Skip to main content
Early access: while the developer tier is in beta, the API serves this route to Investorlift’s team and trusted partners. Investorlift will restrict the route further before it opens to every key.
A lender is the lender of record on a recorded mortgage. Its id is len_ plus 12 hex characters. The lender has the same id on every instrument and in every market, whatever spelling the document used. KIAVI FUNDING INC, KIAVI FUNDING LLC and LENDINGHOME FUNDING CORP are one lender under one id. names[] lists every spelling the recorded instruments use. Store the id, and keep the name for display. After a registry run, an old id goes one of two ways. Either it redirects: you get the current profile, and meta.resolved_from names the id you asked for. Or the run retired it: you get 410 gone with superseded_by: null. Then search by name again, as Retired ids explains. Every value is a snapshot at as_of, the financing slice date in meta.coverage[].lenders.as_of. Every response that carries one says so in meta.dated[]. See dated data.

​
How a lender is identified

identity_basis says how the registry tied a spelling to its lender: The registry never merges acquisitions. A lender that bought another’s book keeps its own id, and so does the lender it bought. Successor says which is which. The id is a function of the folded name, not of the market or of a data refresh. So the id is the identity to store. name can change under it when one spelling overtakes another. A hand-decided merge joins two spellings by a dictionary row and retires the smaller id. The retired id still answers. See Retired ids.

​
Class and the hard-money flag

lender_class is a reading of the name, never a legal status. It is the same reading the parcel’s financing block carries. The five classes:
  • BANK
  • NONBANK
  • PRIVATE: a trust, a seller carry-back or another private party
  • INDIVIDUAL: a person’s name
  • GOVERNMENT: HUD, the VA, a housing authority
Since 0.17.0 the registry reads the class on the folded name. So punctuation, initials and a missing space cannot defeat a rule. KS STATEBANK and MANUFACTURERS & TRADERS TR CO read BANK. A person whose surname is Banks does not. A securitisation trust, or a trust with lender words in its name, reads NONBANK, not PRIVATE. A housing authority, a city or a county reads GOVERNMENT unless it is a credit union. PRIVATE is not the hard-money sense of “private lender”. For that sense, read is_hard_money. It is true when the name carries hard-money vocabulary or is a known brand (hard_money_basis: NAME). Since 0.16.0 it is also true when the book behaves like a hard-money book, that is when three things hold:
  • the lender has 20 or more loans since 2015
  • a quarter or more of them have a recording date within 45 days of a registered investor’s purchase
  • the lender is not a bank
The basis is then BEHAVIOUR, or BOTH when the name says so too. The second rule catches a lender that lends hard money under a plain name, once the deed link of its market exists. A market without a deed link has the name rule alone. investor_lending.is_investor_lender is the neighbouring flag: half or more of the window’s loans went to registered investors, whatever the name.

​
Class basis

class_basis (since 0.17.0) says why the class reads as it does: A class is a reading, never a verification: a name that reads BANK is a name with a bank word in it. When a registry run reads a lender’s name as a person’s where it did not before, the lender falls under the person gate. If it loses its id, the run retires the id with an alias row, and the id answers 410 gone. So the change is never silent.

​
Successor

successor (since 0.17.0) is the lender this one is now part of, on a market profile. The link is an ACQUISITION, where the acquirer bought the lender or its book, or a SUBSIDIARY, where the parent owns it. The block carries the successor’s id, name, lender_class and is_hard_money. The registry never merges the books: BANK ONE’s loans stay BANK ONE’s, and WELLS FARGO BANK NORTHWEST NA keeps its own profile. The profile of each points at the lender it is now part of instead. It is null when the registry holds no link, and null when the linked lender has no profile in the market. Investorlift keeps the links by hand. So successor is present on a few dozen lenders, not on every lender an acquirer ever bought.

​
NMLS id

nmls (since 0.17.0) is the lender’s NMLS id and the NMLS Consumer Access page it opens (id, url). Investorlift curates it by hand for the largest lenders, and starts from the most active in the last 24 months. It is null for every lender no one curated yet, which is most of the registry. A null is “not looked up”, never “no NMLS registration”. The id is the lender’s, so it is the same on every market profile and on the union object too.

