To be clear I’m not necessarily wedded to a finite supply cap. That said, I do think that you are missing one of the big arguments in favor of finite supply: mistrust of future governance.
The fact is that we have not solved blockchain governance, and I worry that there is a large chance that we never will come up with a blockchain governance process that is sufficiently robust that it will be capable of regularly doing things like adjusting fundamental economic parameters. Right now, we are living in a unique time where blockchains have become a multibillion-dollar force, but the kinds of usual powers-that-be that typically try to manipulate multibillion-dollar things for their own interests haven’ t fully figured this out, but we can’t rely on this to continue to be the case in the long term.
If Ethereum becomes part of mainstream global internet infrastructure, then hodlers, staking pools, large corporate users, etc will all be demanding their seats at the table, and we should absolutely assume that they’ll use all sorts of techniques, including social media manipulation, trolling, influencing client development companies, backroom deals, etc to try to get their way. If Encrypted Media Extensions can get into W3C standards, then I absolutely expect that some issuance mechanism that seems on the surface reasonable but is actually totally capturable by oligarchic interests (cough cough DPOS) that push for it could get into the ethereum protocol, unless we armor ourselves to preclude such possibilities. Subjective social protocol governance is NOT incentive compatible and is NOT a solved problem; if the US government can’t figure it out, I see no reason why we should assume as a certainty that we will.
This is why Schelling fences are important. Cargo culting around numbers is just the way that communities make Schelling fences work in a way that allows people to contribute to the Schelling fence’s strength even if they themselves do not deeply understand the technical arguments; the point is that if you try to transgress some boundary then hordes of noobs will start yelling “hey, you’re overstepping our magic number we don’t really understand but definitely agreed upon!!!1!” and this will make your life harder (BTW this also applies to political norms like “free speech”).
That said, simply stating “Schelling fences can be useful” is not an argument that any specific Schelling fence being adopted today would be good; all Schelling fences have costs in that there exist possible futures where overstepping a fence becomes optimal but is prevented because the fence exists. There are options outside of finite supply, like “maximum 2 million new coins per year” that could work too (note: the original Ethereum whitepaper specified a limit of ~15.6 million new coins per year; I think this is far too high a limit but it’s certainly worth respecting).
For consensus, I do at present believe that long-run consensus security expenditure should be equal to or less than zero (less than if transaction fees are burned); the reason is that the Ethereum blockchain contributes to the security of both ETH and every ERC20, and so I don’t think it’s reasonable to uniquely burden ETH holders specifically with the task of paying for the security, and I do now believe that it will be possible in the long run to pay for sufficient security with txfees alone. That said, it may be wise to delay setting a limit until a proof of stake algorithm has been launched and has worked for some time.
As far as issuance for public goods goes, I personally am open to it and we should perhaps discuss that more explicitly (I wonder what the results would be of a community poll like “if there was a reasonably exploitation-resistance algorithm for issuing 1m ETH per year and allocating it to protocol-relevant public goods like research and development, would you be okay with adding it into the protocol?”). That said, at least I personally would be in favor of only allowing such a thing if we’re reasonably certain a good algorithm exists, and would also be in favor of Schelling fences that constrain it to at most something like 1–4 million coins per year.
But it may well be the case that we’ll become more and more convinced over time that such algorithms don’t exist; voting is exploitable, it’s hard to make trustless prediction markets for anything other than proof of work difficulty, so there may be no solutions other than those that basically amount to call-out assurance contracts.