Our verified-dse programme searches chip-flow settings and asks whether a formal equivalence proof can be afforded for each candidate. Earlier we found that tool upgrades reshuffle routed rankings a lot. Does the same happen to proofs?
We reran the 15-design proof table (SAT miter, 10 s timeout) with Yosys 0.70 instead of 0.33. Result: no counter-examples, and "proved within 3 s" agrees on 750 of 751 points. Time is set by the design (4 to 6-bit multipliers prove in under half a second; 10-bit ones never finish in 10 s), not by the version. Netlists differ by a few cells between versions, yet the proof outcomes do not.
Caveats: same designs, same timeout, one solver family. The numbers and script are in research/verified-dse/experiments/proof_success/ (compare_y070.txt); the log entry is milestone 52.