Availability statistics — methodology

The figures themselves are published at /xs2a/statistics, with a CSV download. This page explains how they are produced and how to read them.

Under RTS 2018/389 Art. 32(4) we publish quarterly statistics on the availability and performance of this interface. They appear at https://platform.paysaxas.com/xs2a/statistics and are archived there.

What we measure, and against what

RTS Art. 32 sets no absolute uptime target. The test is parity: the dedicated interface must perform no worse than the channel PaySaxas customers use directly. So every figure we publish for XS2A is published beside the same figure for the customer portal, measured the same way, on the same schedule.

Availability — a probe every minute, against both channels. An interface counts as unavailable after five consecutive failed probes (Art. 32(2)), and stays so until a probe succeeds.
Performance — response-time percentiles (p50, p90, p99) per service: account information, payment initiation, funds confirmation.
Error rate — server failures only. A refused request is the interface working correctly, and counting refusals would flatter a broken interface while punishing a well-defended one.

How to read a blank

A period we did not measure enough of is published as blank, with the number of days actually measured beside it. It is never rendered as 100%.

“Enough” is a day, not the quarter: a day whose probes covered too little of it is excluded from the figure rather than averaged into it, so a blank does not only mean “we measured nothing at all” — it can also mean the days we did measure were too few to support a claim about the period. Either way the days measured are shown, so you can see which.

That is deliberate. “We did not measure” is a statement a supervisor — or you — can act on. A fabricated availability figure is not, and a thinly sampled day extrapolated to twenty-four hours is a fabrication with arithmetic in front of it.

Incidents

Interface problems are reported to FIN-FSA under Art. 33(3). Where an incident affects providers, it is also posted to the status channel listed in support.md.