A chargeback arrives in its own time.

Fictional scenarios and synthetic data. Reviewed 6 November 2026.

Brightwater Tickets had its biggest month in August and its quietest in October. In the first week of November, Stilling Bank's letter shows the dispute ratio climbing for the second month running. The merchant's new fraud rule went live on 1 October, and the Brightwater team is certain it is working, because October's own disputes look tiny. Both readings are made from the wrong month.

What is actually being decided.

Whether a number that is going up means behaviour has changed, or only that time has passed. The merchant carries the decision about rules, refunds and pricing that will be made on the strength of the number. The PSP carries the duty to report it to the acquirer in the scheme's form, and the duty to explain it internally in a form that is actually true. The acquirer carries the scheme ratio as the scheme defines it, lag and all, and will not accept a cohort chart in place of it.

The lag itself is not controversial. In Visa’s dispute-resolution guidance, a cardholder raises the dispute with their issuer some time after the charge, the issuer raises it with the merchant, and the merchant then has a window to reply. Those steps need not happen on the day of sale. What the lifecycle does not give you is a distribution, and no two merchants have the same one. Ticketing disputes cluster around the event date, which may be months after purchase. Subscription disputes cluster around the renewal that was noticed. Do not assume casino-deposit disputes have a universal short deadline or lag. The first job is to measure your own.

Two ways to count the same disputes.

A cohort is the set of transactions that share a month of sale. The cohort view asks: of everything sold in August, how much has been disputed so far, and how much will be by the time the cohort is mature?

The calendar view asks: how many disputes arrived in October, divided by some month's sales? This is the view the scheme programmes take. As summarised by acquirer guidance, Visa's VAMP ratio divides the month's fraud reports and disputes by the same month's settled card-absent transactions; Mastercard's ECP divides this month's chargebacks by the previous month's transaction count. Two ratios, two denominators, one portfolio sets both out. Neither looks back further than one month, and neither cares which cohort a dispute came from.

The two views answer different questions. The cohort view tells you whether August's customers behaved differently from July's. The calendar view tells you whether the scheme is about to write to you. Report both; never let one stand in for the other.

August, measured at four ages.

In this illustration, Brightwater sold 60,000 tickets in August. By the time the cohort is fully mature, 300 of those sales will have been disputed, a cohort rate of 300 / 60,000 = 0.50 per cent. The disputes arrive on Brightwater's usual curve: 20 per cent in the month of sale, 50 per cent in the following month, 25 per cent in the month after that, and 5 per cent later still. In counts: 60 in August, 150 in September, 75 in October, 15 from November onwards. Check: 60 + 150 + 75 + 15 = 300.

Now read the August cohort at four points in time.

• Mid-September, with half of September's arrivals in: 60 + 75 = 135. 135 / 60,000 = 0.225 per cent.
• End of September: 60 + 150 = 210. 210 / 60,000 = 0.35 per cent.
• End of October: 210 + 75 = 285. 285 / 60,000 = 0.475 per cent.
• Mature: 300 / 60,000 = 0.50 per cent.

Observation: the August cohort rate more than doubles between mid-September and maturity, with no change in anything August's customers did. Inference: any comparison between August at four weeks old and July at twelve weeks old is a comparison between an immature number and a mature one, and it will flatter August in some cases. The Volume 01 note disputes and time made this point about a single queue; here it is the same point about a ratio.

Why October's ratio climbed while nothing changed.

Give September 30,000 sales and October 20,000, each with the same underlying 0.50 per cent cohort rate and the same arrival curve. September's cohort will eventually produce 150 disputes; October's will produce 100.

Count the disputes that arrive in October, by cohort:

• From August (month 2 for that cohort, 25 per cent of 300): 75.
• From September (month 1, 50 per cent of 150): 75.
• From October (month 0, 20 per cent of 100): 20.

October's arrivals: 75 + 75 + 20 = 170.

On a same-month denominator, 170 / 20,000 = 0.85 per cent. On a previous-month denominator in the ECP style, 170 / 210,000 × 10,000 = 57 basis points, 0.57 per cent. Every cohort in this illustration runs at 0.50 per cent. The calendar ratio for October is higher on either denominator, because August's large cohort is still sending its disputes into a month with far fewer sales to divide by.

That is the second month of climbing in Stilling Bank's letter. It is lag arithmetic. It is also real, in the sense that the scheme counts it exactly this way and the assessments follow the calendar, not the cohort. The alternative we keep alive: behaviour may also have changed. The test is the cohort view. If September's cohort at one month old (150 × 0.7 = 105 disputes so far, 0.35 per cent) matches August's at one month old (0.35 per cent), the behaviour is stable and the calendar climb is volume shape. If it does not, something happened in September.

And the rule that "worked" on 1 October: October's cohort, at the start of November, has seen 20 per cent of its eventual disputes. 20 / 20,000 = 0.10 per cent. August's cohort at the same age was 60 / 60,000, also 0.10 per cent. The rule has had no visible effect yet, and could not have. It will be a fair comparison at the end of January. The article on baselines and seasonality is the general case: compared with what, exactly.

A failure mode to watch.

Judging a change by the cohort it was applied to, a few weeks after applying it, against a mature cohort from before. The new rule, the new 3DS policy, the new checkout: all of them look excellent at thirty days because the disputes have not arrived. The team reports success, the budget moves on, and the ninety-day number lands in a quarter nobody is watching. The fix is dull: fix a comparison age, usually the point where your own curve has delivered most of its disputes, and do not report a cohort as comparable until it reaches it.

A way to work it.

1. Tag every dispute with the month of its sale as well as the month it arrived. If the data model cannot do this, that is the first ticket.
2. From at least six mature cohorts, build your arrival curve: the share of a cohort's eventual disputes that lands in month 0, 1, 2 and later.
3. Choose a comparison age at which most of the curve has arrived. Mark cohorts younger than that as immature in every report.
4. Report the calendar ratio in the scheme's own form, with the denominator the scheme uses, for the acquirer conversation.
5. When the calendar ratio moves, decompose the month's arrivals by cohort before explaining it. Volume shape first, behaviour second.
6. For any change you want to evaluate, write down today the date on which the comparison becomes fair. Put it in the calendar with an owner.

Take it to the desk.

When we last said a change had reduced disputes, how old was the cohort we were looking at, and what did the same cohort look like at ninety days?

Try it: A merchant's ratio heads toward the threshold and, from Volume 01, Dispute spike. Look up: cohort, chargeback ratio, baseline.

Fictional scenario. Synthetic data. No real customer records.

Edited 6 November 2026. Primary references are listed above; fictional scenarios are not claims about Antifrog customers or operating experience.