Skip to main content
Model, not forecast. Every input here is an assumption. Replace each figure with a real number before quoting any of it.One of them is no longer a guess: the insurance reserve line is now set by the reserve model rather than picked, which moved it from 0.20% to 0.29% and the net take rate from 0.70% to 0.61%. The payout spread was then set at 0.80% (from 0.50%), which puts the net take rate at 0.91%.

How it makes money

Four lines, in order of how soon they arrive:
1

Platform fee on settled volume

Charged to the merchant. Deliberately below typical card acceptance cost — that is the entire commercial argument.
2

Spread on the payout conversion

The gap between the rate the partner gives and the rate the merchant is quoted. Shared with the disbursement partner.
3

Instant settlement

A fee for instant rather than nightly payout.
4

Lending against observed cashflow

Later. This is where acquiring businesses actually make money, and it is unlocked by owning the cashflow record, not by charging more.
The same sequence Square walked.

The table

Thin per merchant and entirely normal for acquiring. The business is volume and retention, and the retention argument is specific: a merchant who has been paid on an afternoon the bank terminal was down does not go back.

The line the model set

The reserve line is the one figure here that is no longer a judgement call. The reserve model puts the line that expected loss actually implies at 28.07 bps, against the 20 the plan assumed. The rate charged rounds up to the next whole basis point:
Why up rather than to nearest. At 28 bps the line raises $3,192 against $3,200 of expected loss — it still does not cover, so rounding to nearest would have preserved the exact shortfall the model was built to find. A reserve that over-collects slightly is the survivable error.It is also why the constant is a whole number. INSURANCE_RESERVE_BPS is a bigint against BPS = 10_000n, so 28.07 has no representation at this scale, and 0.01 bp of precision on a figure resting on eleven unmeasured inputs would be false anyway.
The cost of the correction is 9 bps off the take rate — 0.61%, not 0.70% — and the stock now funds in 2.4 months rather than 3.5. Still outstanding, on the SKR premium:
Reserve relief funds a premium of about 1.001×, not the illustrative 1.5×.The premium may still be worth paying — out of the rebate budget as acquisition cost, or justified by Guardian yield accruing to the merchant. But it cannot be described as priced off capital relief, which is how the deck currently frames it.

How the ledger books it

@nelo/settle accrues all four deductions on every settled sale, in one transaction, because it is one economic event — the money arrived, the fee was earned, the reserve was funded and the rebate accrued at the same instant. Splitting it would let three of the four land and the fourth fail. On a $100 sale at the rates above:
The reserve and rebate are funded out of the platform’s own take, not charged on top of the merchant’s fee. After a full payout, what stays in custody is exactly the platform fee — and that fee has to cover the reserve and rebate promised out of it. At 50 bps fee against 29 + 10 bps promised, 11 bps is genuinely free — down from 20 before the reserve line was corrected. That relationship is a committed test.

What the rates are, in code

They live as named constants in one place so replacing one with a real number is a change somebody can find. The reserve line is the first that has actually been replaced.