I think this aphorism has done much more harm than good in the last 5 years.
Imagine if this same bar had been applied to the deployment of CLTV/CSV in the context of Lightning. We may still be waiting on a “proof” version of Lightning that delegates locktimes to a multisig oracle. Who would invest the time to build such a thing?
Could e.g. Lightning Labs have raised capital absent those consensus changes?
Some have used this kind of rationale as a cudgel to claim that if there were “real demand” for, say, vaults that we’d see some similar multisig-crutched prototype. While Revault was essentially this – but likely failed because vaults without consensus support are too complex. Meanwhile many companies are asking for consensus-driven vaults (AnchorWatch, Anchorage, NYDIG, River).
If we have soft fork proposals that are simple to verify for safety, have mature implementations and demand, and potentially wide use – as in the case of CLTV – we should deploy them. I don’t think we should handwring about a lack of multisig-enabled Rube Goldberg machines.