The reducing is the XWS-15 and XWS-19, not serial production of WS-15 and WS-19. I use X to designate the test-production variant. These engines from 2020 to 2026 are NOT production articles, every batch of them are slightly different with later batches being closer to the final articles for mass productions. It is natural to expect that the divination being reduced, so the number of test articles being reduced in total. At the end of the phase, only the last batch (say 5 in 2026) would be identical to the version in mass production. Most early makes of these engines are not going to be used by J-20.
Let me illustrate in total numbers. In 2020 the total is 15, if 5 as a batch, there will be 3 batches. I will make 3 batches in sequence. The first batch results to deviation that only 2 is up to spec, I will improve my process and make 2nd batch which results 3 good, so by the 3rd batch I may get 4 good. That means out of 15 in 2020, I have improved my process to make a batch (5) of quality 80%. That is not good enough, so I start another batch of 5 in 2021. Do I still need 15? Probably not because now I have a much higher baseline. If everything goes smoothly, I may only need 2 batches to reach 5/5 where I can settle my process.
This is to illustrate how pre-production development are planned. How they spread out these numbers (size of batches and number of batches) are dependent on their R&D methodology, but the overall trend is gradual reduction of total articles until the last batch.