Public ZSA testnet still verifies with the pre-fix Orchard circuit (halo2_gadgets < 0.5 soundness advisory)

[0xpierre-dev]

Hi! An observation from building against the public testnet, and a question about redeploy plans.

What we observed

The halo2_gadgets soundness advisory (missing copy constraint in variable-base scalar multiplication) is fixed in 0.5, and your current zsa1 lines already pin it:

Repo	Rev	Requires
QED-it/librustzcash @ zsa1	c5c232db	patches orchard to cf801a5d
QED-it/orchard	cf801a5d (0.14.0)	halo2_gadgets = "0.5"
QED-it/halo2	ef3d0ba2	gadgets 0.5.0 (fixed circuit)

We migrated our client stack onto exactly that set: Cachet, a browser prover and verifiable registry for ZSA, live at cachetzec.com. It compiles cleanly and all tests pass, but proofs from the fixed circuit are rejected by the deployed consensus:

• our regtest node (zebra-regtest-txv6, built from 2b036fd6) rejects them with could not validate orchard proof
• dev.zebra.zsa-test.net (Zebra 4.2.0) accepted our old-circuit proofs as recently as yesterday, so it verifies with the pre-fix key too

Which makes sense: the soundness fix changes the verifying key, so it is a consensus change, and no client can ship it unilaterally.

Questions

1. Is a testnet redeploy on the fixed circuit planned, and roughly when?
2. Would it be a chain reset (rules enforced from genesis, as at launch)?

Our migration branch is ready and tracked in cachet-zec/cachet#3. Happy to test against a new network the day it exists.

Note

Adjacent question: is a rebase of the ZSA branches onto the current trait generation (ff 0.14) on the roadmap? That is what would let browser provers pick up Zakura Common's speedups by changing version pins.