicer reports that give a conduit loan its payment status, special-servicing flag, balance and prepayment dates — failed every night from September 13 to October 23. A loan known from those reports kept the status it had in mid-September: a transfer to special servicing or an extension reported since then did not reach it, and no loan disclosed for the first time in that window was added. The cause was a county listed ahead of its launch, not the filings; the Freddie Mac disclosure read stopped the same way on October 20. Both are running again and the six weeks of reports are being applied, so every figure drawn from them will move as they land. Recorded loans were unaffected: the county record is read weekly and never stopped.
• Oct 23, 2026 — Lender names reviewed by hand: mergers, vehicles and non-lenders. Lender rankings, the lender search and the lender pages now reflect a hand review of every lender whose recorded names share a website — 93 groups, decided one by one against the bank regulators' records and the lenders' own filings. A bank's former name counts under its successor: Flushing under OceanFirst, Veritex and Cadence under Huntington, Independent Financial under SouthState, and every spelling of Dime under Dime. A fund's lending vehicles count under the firm that runs them, and three vehicles that had been credited to the wrong firm are now under the right one. Thirteen names that lend nothing — loan servicers, bank trust departments, bond trustees, reverse-mortgage lenders — are out of the rankings. A lender's count can move either way; its loans did not. Lenders by county.

All 55 notes, by month.