Every decision NASA makes is a political one. Not L-R politics, but there are at least four positions that are appointed and need Senate confirmation, and they have to consider Congressional reaction in every decision because the House is in charge of the budget.
Killing two astronauts when you know you have problems you can't understand or replicate is going to kill the program in Congress, regardless of how badly they want Starliner.
This decision covers their tail politically. Boeing maintains Starliner is safe for reentry, and they'll get a chance to prove it. If it fails to deorbit or burns up, the program can still survive. Failure will hardly make a blip right now on the news cycle or Boeing's reputation. Success obviously won't hurt the program, and saying, "Better safe than sorry with our astronauts lives" will always be an answer that diffused questions.
Assuming the program survives, they should do the next flight unmanned, bring up some cargo. Then before they deorbit, deliberately abuse the thrusters and see if they can get a handle on when/why the thrusters are failing since they cannot reproduce the problem on the ground. And see how many you can lose and still maintain a safe deorbit burn and maintain attitude. Even if they cannot get a handle on "why", if they get enough data on "when", it'd make future green lights a lot more likely. And if your goal is to abuse it first, then no-one cares if that one comes back safely or not. You're just gathering data.
Bottom-line is no-one likes "We're not quite sure why, but we're extremely confident it won't be a problem" after 5/28 thrusters experience failure on your first crewed flight to ISS. It'd be nice to know more about the helium leaks if you plan on being up there for several months in the future, too, but the priority now should be nailing down why your thrusters are overheating and unpredictable. If you can't reproduce it on the ground, you need to go abuse it in space and make it happen. Figure out the limits even if you can't figure out the cause precisely.