​
Retired ids

Ids are a pure function of the folded name, so a registry rebuild re-issues the same id for the same name. A run retires an id only in three cases. The run decides that two ids were one lender: a spelling pair or a truncated spelling joined by a dictionary row, reason: MERGED or MERGED_PARTIAL. The run splits an id (NEW_SPLIT). Or a lender’s rows leave the registry altogether (RETIRED). The registry keeps every retired id since the first run, in every market that served the lender:
  • On GET /v1/lenders/{id}, /loans and /borrowers a merged or split id answers 200 with the current lender in data. meta.resolved_from lists the ids the API followed, oldest first. Store the current id.
  • A RETIRED id answers 410 gone with superseded_by: null, the reason and the registry run that retired it. Search by name with GET /v1/lenders/search.
  • An id that no run ever issued answers 404, and so does a chain that no longer leads to a lender.
  • id_history[] on a market profile is the other direction: every id retired into this lender, oldest run first. Each carries its reason and the lender_run that retired it. It is empty on nearly every lender.
The financed_by and filters.financing.lender_id filters follow an old id the same way and stamp meta.resolved_from on the list. A registry built before 0.17.0 has no alias table: an old id answers 404 there, and meta.resolved_from is absent.

​
What the counts are

Two tables feed the registry. The first is the open lien positions at the slice date, up to 4 per parcel. Each comes with the balance, rate, term and maturity the open-lien table carries. The second is the recorded mortgage history: the 5 most recent mortgages per parcel. The history captures about one open loan in five (coverage.history_capture_share), and a paid-off loan is in neither table. The consequences:
  • n_loans, n_loans_12m and n_loans_24m count instruments once across both tables. The same parcel, date and amount is one instrument. They are the origination counts, and the lender list ranks by n_loans_24m.
  • n_in_history and its windows are what the history holds, duplicated slots included. They are a floor, never an activity measure.
  • A bridge lender’s book is the most undercounted. The registry counts a short-term loan only while it is open, and the lender’s paid-off loans are in neither table. Compare hard-money lenders with each other, not with a bank.
  • Every by_year row before history_floor_year is a floor: the history keeps 5 slots per parcel. On a parcel with all five slots filled, the older mortgages fell out.
  • Volumes leave out amounts above $5,000,000. The file records a commercial facility’s whole amount on every parcel it covers, so one facility can add billions to a bank’s window volume. volume_24m, every volume in by_year, by_month and geography, and the terms.amount quartiles leave those instruments out. n_amount_excluded_24m counts them, and the loan row keeps its amount with amount_outlier set. A blanket loan, one loan recorded on 5 or more parcels, counts once in the same sums, and n_blanket_positions_24m counts its positions. Loan counts and the ranks by loans never change.
  • Null means not measured, and zero means measured and none. n_open_maturing_12m_from_as_of is null when fewer than half of the open positions carry a maturity date (share_due_known). n_purchase and n_investor on a year, and every purchase, borrower, investor and flip figure below, are null without the market’s deed link or borrower match. purchase_measured, investor_lending_measured, flips_measured and borrowers_measured on the market profile say which of these the market measures. takebacks is null as a whole while takebacks_measured is false, and n_counties_ranked_first_24m is null while counties_measured is false (since 0.19.0). A margin is null under 20 resold flips with both prices, and a quartile under 3 values.
n_loans_24m counts every instrument of any kind, purchase money, refinance, HELOC and construction alike. For a GOVERNMENT lender it counts the insurance claims and partial-claim seconds recorded under the agency’s name, not originations.

​
One profile per market

The object carries the union totals: n_loans_24m, volume_24m, n_open_liens and open_balance, summed over markets[]. It also carries one market profile per loaded market where the lender has loans, most loans first. Each market profile carries these blocks:
  • rankings: the rank and share over the 24 months that end on as_of, overall, within its class and among hard-money lenders. The registry computes them over every lender of the market, the ones the default list hides included. So a profile’s rank can be absent from a default page of the list, which numbers its own rows.
  • coverage: the counties the registry covers, the parcels left out, the slice date, recordings_through and the capture share.
  • the by_year and by_month series against the market
  • the product, term, label and property-type mix
  • terms: the amount, rate and term quartiles, each with how much of the book carries it
  • geography
  • a summary written from the served numbers at build time
recordings_through is fourteen days before as_of. The two weeks between are unmeasured, not empty. The registry measures every recency on a lender against recordings_through, not against the slice date.

