Summary: 1. Don't read too much into the physics of a simulated flight. 2. The accel-based altitude normally stops being more accurate than the baro-based altitude early in the flight. 3. The displayed measurements in the brackets were calculated in post-processing the recorded data, rather than on-board, and so were spoofed by a condition that's specific to performing the simulated flight with the altimeter upside-down.
For the gory details of 1&3: There are a few things going on here. The effect of the test mode in this case is complicated by the fact that the Raven was upside down when the simulated flight took place. That made a constant reading of -5Gs, so the Raven figured it was upside down and changed the sign of its acceleration. It also adjusted its offset to compensate for what looked like an offset error. Then when the test mode was turned off, that made an apparent 4 Gs, and it interpreted that as a liftoff type of acceleration, but it took awhile for the velocity to catch up to where it was positive, and in the meantime the original G reading was not recorded. Thus the post-processing sees the initial G reading of about 4 Gs and assumes that's an offset to be subtracted from the rest of the measurements. That threw off the post-processed velocity and altitude. The on-board velocity estimate "Velocity (Accel-Ft/sec)" a long averaging period before liftoff detection to get a better estimate of the prelaunch accelerometer readings, and you can see that it performed fine.
The discussion behind summary point #2 is probably more interesting:
The accel-based apogee altitude is very sensitive to nonlinearity and resolution errors in the accelerometer and the microcontroller A/D converter. Not to mention flight angles and coning that can also contribute to errors in the accel-based altitude error. So the accel-based altitude really should be ignored after the first part of the flight. So why display it at all? It's actually pretty good early in the flight, particularly during the initial boost when shifting parachutes, accelerating air columns, high angles of attack and Mach transients can make the initial baro readings pretty useless. In fact, you can often measure the length of your launch guide rail or tower by zooming in and looking at the lateral accelerometer to see when the launch rail stops bouncing the rocket around, and comparing that to the accel-based altitude at that time. For example, look at the NSL kraken accel data in post #69, and you can see that the rail was about 6 feet long. The Oh-S-Kraken accel data in post #74 shows the rail was about 10 feet long. If you want to tell people at what altitude your rocket broke the speed of sound, the accel-based altitude is the one to look at.