BOAC
........ Most of us would have expected a departure, I'm sure.
Sorry to object here.
Imho with the term "departure" we adress the point, where the aircraft leaves the region of controlled flight (i.e. departs the flight regime).
There is no doubt, that the aircraft was outside this controlled flight region, was uncontrolled and therefore had departed. It stalled and departed with a high sinkrate and some minor roll oscilations.
Unaccelarated stall and departure like we had here (1g-stall) is less liable to violent maneuvers / g-loads / rolls/ oscilations and therefore hard to discover from the aircraft behaviour alone (without reference to the instruments or to the outside). First sign might be a gentle nose drop in a conventional aircraft, but not in AF447 case, as the FCPC´s where acting to maintain the demanded load factor. This stall has besides the AOA problem (AOA too high) also a speed problem (speed too low for 1g flight). To recover it is necessary to reduce the AOA and increase speed up to flying speed again.
Accelerated stalls (normal asociated with turning) are more sudden and more violent. The flow separation over the wings is not equally distributed and therefore yaw, roll and pitch changes in any random combination are induced. There is a primary AOA problem (g-induced too high AOA) but lesser or no speed problem (speed too low for acelerated flight, but not for 1g flight) and relaxing of g-load does the trick to solve the problem.
The entry into the stall of AF 447 was very smooth, as the speed decay below the stall speed occured in the timeframe, where the PF tried to correct his pitchup by some ND input (g load around 1 g). But the mass of the airframe in combination with the pitch upward vector smoothed the airframe close to the stall (second stall warning). It was initially no AOA problem, but a speed problem (running out of speed in the climb with not enough power applied). At that point speeds where already unreliable. The aplication of TOGA at that critical phase delayed the departure and held the AC in the pitch and in a little bit of climb (for further 500 feet) while further shedding vital speed. But finally the AC just departed by picking up a high descent rate and some roll oscilations.
The slowing during the climb might also have caused some SI (if you slow, you feel tilt forward) and induced doubts about the validity of the pitch information, as airspeed was unreliable already. That would also explain the missing ND inputs (as he felt already ND pitch) and the thinking of the PF at some point later, that they where fast and thus descending fast (not believing the pitch information), not falling / stalling downward fast.
All three pilots saw their problems obviously in the roll oscilations not in speed and AOA. Otherwise primary concern should have been the reduction of AOA and rolling wings level later on, when AOA reached an non stalled value.
Later on when it came to their mind that they are not in some high speed descent, that commanding climb with the SS did not work, they couldn´t get their mindset squared to "being stalled" and the apropriate recovery procedure.