What counts as a position, and what only looks like one
Every MCA file eventually comes down to a number: how many positions are on this merchant. It gets quoted like a fact, and it is a verdict — the output of a rule about which repeating debits are an advance being repaid and which are a business paying its bills. The rule is where the argument actually lives, so here is ours: the evidence that makes a stream a funder, the order in which a position is typed, and the third answer we keep for streams that have not earned either label.
- Fixed amounts or varying ones — never credit for both
- The registry outranks the model; cadence decides last
- A stream that has not earned a label stays a candidate
A count is only as good as its definition
Two mistakes are available, and they are not symmetric in how they feel. Calling a vendor a funder adds a position that does not exist, inflates the burden, and kills a deal that was fundable. Missing a funder does the reverse, and nobody finds out until the merchant is servicing one more advance than their cash flow can carry. The first mistake is visible immediately and argued about; the second shows up months later as a default.
Both come from the same place: a definition that is doing less work than the number it produces. A payment that repeats is not evidence of financing. Rent repeats. Payroll repeats. A software bill repeats, on the same day, to the cent, more reliably than most advances.
What repayment actually looks like in a debit column
An advance being repaid has a shape. The same counterparty appears again and again at short intervals, and the amounts follow one of two patterns: they are fixed, because the contract set a daily or weekly debit, or they move, because the funder is taking a percentage of what the merchant took in. Those two are alternatives. A stream is one or the other, never both.
That last sentence is not pedantry. When the test for “fixed” and the test for “varying” overlapped even slightly, a dollar of jitter on an otherwise-fixed weekly payer collected the credit for both patterns at once, and the combined score was enough to promote an unknown counterparty into a confirmed advance. One dollar of noise, one extra position on the file, one declined merchant. The two tests are now strictly complementary, which is the sort of fix that changes no headline and quietly stops a category of wrong answers.
What a stream has to show
No single signal decides. They are weighed together, and the weights say which ones carry an argument on their own.
Identity — who is being paid
A name on the recognised funder list, a funder token inside an electronic descriptor, or a descriptor field that names the counterparty behind the rail. This is the heaviest signal, because it is the only one that is about the party rather than the pattern.
Rhythm — how the payments sit in the week
A daily-like cadence landing on at least three separate days, or a weekly cadence with a fixed amount. Cadence is measured per calendar day rather than per transaction, so a funder that pulls twice on one day is not mistaken for a daily payer — and the monthly projection is not doubled by it.
Repetition — how many times in the month
Four or more payments inside thirty days. On its own this is weak evidence, which is exactly what its weight reflects: plenty of ordinary suppliers clear that bar.
The shape of the amounts
Amounts that move across a narrow band read as a holdback percentage of receipts. Amounts fixed to the cent read as a contracted debit. Either supports the case; neither counts twice; and cents plus an electronic-payment token, alongside a funder name, add a little more.
Who gets to say whether it is an advance or a loan
Knowing a stream is financing is only half of it. An advance and a term loan behave differently, are underwritten differently, and belong in different places in a report — the stacking count is a count of advances, not of every company that lends money.
The type is settled in a fixed order of authority. The recognised funder list decides first, because a name we have confirmed is better evidence than any inference from behaviour. The payee classifier comes next, then what was read from the statement line itself. Cadence is the last tier, not the first: monthly spacing on an otherwise unknown financing payer reads as term-loan servicing, and a fortnightly rhythm is a genuine coin flip between a marketplace holdback and a loan, so it is not guessed at all.
There is a guardrail on the top of that order, and it exists because of something that actually happened. When the registry says one product type and the stream’s own behaviour points firmly at the other, the registry still wins the label — but the confidence is capped so the position surfaces for review. A single mistyped row in a list of funders must not be able to invert a merchant’s burden while presenting itself as certain.
The things that repeat beautifully and are not financing
Some payees are excluded from funder detection outright, however convincing their rhythm: tax collections, credit-card balance payments, payroll runs, retail and point-of-sale activity, card-processor fees, personal payment apps, and consumer lenders. These are the false positives that keyword-driven tools produce most often, and they are the ones that do the most damage, because a merchant with a healthy card processor and a disciplined tax schedule can be made to look stacked by nothing but good habits.
One exception is written into that rule on purpose. Capital products sold inside a payments platform look like processor traffic and are a real advance on the debit side, so they are pulled back out of the exclusion rather than dismissed with the processor they are named after.
Candidate is a verdict, not a failure
A stream that carries some of the evidence and not enough of it is not forced into a class. It is marked a candidate: shown in the report, listed with what it does and does not have, and kept out of the stacking count until something settles it. The count stays what it claims to be — distinct concurrent advances — rather than quietly absorbing every debit that looked financial.
This is the same instinct as the rest of the report. An underwriter can look at a candidate and know in ten seconds what the software could not: that the payee is their cousin’s equipment lease, or that it is the funder from the deal that never made it into the submission. What they cannot do is unpick a confident number built on a guess. So the guess is not made, and the evidence is handed over instead.
Common questions
How are MCA positions detected in a bank statement?
By grouping debits with the same counterparty and weighing four things: who is being paid, how the payments sit in the week, how many there are, and whether the amounts are fixed or move with receipts. Identity carries the most weight; repetition alone carries the least.
What does it mean when a position is marked a candidate?
That the stream has some of the evidence of financing but not enough to be typed as an advance or a loan. It stays visible in the report and outside the stacking count, so an underwriter can settle it with context the statement does not contain.
Do term loans count towards the stacking number?
No. Loans are surfaced in their own right and counted as debt service in the burden math, but the stacking count is a count of concurrent merchant cash advances — which is the thing it is used to mean.
Why would a well-known funder show low confidence?
Usually because the recognised list and the stream’s own behaviour disagree about the product type. The list still decides the label, but the confidence is capped so the position is reviewed rather than trusted — a single wrong entry should not be able to rewrite a merchant’s burden silently.
What if a position is wrong?
Retag the transactions and recalculate: revenue, burden, the position list and the risk summary are all re-derived from the corrected rows. Confirmed corrections also make the recognised-funder list better for the files that follow.
Related
Count the positions from the transactions
Upload a statement and see every stream, what typed it, and what stayed a candidate.