the curved track cannot be flown with autopilot in HDG mode as magnetic variation is not consistent (maybe with any lateral mode off?)
The report analysis does not require a 'curved' track, that is to say a continuous turn. As I noted, the rate of change of heading (averaged over each leg) is not constant so each leg just averages to a heading. Small heading changes make little impact on the average speed over that leg so do not invalid the analysis.
All the way through the AMSA/JACC maps, the possible tracks have been labelled with speeds, implying an assumption of constant speed (but not heading).
So, the following speculation (and I know nothing about B777s so this bit could be gibberish).
a. The analysis (after the turn) is based on the autothrottle being engaged, but at an unknown speed.
b. There is no heading autopilot maintenance mode engaged and the aircraft heading is wandering left and right, but with a general trend to the left which reduces with time during the flight.