Thanks LunaWei — here are the numbers from the build I described, so the state table is concrete rather than a general idea.
I ended up with five states, not two: fully open, about 135 degrees, 90 degrees, about 45 degrees, and fully closed. I tried three states first, open, half and closed, and it was not enough, because the worst mismatch sat between the half-fold state and the endpoints, where the ground length was changing fastest. Adding the 135 and 45 degree states caught those transition regions, and each state stores its own tuner setting, a return-loss target, and the bands it covers. I did not add a sixth state, because the extra state bought very little once the 45 and 135 degree points were in, and it made the switching logic harder to validate.
For hysteresis, I use about 5 degrees of hinge angle and about 2 dB of return loss as the guard band. The tuner only switches state when the angle has moved past the boundary by that 5 degrees and stayed there for a short dwell, and it only switches back when the angle has returned past the boundary by the same margin. Without that, the tuner chattered near a boundary when the user held the phone near half-fold, and every switch produced a brief mismatch spike. With the guard band, the switching is stable, and I also freeze the tuner during the actual hinge movement and only apply the new state once the angle sensor reports a settled position.
On TRP, the worst point in my measurements was not at the endpoints. It was around 90 degrees, and more broadly in the 70 to 100 degree range, where the two halves of the phone form a poor ground and the antenna efficiency drops at the same time as the match degrades. At that angle I measured about 2 to 3 dB lower TRP than at fully open, even after the tuner had settled on the correct state. At fully closed the TRP was also lower, but it was more predictable, because the state was stable and the tuner setting was fixed. The mid-fold region is the one to test carefully, because that is where a user actually holds the phone while unfolding it, and a short TRP dip there is easy to miss if you only measure the two end states.