I've rarely seen directly working on tech debt actually pay off. You either get stuck fixing last year's problem, or you lose the race to feature parity with where you were.

To that end, keep cleaning as you go. And accept that good development can be like a good batter. Probabilistically good. Not certain.