Where every number on this site comes from

One source, read once a week. Every figure recomputed from the published numbers. Every check shown with its count. No verification, and no badge pretending otherwise. This page is the ledger behind the four commitments — the place to catch us if a number here says more than the data behind it supports.

Every figure on Main Street Diligence comes from the 2,079 businesses actively listed on the Business Brokers of Florida MLS on September 14, 2026, read once a week from the public search page, recomputed from the published numbers, and checked against themselves. 124 of them (6%) fail at least one arithmetic check, and every failure is printed on the listing. The site’s own checks were re-run over all 2,079 of them when this page was built and graded 2,079 of 2,079 the same way the ranking does. Counted from the crawl of September 14, 2026; recomputed on every build, never typed in. Nothing on this page is verified against a document, because nothing on this site is.

One source

Every listing on this site is an active business-for-sale listing on the Business Brokers of Florida MLS, published at bizmls.com on a public search page that needs no account and no login. We do not accept listings, we do not take a feed from anyone, and no broker can pay to appear, to rank, or to carry a badge. Florida only, by design: we would rather be able to defend every figure in one state than reprint feeds from fifty.

What is taken from each listing is what the broker published: the asking price, the down payment, the discretionary earnings, the revenue, the category, the location, the broker’s own description, and — from the listing’s detail page — up to three fiscal years of figures with the label the broker put on each one, the asset schedule, seller-financing terms and owner hours. Every listing here carries its BizMLS number and a link back to the source, and the description is quoted as written and attributed, never rewritten. What we add is arithmetic on top of those figures, and the arithmetic is the whole product.

How often it is re-read

Every Monday at 07:00 Eastern the crawler sends one request to the public search page, which returns the whole BBF inventory, and two hours later reads each active listing’s detail page one at a time with a pause between requests. It identifies itself by name with a contact address on every request, honours the site’s robots file, never rotates addresses or disguises itself, and has a single stop switch that halts every crawl before its first request. The full policy, including what happens if the source ever asks us to stop, is written down in the repository as docs/crawl-policy.md.

A listing that leaves the source index is marked inactive here, not deleted, and every change to a listing’s price, down payment, earnings, revenue or status is recorded in a history table from the crawl that saw it. That is what lets the site say how long something has been listed and when its price was cut — and it can only accumulate forward, which is why the crawl is weekly rather than monthly.

The date of the crawl each figure describes is printed in the footer of every page of the app, next to the number of active listings it covers. If that date is more than eight days old the footer turns amber, because a figure that has quietly aged is worse than one that says so.

What is computed, and from what

Nothing on a card is reprinted from the listing except the four figures the broker published. Everything else is recomputed here from those four, so that a reader can check it:

• SDE ÷ price, the payback period and the earnings multiple are the published earnings and the published price, divided. They are not returns — they are two seller figures in a ratio, before the owner’s labour, financing or tax — and the SDE page says why the site stopped calling the first one “ROI”.
• The Overall Score is a percentile rank across the active inventory: 70% on SDE ÷ price, 30% on revenue to investment. It ranks listings against each other and says nothing about value. It is shown as a band (Top 5%, Top half) because two decimals of a percentile imply a precision the inputs do not have.
• Medians, never averages, for every headline figure, with placeholder prices under $10,000 excluded — a “call for price” listing carries a token figure and a $1 price inside a median is a claim about the market rather than a measurement of it. The definitions are checked into the repository as SQL so they can be read.
• The lender’s screen deducts a standardized $75,000 manager’s wage from the most recent published year, then asks whether debt at 90% of the asking price would come to more than 3.5× what is left. Part of that carries the SBA’s name and part of it is ours, and the difference matters: the eligibility rules and the 10%-down minimum equity injection are written down in 13 CFR and the SOP, and every badge cites the one it was graded under, but the 3.5× ceiling is ours — a deliberately conservative first-pass leverage cap, not an SBA threshold and not an underwriting decision, which turns on debt-service coverage at a lender’s own rate and term. The $75,000 is a convention too, not the market rate for a manager in any particular trade. Neither number is load-bearing on its own. Hold the wage at $75,000 and move the ceiling from 3.0× to 4.0× and the number of listings that pass runs from 262 (13%) to 473 (23%), against 375 (18%) at the setting the site publishes. The headline is a reading taken at one standardized setting, not a property of the market, and the setting is the first thing to argue with. Eligibility defaults to the SOP 50 10 8.1 rules a deal starting today would close under; the book in force until 1 October 2026 is SOP 50 10 8, and the sidebar’s “Underwrite to” control grades every listing to it. How it works walks through it; the research has the rule change in detail.

