The simplest solution is for the early exit penalty not to decay to zero. With a minimum penalty on every early exit, all of the design problems appear to disappear.
The claim clamp starts at 40% for a 28 day commitment and decreases linearly to 0% at the maximum lock duration. An infinite lock can claim 100%. The early exit penalty starts at 75% when the entire lock remains and decreases to a minimum of 25% just before maturity. It applies to the amount remaining after the claim clamp. The penalty never falls below 25% before maturity, and only reaches 0% once the lock has fully matured.
The key invariant is that a user who chooses a longer lock and exits after X days must always receive less DUST than a user who committed to exactly X days and completed that commitment. For periods shorter than 28 days, the comparison is against the instant claim, which returns 50%.
I modeled every whole day commitment from 28 to 364 days against every possible whole day early exit, covering 66,052 scenarios in total. There were zero invariant violations. Even the tightest case maintained a 5.01% margin in gross DUST. Voting power and revenue sharing are not included in this comparison, which further strengthens the case for keeping a nonzero minimum early exit penalty.
Make a copy and play.