The pair efficacy/effectiveness is separated entirely by the conditions of measurement, so anchor each number to its setting.
A randomized controlled trial is the gold-standard, tightly controlled environment: eligible volunteers, supervised dosing, intact cold chain. The protection observed there is the efficacy, and in this question that controlled figure is $90\%$.
When the vaccine then enters everyday programme use, real-world friction — partial coverage, missed boosters, storage lapses, a sicker and broader population — erodes some benefit. That field value, $85\%$, is the effectiveness, and it is expected to sit below the efficacy.
Hence the trial-based, idealized number is the efficacy. A useful memory hook: efficacy answers "can it work?" (best-case, in a trial), while effectiveness answers "does it work?" (everyday programme). Because the everyday answer is dragged down by missed doses, storage lapses and a broader population, effectiveness almost always trails efficacy, which is why 85% is below 90%. The 5% difference and the 87.5% mean are nonsense distractors with no biological meaning, and 85% specifically names the real-world effectiveness rather than efficacy.
\[\boxed{\text{Efficacy} = 90\%}\]