The key to understand failures is to understand the non-ideal characteristics of your layout and each component against assumed V:I(t) response. This includes primary core saturation, wire inductance, inductive and capacitive stray coupling, effect of Common Mode noise on differential induced noise, and the benefits of CM chokes. (ferrite beads around conduction pairs)
All high current paths must become balanced by twisting the DC in with return together. Examine the current loop of your layout and minimize the area of the loop ASAP.
In order to monitor power consumption you need a reliable DC current shunt < the RdsOn of your drivers (6 mOhm) I would suggest 1 mOhm ground side, low-ESL, current shunt using magnet wire. To make a non-inductive wire shunt, I use magnet wire , folded twisted so the ends are together then the inductance cancels out. Then add a connector to the board to your common ground point ( BNC bulkhead receptacle) so you can attach 50 Ohm coax from board to scope and terminate the coax with 50 Ohms at the scope using a BNC T connector ( if not avail. in scope) This will give the cleanest DC current sensing possible to monitor overcurrent conditions. ( YOu can build an OCP circuit to automatically shutdown and protect your expensive parts. ( Over current protection)
Since AWG 20 is 33 mOhm/m to make 1 mOhm you need 30 mm of AWG20 magnet wire or thin insulated wire, folded in half, twisted 5 times or more and connected close to your star ground system to your supply return. If you can use a lab supply to put in a calibrated 10A CC limit or any low voltage , you can verify/calibrate the mOhm shunt with a DC meter.
From this point on it is important to view every high speed LdI/dt wire as a potential voltage drop and examine why the more expensive drivers that are 10x faster response work better in order to make the slower cheaper ones work safely without cross-conduction MOSFET failures.
Understand what makes them slower and how that affects MOSFET power dissipation.
How can you speed the gate drive transition to protect the MOSFETs by switching Vgs faster?
How does the dead-time change and how do you control it?