​
Who borrows, what was financed and how it ended

Since 0.16.0 the registry joins the recorded deeds to the loans. Three blocks and some counts read the join. The registry writes all of them into the profile at build time, and never computes them on read. Each is null as a whole where the market lacks its input.
  • Purchase money. The deeds decide purpose on every loan row, never the file’s label. It is PURCHASE when a priced deed on the parcel is dated in the 45 days up to and including the recording date. It is NOT_PURCHASE when the deed registry covers those days and none falls in them, and UNKNOWN when it does not.
  • Purchase money on the profile. by_purpose_24m splits the window, and sort=purchase_desc orders by n_purchase_money_24m. rankings.rank_purchase_24m places the lender by it. terms.loan_to_price is the loan against the deed price on the window’s purchase loans. A value above 1 is a rehab holdback on top of the price, not an error. purchase_measured says if the market serves any of it.
  • borrowers. Who borrows. The block counts the borrowers, all time and in the window, and how many resolve to a registered investor. borrower_measured_share is the share of the book with any borrower source at all. Not every loan has one: the recorded history names borrowers, and the open-lien table does not. The block also carries repeats, newcomers, churn, the ten largest borrowers and the ten lenders that share the most borrowers with this one.
  • A borrower is a key, 12 hex characters hashed from the folded name, the same key on every lender’s rows. The profile serves a company’s name beside it. A person’s name is never on the profile, and person_names_redacted says when the registry dropped one.
  • n_churned_12m counts the borrowers whose last loan here is 12 to 36 months old and who took a later loan elsewhere. It counts them only when no open lien of this lender follows that later loan. On a feed that sees one loan in five, that count is a lower bound of leaving. It is not an upper bound of anything. churn_measured is false where the registry cannot read churn at all. GET /v1/lenders/{id}/borrowers lists every borrower.
  • investor_lending. How much of the book funds registered investors. The block carries the investor loans and their share of the window, and the investors and their kinds. It carries the deals whose purchase loan is this lender’s, by what occurred on them, and the auction and wholesale purchases among them. It also carries the ten largest investors and is_investor_lender. The purchase loan of a deal is a loan recorded in the 45 days from the purchase, the largest when several.
  • by_outcome_24m and median_months_to_payoff beside it say what became of the captured loans. The registry reads them from the later deeds and mortgages. The outcomes are RESOLVED_BY_RESALE, REFINANCED, OPEN, UNKNOWN and, since 0.19.0, FORECLOSED. The last one is a take-back deed on the parcel before any resale or later mortgage. They are a performance proxy and never a payment history.
  • flips_financed. The flips bought in the 36 months to 45 days before recordings_through whose purchase loan was this lender’s. The block counts them, and it also carries the lender’s share of every flip the slice can attribute. capture_share measures them against the flips by investors who borrow from this lender elsewhere, whose own purchase loan the slice did not see. Where 20 or more flips resold with both prices, the block carries the resale margins: median_gross_profit with its quartiles, median_spread_ratio, median_hold_days and median_loan_to_price. Under that the margins are null, and a gross profit is never net of rehab or costs.
The capture behind all three is partial. The open-lien table sees about one purchase loan in five on holdings, and one in fifteen on flips. So every count is a floor, and a NONE_CAPTURED purchase on a deal is not a cash purchase. The summary paragraph never quotes a margin.

​
Take-backs

takebacks (since 0.19.0) is what the lender took back at foreclosure. The registry reads it from the county’s recorded deeds since 2016 and never from a delinquency or default feed. A take-back deed is a trustee’s deed, a sheriff’s deed or a deed in lieu of foreclosure whose grantee is this lender. The registry matches the grantee under any of the lender’s spellings, through the same fold that gives the lender its id. The grantee on such a deed is usually the trustee or the servicer. So the originator of a securitised loan is invisible here, and a bank that services a large book reads large. The block does not measure delinquency, only the deed that ended it.
  • n_takebacks counts the lender’s take-back deeds since 2016 and n_takebacks_24m those in the 24 months to as_of. by_instrument[] splits them into TRUSTEES_DEED, SHERIFFS_DEED and DEED_IN_LIEU, each with its share.
  • n_loans_foreclosed is the take-backs on a parcel that carries a loan row of this lender recorded before the deed. n_takebacks_unlinked is the rest, for one of three reasons. The loan predates the five slots the history keeps. It sits in neither table. Or another lender made the loan, and this one is that lender’s trustee or servicer.
  • foreclosure_share is the share of the lender’s captured loans whose outcome is FORECLOSED. A captured loan is one tied to an investor deal. The share is null under 20 captured loans. median_days_takeback_to_resale is the median days from the take-back deed to the next priced sale of the parcel. It is null under 3.
  • is_auction_lender is true when the lender took property back at a trustee’s sale by credit bid or by deed in lieu, whatever the name says. The flag reads the deeds the auction registry classifies, and those deeds alone, never the class. A bank with no take-back reads false.
The block is null as a whole while takebacks_measured is false on the market profile. Every count leaves out every take-back deed of a bulk judgment sale held for review. foreclosed on the loan row is the per-loan side of the same deeds.

