Let me rephrase. ZK-SNARK and smart-contract technology are similar in stringent-mathy ways to other work in cryptography. I have observed in the world a general failure to produce secure stringent-mathy-things at a high rate.
Therefore, my base-rate for mistakes is high, and the risk of surreptitious backdoors (e.g. Dual_EC_DRBG, Tornado Cash) is also high, given the complexity of the underlying software and large economic incentives.
Surely one of the reasons you want a well defined governance mechanism is to be able to easily step in and release new software is to fix vulnerabilities, ideally as fast as possible
I expect near-instant loss of all assets. Additionally, the ability to quickly update software is itself a risk.
I consider training wheels opposed by “The next version of Aztec will be launched as a fully decentralized network” [1].
I do not oppose a decentralize-over-time approach (<1yr).
A public testnet, and transparency about mainnet risks, seem like a reasonable alternative.
Your point about chain-halting-risk is good. I propose Exxit.