Call check-status and you get one string per country, describing this instant [3]. Nothing behind it is a time series, so "is Germany up" has an answer and "is Germany up at 02:00" does not [1][2]. The Commission publishes the present tense and keeps none of it, which makes it a smoke detector you have to stand next to. The sampler exists to fix that: 96 reads per state per day, every one retained [4].
The parsing rule is the part I would copy into my own client. Only the literal string `Available` counts as up, and every other value the endpoint returns, including values the author had not seen before, is counted as down rather than quietly discarded [6]. The other default, where an unrecognised status reads as healthy, converts a wording change at the Commission into a gap in your own availability data that nobody notices for a quarter.
The residuals are where the engineering is. 198 unavailable samples out of 15,344 is 1.29% [1]. Germany's 88 misses out of 548 samples is 16.06% [3], which is 12.4 times the fleet-wide rate [3]. Eighty-eight samples spaced fifteen minutes apart span 22 hours of the window [4]; Latvia's 61 span about 15 hours and Ireland's 25 about six [5]. The ten remaining states that missed anything sum to 24 misses, exactly 198 minus 174 [7], so the table closes against its own total.
For Germany's 83.94% to say anything about your billing flow, two conditions have to hold. Your calls have to land in the hours where the misses actually sit, which is why the per-country hour profiles in the report are the usable output and the percentage is the summary [19]. And the Commission's self-report has to agree with what a live `checkVat` does for your request, which the author does not claim: the status feed can disagree with a real call, and it can lag a member state going down [16].
The post is unusually honest about its own shelf life. A comment left in the source says every figure comes from a single script run and is to be replaced by a 30-day window, and it sets the rule that a claim absent from the script output does not belong in the post [18][20]. That rule is doing work already, because the sample count drifts from the label: 548 samples at 96 a day is 5.7 days, not five [8]. The author says five days establishes a pattern that repeats daily and declines to extrapolate an annual figure from it [17]. That is the right reading of a window this short, and it is more restraint than most vendor status pages manage.
The figure worth putting in a runbook, then, is not the availability percentage. It is the hour range per country where the misses cluster, because a job scheduler can act on that.