​
Rankings by place

Since 0.19.0 rankings also carries share_by_volume_24m: the lender’s window volume over the market’s, under the $5,000,000 rule. It also carries three counts of the places the lender leads, by ZIP, city and county: n_zips_ranked_first_24m, n_cities_ranked_first_24m and n_counties_ranked_first_24m. These count the places where the lender recorded 10 or more loans in the 24 months to as_of and ranks first by that count. The rank is among every lender of the market there, persons and government included. The county count is null while counties_measured is false, as it is with one loaded county. GET /v1/lenders/{id}/rankings gives the places themselves, with the lender’s rank and share in each, over any window and at the hex-cell grain too. GET /v1/lenders/{id}/cells gives its loans per cell. The summary paragraph says how many ZIP codes the lender ranks first in when that count is 1 or more. On a registry built before the place rankings, the four fields are null.

​
The person gate

A lender whose name reads INDIVIDUAL or PRIVATE gets an id and a profile only when it passes a gate (profile_gate: SERVED). The gate is 10 or more loans on 5 or more parcels. Below it, the name stays on the parcel’s financing block alone, with no id. The list and the search hide INDIVIDUAL and PRIVATE lenders unless include_persons=true. They hide GOVERNMENT lenders unless include_government=true. This is the way investor lists hide institutional buyers.

​
Contact data

Nothing in any delivery carries a lender’s phone or email, so no lender object has a contact block. Lender names are business records of the loan. The API serves them in full to every key, whatever the class. The API serves a borrower as a key and, when it is a company, its name. A person borrower’s name is never on the profile, and this host does not serve it on the borrowers route. The profile names the investor behind a borrower by its registry name, the deed spelling, as every deal row names it.

​
Availability

The counties in meta.coverage[].lenders.counties can be narrower than the market. Counties, cities and ZIP codes lists them. A ZIP, city or county outside them answers 422 outside_coverage and names the covered counties, never an empty page. A geometry outside every loaded market’s tolerance answers the same. A market without a published lender registry answers 422 lenders_unavailable on every lender route. meta.coverage[].lenders is null there. A registry built before a phase serves that phase’s fields as null. It answers 422 lenders_unavailable on the routes and parameters that need them. meta.coverage[].lenders carries the flags: purchase_measured, investor_lending_measured, flips_measured, borrowers_measured, takebacks_measured and counties_measured.

One lender: who it is, its totals across the loaded markets, and a profile per market. The identity is the names, the class, the hard-money flag and how the registry established the identity. Nothing in any delivery carries a lender's phone or email, so no such block exists.

​
id
string
required

Lender id: len_ followed by 12 hex characters, for example len_3f9a1c27b4e0. The registry mints the id from the lender's folded name, so the same lender carries the same id in every market, stable across data refreshes. A hand-kept dictionary row can merge two spellings into one lender. The API then follows the old id from the earlier registry run to that lender and lists it in meta.resolved_from. A retired id with no successor answers 410 gone. Store the whole id, prefix included, as the identity.

Pattern: ^len_[0-9a-f]{12}$
​
name
string
required

The lender's display name in its lead market, upper case as the file writes it, for example KIAVI FUNDING INC. The lead market is the market with the most loans in the window. The name can change under a stable id, so it is never a key. It is a business name, and the API serves it to every key.

​
lender_class
enum<string>
required

A reading of the name, never a legal status. The parcel financing block carries the same reading. How the lender name reads. BANK: a bank, credit union or thrift. NONBANK: a mortgage company or other lending business. PRIVATE: a trust, a seller carry-back or another private party, not the hard-money sense of private lender, which the Lender object carries as is_hard_money. INDIVIDUAL: a person's name. GOVERNMENT: an agency or a public body.

Available options:
BANK,
NONBANK,
PRIVATE,
INDIVIDUAL,
GOVERNMENT
​
class_basis
enum<string> | null
required

Why the class reads as it does. NAME: the name rules of the registry, read on the folded name. The rules apply the government, bank, trust, lending and person vocabularies in that order. So KS STATEBANK reads BANK, and a trust with lender words reads NONBANK. DICTIONARY: a hand-kept row overrides the name's reading, for example BBVA USA as a BANK. Null on a registry built before the registry recorded a basis.

