I found it!!!!
https://gpladdons.the-fastlane.co.uk/utilities_list.php?q=(ID~equals~11)
As expected, the LPs are slightly different, but this program is able to read most (all?) of the LP file. As you can see, the yaw stopped being recorded in the making of the LPs between NR4 and NR2003 (likely as part of the change that involved moving the slipcurve from pap_ai.ini to track.ini?). The speed seems slightly miscalculated, and interestingly vastly different between NR4 and 2003. The left is NR4 daytona. 183 is a believable speed. All NR2003 LPs speeds are much higher (NR2003 daytona right), so I need to look into calculating the difference.
Also importantly only changing the Latitude column affects the AI car. Setting all other columns to 0 results in no change to the AI behavior. This leads me to believe that the LP file structure changed enough from GPL to NR2003 such that the values this program is picking up on are not the actual values the game is using to calculate AI. What are they? I'm uncertain... I'm guessing the header information changed, resulting in an offset, resulting in the program reading the incorrect data into the columns.