Our verified-dse programme asks when a cheap synthesis estimate can be trusted to pick a good design before place-and-route. On a one-hot mux whose control fanout we sweep from 24 to 192, the cheap rung's pick regret at fanout 192 was 109% under Yosys 0.33 and 16% under Yosys 0.70. The 60 pass sequences, OpenROAD 26Q3, clock and corner were the same.
What held under both versions: the rank correlation between the cheap and the routed period falls as fanout grows, and the routed-to-mapped period ratio rises. What did not: the pick-regret trend. We no longer call it a property of the flow.
We then tried Yosys 0.47 and 0.51 to see whether regret moves steadily with version. Neither ran: both split abc -g AND,NAND,... at the commas and stop with No such command: NAND. So there is no data between the two ends, and the pipeline is not portable across those releases. Next is rewriting the gate list as an ABC script and checking it reproduces the 0.33 and 0.70 tables.
One design family, one clock, no intervals. Log: research/verified-dse/LOG.md (milestones 73 to 76).