Available options:
NAME,
DICTIONARY
​
identity_basis
enum<string>
required

How the registry ties the spelling to its lender: NAME_FOLD means the folded name alone joins them. The fold uses upper case, drops punctuation and entity suffixes and spells out abbreviations: WELLS FARGO BANK NA and WELLS FARGO BANK N.A. are one. TRUNCATED_PREFIX: the file cut the name at 30 or 40 characters, and the registry joined it to the one longer name that starts with it. ALIAS_DICTIONARY: a hand-kept row joined it, for a rename such as QUICKEN LOANS to ROCKET MORTGAGE, a trade name or a charter form. A row also covers a spelling the same instrument carries both ways. The registry never merges acquisitions.

Available options:
NAME_FOLD,
TRUNCATED_PREFIX,
ALIAS_DICTIONARY
​
is_hard_money
boolean
required

True when the name carries hard-money vocabulary or is a known hard-money brand, or when the book behaves like one. The field hard_money_basis says which, and the behaviour test needs the market's deed link. A PRIVATE lender_class is a trust or a seller carry-back, not this. The hard_money filter of the list reads this flag.

​
hard_money_basis
enum<string> | null
required

Null when is_hard_money is false. Why is_hard_money is true: NAME means the name carries hard-money vocabulary such as BRIDGE, FIX AND FLIP, REHAB, PRIVATE MONEY or a numbered FUND. A known brand such as KIAVI, LIMA ONE or RCN CAPITAL also counts as NAME. FUNDING, CAPITAL and LENDING alone never qualify. BEHAVIOUR: the lender is not a bank and has 20 or more loans since 2015. A quarter or more of those loans sit within 45 days of a registered investor's purchase. BOTH: name and behaviour.

Available options:
NAME,
BEHAVIOUR,
BOTH
​
names
string[]
required

Every spelling across the markets, the lead market's first. These are business names, and the API serves them to every key.

One spelling, as the file writes it.

​
n_names
integer
required

How many spellings names holds.

Required range: -9007199254740991 <= x <= 9007199254740991
​
profile_gate
enum<string>
required

The profile gate of the identity: SERVED means it has a profile, the only value a served lender carries. BELOW_PERSON_THRESHOLD: a person-named lender, class INDIVIDUAL or PRIVATE, with fewer than 10 loans or fewer than 5 parcels. Such a lender has no id, no search hit and no profile, and its name stays on the parcel financing block only. PLACEHOLDER: a name such as NOT AVAILABLE, NOT PROVIDED or UNKNOWN, with no id. The market's coverage counts the two hidden values, and the API never serves them as rows.

Available options:
SERVED,
BELOW_PERSON_THRESHOLD,
PLACEHOLDER
​
nmls
object | null
required

The NMLS id and the NMLS Consumer Access page it opens. The registry curates the id by hand for the largest lenders. The block is the lead market's, and the same lender carries one id everywhere. Null where the lender has no curated id, and on a registry built before phase 4.

​
n_loans_24m
integer
required

Instruments recorded in the 24 months ending on as_of (730 days, inclusive), as of as_of, the slice date, never the clock, summed over markets[].

Required range: -9007199254740991 <= x <= 9007199254740991
​
volume_24m
integer
required

Sum of the loan amounts of the priced instruments in the 24 months ending on as_of (730 days, inclusive), as of as_of, the slice date, never the clock, whole dollars, summed over markets[] after each market profile's exclusions. The exclusions count a blanket group once and leave out amounts above $5,000,000. 0 when none carries a price.

Required range: -9007199254740991 <= x <= 9007199254740991
​
n_open_liens
integer
required

Open lien positions at as_of, summed over markets[].

Required range: -9007199254740991 <= x <= 9007199254740991
​
open_balance
integer
required

The lender's open balance after the exclusions each market profile describes, whole dollars, summed over markets[]. 0 when none.

Required range: -9007199254740991 <= x <= 9007199254740991
​
n_markets
integer
required

How many markets the lender has a profile in: the length of markets.

Required range: -9007199254740991 <= x <= 9007199254740991
​
markets
object[]
required

The per-market profiles, most loans in the window first. One entry when market= names one. Each carries its own ranks, coverage, as_of and dated. The union totals above are their sums.

Minimum array length: 1