From Page 110 of the report (on my PDF copy there are no page numbers; p 110 is the PDF location)
The CVR indicates that the Commander (PNF) took control of the power levers during the final approach, this action being acknowledged by the PF. This was significant, as both power levers were subsequently retarded below Flight Idle — an action which would have been unexpected by the PF.
The recorded data shows that the No. 1 engine reached a minimum torque value of -9% in Beta range, while No. 2 engine reached a minimum value of 0%. This thrust asymmetry was coincident with the aircraft commencing a roll to the left (maximum recorded value 40 degrees of bank). It is possible that the PF may have made a control wheel input to the right in response to the unanticipated left roll. However, without the FDR parameters of control wheel or control surface position the Investigation cannot determine if such input was made. The subsequent application of power to commence the go-around, at approximately 100 feet, coincided with the commencement of a rapid roll to the right and loss of control. This roll continued through the vertical, the right wingtip struck the runway and the aircraft inverted.
So although they had corrected from the known torque difference, at this point the asymmetry, due to the undetected thrust fault I previously mentioned was nine percent, sufficient to produce the 20 degree bank to the left.