The checks, with today’s counts

Every active listing is run through the checks below, in the browser, every time it is drawn. The outcome is printed on the card as one of three grades, and the top grade is deliberately low: figures add up means the published numbers do not contradict each other, and its own tooltip says it tests arithmetic, not truth. A listing with invented but internally consistent figures earns it, and that is the correct behaviour, because the alternative is a badge implying a verification nobody in this market can perform. Across the 2,079 active listings when this page was built: 1,599 read figures add up, 356 read partly disclosed, and 124 read figures don’t add up.

Failures

Two published figures cannot both be right. Printed on the card in amber, and the listing ranks after every listing that passes.

Gaps

Something is not published. A broker’s disclosure decision, not a defect in the business — shown in grey and never used to move a rank.

Notes

True and worth knowing, not a problem.

Checks we measured and did not ship

Two more checks were written and measured before the verdict launched: earnings larger than revenue, and a down payment larger than the asking price. Both caught zero listings, so neither was added — a check that catches nothing would look like diligence while doing none. Re-measured on this build: earnings above revenue would catch 0 listings today, and a down payment above the asking price would catch 0. Both still catch nothing, so both stay out.

What we do not know, counted

A screening tool that hides its misses converts an unknown into a false negative you will never go looking for. So the gaps in our own coverage are counted here, from the same crawl:

• 2,079active listings in the index
• 52name no usable county on either page — blank, “All” or “Withheld”. They appear in every statewide figure and in no county figure, because a missing location is not evidence the business is elsewhere
• 356withhold at least one headline figure or a description, and so read partly disclosed
• 2have no readable detail page. The second crawl returned a bare template or nothing, so the multi-year figures, the evidence class and the cross-page checks do not exist for them. Shown as unknown, not as failing
• 290have a broker naming a document — a tax return or a P&L — behind at least one year. Named, never seen. Nobody here has received or read one

And the one thing no count can close: nothing here is verified. Every financial figure is seller-reported and unaudited. The checks establish that a listing does not contradict itself; two pages can be consistently wrong, and both come from the same side of the table. There is no “verified” badge on this site and there never will be, because we cannot perform the verification it would imply. The diligence — an accountant’s quality-of-earnings review, a lawyer’s read of the lease, a lender’s own underwriting — is yours. This is what comes before it.

If a figure here is wrong

It has happened, and each time the correction is on the site. An outside reviewer noticed that the dashboard and the market pages printed two different median prices for the same inventory; one applied the placeholder-price floor and one did not, and now both read from one definition. The same reviewer objected that a ratio of two seller figures was labelled “ROI”; the cell was relabelled and the reason written up. Building this page found the third: 295 listings read “county withheld” on their card, with a note that they were invisible to every geographic filter, while sitting in the county filter all along — the broker had left the search-page location blank and named the county on the detail page, and the card only read the first. A sentence that says less than the data supports is the same defect as one that says more. A site that asks you to distrust overstated labels has to correct its own, in public, and keep the record.

Corrections, disagreements and questions go to info@mainstreetdiligence.com. If a number on this site says more than the data behind it supports, we would rather hear it than not.

Search Florida listings How the screening works The four commitments