Thanks for the careful review. This is the kind of discussion necessary for making progress on post-quantum Bitcoin.
There's a lot here I agree with, though I come to a different conclusion. I agree that hash-based signatures are a good fit for Bitcoin, for the reasons the review gives. Reducing block space usage is indeed a primary design goal of SHRINCS, and it comes at a cost. The stateful path's safety depends on wallet infrastructure that the protocol does not prescribe and that still needs standards.
I also agree that if block space stays cheap, the stateful path adds complexity for little benefit. But that seems unlikely, since today's fees reflect 64-byte signatures, and stateless SHRINCS signatures are about 90 times larger. High fees matter for a store of value too, because holders eventually need to spend, and we should aim for Bitcoin to remain usable by people who don't hold large amounts.
Congestion and high fees are a strong argument for minimizing signature size, and stateful signatures are up to about 10 times smaller than stateless ones. Wallets can start with stateless signing and enable stateful signing for their existing keys later, without another soft fork. A block-size increase can't fully solve this. Every node has to download and validate every block, so blocks can't grow arbitrarily large.
The review suggests that the stateful path lets the SHRINCS authors say the block-size concern has been addressed. The intent is rather that SHRINCS doesn't depend on a block-size increase, so that question can be discussed separately without holding up post-quantum signatures.
On custody, I agree that it's unreasonable to expect replicas to share stateful signing slots safely. Some high-availability setups can still use the stateful path, for example one where a designated signer signs statefully while the replicas sign only statelessly, so failover costs larger signatures but requires no coordination of state. In practice, custodians may simply sign statelessly.
The review describes stateful signing as an optimization, which fits better than the BIP's "primary path" wording. It also made me realize that calling the stateless path a "fallback" is misleading, and we're considering changing the terminology in the BIP.
The review proposes considering a stateless-only scheme with a larger block size. Modified algorithms could reduce stateless signatures to roughly 4,000 bytes, about 60 times the size of a Schnorr signature instead of 90. Would that be worth departing from the SLH-DSA algorithms?