Escuan, it's partly logic, partly trial and error. I had a bit of trouble reducing from 44 blocked cells to 40 because I wasn't sure one resulting entry was an allowable Chambers word, not having the latest Chambers to hand and not finding it on the online Chambers.
I congratulate the setter on some excellent grid construction. It's not easy to achieve a series of changes that all produce real words. However there were aspects of the puzzle I didn't like. The preamble was distinctly unhelpful in some respects. I cannot see why solvers weren't told that some of the samples of wordplay were deficient by more than one letter, in one case five. Nor were solvers told whether the final step (drawing lines) left real words. Since two-letter words are an anomaly in a Listener grid, I found the lack of information about that a bit annoying.
I wasn't all that impressed by the clues, either. A few good ones, some awful ones. Rather too many had wordplay involving proper names, sometimes curtailed.