In the discussion around James' OP_VAULT proposal, it was implied that
precomputed vaults must use ephemeral keys that must be deleted as part of
the vaulting protocol, like Bryan Bishop's proposal
<https://lists.linuxfoundation.org/pipermail/bitcoin-dev/2019-August/017229.html>.
Looking around, I haven't been able to find any wallet vault proposal that
doesn't require ephemeral keys, so at the risk of posting something that's
obvious to everyone, I wanted to share a simple way to do a wallet vault
without requiring any key deletion.

The basic idea is to create an N-of-N multisig address, and pre-sign some
transactions from it with N-1 keys to an address with several timelocked
spend paths. This has the fallback that funds can always be spent
immediately if you use all the keys, just like a normal N-of-N multisig
address (since that's what it is at its core), but the usage of any of the
pre-signed transactions leads to an address that guarantees a clawback
within a time window. Here's a 3-of-3 example:

*Vault Initialization*:
1. Create 3 of 3 Vault Address
2. Create an Interim Address that can send with:
 * 1 of 3 keys after a timelock of 1 month
 * 2 of 3 keys after a timelock of 1 week
 * 3 of 3 keys with no timelock

*Vaulting*:
1. Create a transaction sending an output to the Vault Address
2. Create a transaction spending that Vault Address output to the Interim
Address
3. Presign one copy of the step-2 transaction for each of the three
combinations of two keys.
4. Store seeds separately, store the wallet config as well as the three
presigned transactions with each seed.

*Unvaulting*:
1. Sign one of the pre-signed transactions with the missing signature.
2. Broadcast
3. Wait the appropriate timelock for the number of keys you want to sign
with.
4. Create a transaction sending from the Interim Address.
5. Broadcast

*Recovering *(after unvaulting step 2 after the broadcasted transaction to
the Interim Address has been mined):
1. Sign the utxo with all three keys to any destination. Alternatively sign
with two keys, wait 1 week.
2. Broadcast it

This has the usual downsides of pre-signed vaults that you need to backup
these transactions for each vaulting, complications involving the
flexibility (or lack thereof) of fees, and inflexibility in how much to
unvault (must be the whole utxo, no change). This could of course be
augmented with various techniques to make fee handling more flexible
(anchor outputs, multiple versions of the presigned transactions with
different fees, etc). More complicated presigning schemes could allow for
some flexibility in unvaulting amount (eg by sending change back into the
vault, and creating additional pre-signed transactions for that new output).

It also has the same downside that OP_CTV vaults have, where a stolen key
can steal funds from the interim address by racing the owner with their own
transaction once the necessary delay has passed. Note that James' OP_VAULT
opcode wouldn't have this problem.

But not requiring any toxic waste keys seems like a pretty good benefit
over Bryan Bishop's original proposal.

Anyways sorry if this was already on people's radar and just too obvious to
post about.