The A350 and the B787 are the only passenger airliners that can be seen cruising as high as 43,000 ft on flightradar24 (with the exception of a B747 I once saw). However, the B787 is more often found cruising at 42,975 ft rather than at 43,000, unlike the A350 which is always seen cruising at exactly 43,000 ft. Sometimes the B787 flies at 43,000 ft but more often at 42,975 according to fr24. If so, why? If not, if it is actually cruising at 43,000 ft, does its portrayal on fr24 have to do with its transponder's position or something?
This is a guess based on superficial understanding of the systems involved, but I'll make it an answer as it seems to be all that's available unless an engineer who actually worked on the 787 autopilot chimes in, because it's details I don't expect to be documented anywhere public.
So: there are two systems involved in this, the autopilot and the ADS-B encoder.
The autopilot is a feedback-loop controller. Probably a variant of a PID controller or maybe two chained controllers, the altitude one commanding a either pitch or vertical speed one. These controllers take the altitude, vertical speed and pitch and calculate some formula that determines whether to wind the elevator up or down. The coefficients are estimated from design and then fine-tuned in test flight.
Depending on how exactly the tuning worked out, the aircraft can settle at altitude few feet below or above the target – either the vertical speed is reduced fast and the airplane converges to the target altitude from below, but remains a few feet lower for long, or the vertical speed is reduced a bit more slowly and the plane overshoots a couple of feet and very, very slowly comes back down – the plane is a bit over 50 ft high, so a couple of feet off is more then good enough either way.
The other part is the ADS-B¹. The 1090 ES message (which is extension of the mode S transponder) encodes altitude with precision of 25 ft only. So it has to round the value it receives from the altimeter. And if it rounds it down, a couple of feet below might end up being rounded to 25 below.
None of this matters in practice. The controller only sees the altitude rounded to 100 ft. Anything else would be just distracting. The planes are tens of feet high, and the accuracy is tens of feet too. As long as the system can keep the plane ±100 ft of the assigned altitude it is good enough.
Apparently the satellites listen for different kind of messages (the UAT, I suppose) with different precision, so the altitudes are rounded differently for the planes shown in blue.
¹ FlightRadar 24 relies purely on information broadcast by the aircraft themselves, but for altitude so does the ATC. Barometric altitude has the advantage that barometer accurate enough to distinguish tens of feet is a simple and reliable device and was already available in early 20th century. But it measures pressure, not altitude, and due to weather, the geometric altitude can be quite different. That does not matter for separation, since there can't be any abrupt changes in pressure (air being fluid), so we know aircraft in the same area are at the same altitude if and only if they are at the same isobar. But it means we need to know the pressure and the aircraft barometric altimeter is the only instrument that knows that. So ATC also looks at the altimeter reading the aircraft transmits with the transponder mode C or ADS-B.