How often the loan actually goes bad, by vintage
One row per programme per approval year. The denominator is every loan in that year the money actually moved on; the numerator is the ones recorded as charged off within 36 months of their own approval date. That fixed age is the whole point: it is what lets one year be compared with another, because every cohort is being asked the same question at the same point in its life. Where a cohort is too young for the window to have closed, the rate column is deliberately empty rather than optimistic — the loans that would fail have not run out of time to do it. Read the dollar rate beside the loan rate when the question is exposure rather than incidence: the two diverge because the loans that fail are not the big ones.
A row here is a whole year's worth of lending, not a place. Cutting a default rate to a radius would compute it over a handful of loans, which is arithmetic rather than a rate. For the loans themselves around one address, use the loan-by-loan file.
The file
24 cohorts · 7,477 disbursed loans · charge-offs within 36 months of approval · SBA's file as of 2026-06-30
24 rowsCounties in this file: Anoka, Carver, Dakota, Hennepin, Ramsey, Scott, Washington · 15 columns available · free, no account, no rate limit.
What the rows look like
10 real rows from this file, spread evenly across it rather than taken off the top — every file here is written in some order, so its first rows are never a fair sample of it. These are the default columns, which is what the download button hands over.
| programme | approval FY | 36-month window | disbursed loans | charged off in window | default rate % | approved $ | default rate % of $ |
|---|---|---|---|---|---|---|---|
| 504 | 2,026 | open | 10 | 0 | — | 4,485,000 | — |
| 504 | 2,023 | partly closed | 96 | 0 | — | 89,223,000 | — |
| 504 | 2,021 | closed | 184 | 0 | 0.0 | 138,181,000 | 0.0 |
| 504 | 2,018 | closed | 112 | 0 | 0.0 | 77,699,000 | 0.0 |
| 504 | 2,016 | closed | 125 | 0 | 0.0 | 91,920,000 | 0.0 |
| 504 | 2,013 | closed | 131 | 0 | 0.0 | 73,396,000 | 0.0 |
| 504 | 2,011 | closed | 124 | 0 | 0.0 | 74,176,000 | 0.0 |
| 7(a) | 2,025 | open | 903 | 2 | — | 404,955,600 | — |
| 7(a) | 2,023 | partly closed | 957 | 26 | — | 409,731,000 | — |
| 7(a) | 2,020 | closed | 713 | 2 | 0.28 | 315,691,700 | 0.07 |
An em dash means the record is blank, and that is the source's gap rather than ours. Across the whole file: default rate % is blank on 33.3% of rows, default rate % of $ is blank on 33.3% of rows.
The cuts of this file
Every cut is the same columns, filtered. Pick one and the row count changes; the file does not.
Every column in this file
Columns marked default come down unless you pick your own with
&columns=.
| Column | What it is | In the default file |
|---|---|---|
program |
programme | default |
approval_fy |
approval FY | default |
window |
36-month window | default |
loans_disbursed |
disbursed loans | default |
charged_off_within_window |
charged off in window | default |
default_rate_pct |
default rate % | default |
approved_dollars |
approved $ | default |
dollar_default_rate_pct |
default rate % of $ | default |
rate_so_far_pct |
charged off so far % (a floor) | optional |
window_closed_pct |
% of the cohort past the window | optional |
charged_off_dollars |
charged off $ | optional |
charged_off_ever |
charged off at any age | optional |
origination_block |
origination block | optional |
window_closes_on |
window closes on | optional |
window_months |
window (months) | optional |
Where it comes from
U.S. Small Business Administration, 7(a) & 504 FOIA release
Records retrieved 2026-08-20 — the oldest capture date among the layers behind this file, because data is no fresher than its stalest input.
What this layer cannot answer is stated in full on the system card — read it before quoting a figure.
What you may do with it
Creative Commons Attribution 4.0 International (CC BY 4.0). Share it, change it, build a product on it, sell that product — all fine. The one condition is credit: say it came from Brick & Mortar and link back.
Records published by U.S. Small Business Administration. A work of the United States government, which is not subject to copyright protection in the U.S. (17 U.S.C. § 105).
Read the full licence · how we license every dataset, and the two we do not
Without the page
/api/export?dataset=sba-cohorts&format=csv returns the
same file. Call /api/export
with no arguments for the whole catalogue.
Point an AI at it
Every file carries its own source, its capture date and the layer's own warnings in the header, so a model reading one gets the caveats with the numbers rather than the numbers alone. Paste this at an assistant that can fetch a URL:
Read https://brickandmortar.dev/api/export?dataset=sba-cohorts&format=csv
Everything above the header row is provenance and limits. Read it first,
then answer using only what the rows support.llms.txt describes the whole warehouse the same way — what each dataset holds, which counties it really covers, and what it cannot answer.