The problem is probably due to the eccentricity being zero. This is due to my code having trouble converging for orbits with small eccentricities.
Does this TLE work with gpredict?
1 70002U 18599A 18315.16151858 .00000000 00000-0 00000-0 0 07
2 70002 85.1205 90.1568 0012705 292.5520 107.9249 15.20792276 04
This TLE was made by having GMAT propagate the orbit for 12000 seconds and then using the J2000 XYZ positions to fit an orbit to them. This approach is a bit elaborate, but tools from my sattools repository are tried and tested.
As for a delayed launch, you can use launchtle tool from sattools. A git clone and make launchtle should work without having to install all the sattools dependencies. This tool takes an input TLE and launchtime and then corrects the epoch of the TLE for the new launch time, and adjusts the right ascension of the ascending node for how much it would have changed between the old and new launch time.
You’d use it like this: launchtle -c irvine.txt -i 70002 -t 2018-11-11T03:00:00 -T 2018-11-11T05:00:00 -I 70003 to use the 70002 TLE in irvine.txt, which has a 2018-11-11T03:00:00 launch time, to correct it for a new launch time at 2018-11-11T05:00:00 and store it with a new identifier (70003 in this case).