My '17 is at the Mazda dealer now for the exact same symptom, most reproducible on a u-turn or right turn. Just a slow speed-high load scenario. Good throttle response above 40mph or so,
As I use Forscan to collect data, and happen to have a loaner '19 Mazda 3, 2.5L NA to compare it to, the issue is either the MAF is under-reporting, or there something else triggering the Miller cycle mode, which greatly reduces power until the PCM corrects. In my case the CX-5 takes 3 to 4 seconds before the PCM corrects. The low power condition is accompanied by a hollow waaah sound, which is caused, as I understand it, by the greatly advanced (early) intake valve timing.
The issue seems related to a generally lean running condition, most apparent when comparing WOT (Wide Open Throttle) readings. For everyone's reference, here's what the 2.5NA is supposed to do, under WOT from a rolling start.:
Most lines a re self-explanatory. The ones that may not be:
ECT-ACT = actual throttle plate opening, in degrees
APP = accelerator pedal apply in %
SHRTFT1 = short term fuel trim in % above/below desired. pos % = detecting lean increasing fuel. neg % = detecting rich, decreasing fuel
LONGFT1 = long term fuel trim %. pos % = detecting lean increasing fuel. neg % = detecting rich, decreasing fuel
S11_Cur = Bank 1 Sensor 1 upstream (before cat) A/F sensor current output. neg mA = rich, pos mA = lean
O2S12 = Bank 1 Sensor 2 downstream (aftercat) voltage output. 1v= rich, 0v = lean