One request, several credits, and a count that can be predicted

A split is not a fault and it is not a fee. It is the loop working: each rail may return only its own cap, so a request large enough to exceed one cap becomes one payment per rail it touches. The count is arithmetic.

Rail stub

One request: one payment per rail touched
Why: no rail may exceed its cap
On the sample book: 1.40 payments on average
Not a failure: unless an amount is missing

the railA payment method that has carried a completed deposit - not a method saved on the account

the capWhat a rail paid in, less what it has already returned, and never below zero

released moneyThe part of a balance above every cap: payable, and travelling by the operator’s own route

Direct answerA withdrawal is split because each rail has its own cap and no rail may return more than it took in. A request that exceeds one cap is paid in as many payments as there are rails needed to meet it - so the count is decided by the deposit mix, not by the amount alone.

The count, computed

Three numbers produce the number of payments: the request, the rails in the order they are tried, and their caps. If the request is smaller than the first rail’s cap it is one payment. If it exceeds the largest cap it is at least two, and it keeps growing as the request reaches further down the order.

Worked: a 500.00 request across three railsrails and caps card 100.00, bank 240.00, wallet 160.00 total 500.00 order oldest first (wallet, then bank, then card on the sample) requested = 500.00 wallet 160.00 of cap -> pays 160.00, cap -> 0.00, 340.00 still needed bank 240.00 of cap -> pays 240.00, cap -> 0.00, 100.00 still needed card 100.00 of cap -> pays 100.00, cap -> 0.00, 0.00 still needed payments 3 average payment 500.00 / 3 = 166.67 released 0.00 every rail spent exactly to its cap

The same 500.00 against a single 600.00 rail would be one payment. The amount did not change; the number of rails did, and with it the number of credits, the number of settlement clocks and the number of days on which money arrives. 166.67 is the average payment in the worked case and it is a useful sanity check: a split whose largest credit is smaller than the smallest cap has something else happening in it.

The desk’s sample book

The rates below come from the sixth sample - 12,000 withdrawal requests against a book whose deposits are spread across rails. They are invented, and they are included because they give the split rates a size, which a single account cannot.

How many rails a request touches

8,160 requests used one rail (68.0%), 2,880 used two (24.0%) and 960 used three or more (8.0%). Two-thirds of withdrawals are a single payment.

Payments per request

8,160 x 1 + 2,880 x 2 + 960 x 3 = 16,800 payments for 12,000 requests, or 1.40 on average. A third of all requests are split (32.0%).

The rails that were gone

744 requests, or 6.2%, involved a rail that no longer existed, and 408, or 3.4%, met a name mismatch. Those two are the exceptions rather than the norm.

And the average amount

3,220,800.00 withdrawn across the book gives an average request of 268.40 and an average payment of 191.71. A split is a real inconvenience at these sizes, not a rounding artefact.

A split and a failure look alike for about a day. Two credits and a missing third are indistinguishable from two credits and a third that has not settled. The check that separates them is the cap arithmetic: if the caps in the order account for the whole request, the payments are complete and one is simply slower. If they do not, an amount is unaccounted for and the missing payment is a real question - and the first two credits, added to the caps already spent, are the evidence.