docs(adr): record the release gates that block a fleet-ready verdict

Merging puts this in the next release, so the bar is fleet-ready rather
than code-complete, and green CI is evidence for the code gate alone.

Enumerates what is still open: the mutable protocol branch pin, mixed-version
proof over a real connection, deployment ordering, observability of the
adoption counter, NAT/canary, rollback rehearsal and fleet safety.

Calls out that deployment ordering is the one gate that is not inert. The
refusal machinery cannot fire while MIN and CURRENT are both the first
declarable version, but against a fleet that cannot answer a versioned
request every first contact still costs a probe wait, so releasing the
client ahead of the nodes adds real latency to cold uploads.

Author: grumbach

docs/adr/ADR-0010-settlement-version-and-pre-payment-compatibility.md

@@ -159,6 +159,23 @@ The cost of that is roughly two minutes per merkle E2E test for as long as the s
 
 **Still outstanding:** the reverse direction, old client against an upgraded node, and a fleet with both. Those need a testnet built from this branch's `ant-node` rather than the published one, which is not available until the coordinated set lands. Also unproven end to end: structured refusal, lost refusal, and send failure against real peers, which are pinned at unit level only.
 
+### Release gates
+
+Merging puts this in the next release, so the bar is fleet-ready, not code-complete. Green CI covers the code gate and nothing else. **The set is not production ready while any of these is open:**
+
+| Gate | Status |
+|---|---|
+| Code / CI | Proven across all three repos |
+| Dependency | Open: both downstream crates pin a mutable protocol branch |
+| Mixed-version | Partial: new-client against an old fleet is proven by the client suite; the gate itself has never run over a real connection, because the devnet speaks the pre-versioned dialect |
+| Deployment ordering | Open, and **not inert**: against a fleet that cannot answer a versioned request every first contact costs a probe wait, so releasing the client ahead of the nodes adds real latency to cold uploads |
+| Observability | Open: the unversioned-quote counter is the signal that retires the legacy path and has never been read in production |
+| NAT / canary | Open: no canary; relayed and NAT'd peers are the paths this adds work to |
+| Rollback | Argued, not rehearsed |
+| Fleet safety | Open: the corroboration quorum and the client-wide latch have never met a real fleet |
+
+The refusal machinery is inert on arrival, since `MIN` and `CURRENT` are both the first declarable version, so nothing can be refused yet. That lowers the risk. It does not close a gate, and the client-side cost above is live regardless.
+
 ### Re-open triggers
 
 - The unversioned-quote count failing to decay, which would mean a long tail of clients the gate can never protect and would raise the priority of option 3.