[kkroo]

Retires the txfeed vote partition, offset 2 (232.0.3.2, and its IPv6 twin ff3e::232:302). This follows the CTO decision on BLO-22812 (2026-10-25) and is tracked on BLO-48422.

What changes

• txfeed.Partitions gives a simple vote no partition; before, it sent one to PartVote only. shred-txfeed is the only production caller, so the publisher stops sending to 232.0.3.2, and to ff3e::232:302 if -group-base6 is ever set. Simple votes are still counted in votes=.
• PartVote = 2 stays, as a retired and reserved constant. No other partition may take offset 2, so a stale subscriber gets silence, not another feed.
• txfeed-sub drops its -vote selector. txfeed-sub -vote now exits 2 with "flag provided but not defined". A raw -ssm 69.25.95.57@232.0.3.2:5003 still joins and gets silence.
• Comments: the shred-txfeed offset table, the package doc and the RefusedKeys rationale. The Vote program stays refused for W9 channels.

Why this is a behaviour change, not cleanup. The group is not empty today. Consensus votes left blocks at Alpenglow, but the simple-vote test ignores instruction data. That test is Agave's is_simple_vote_transaction shape: a legacy message, fewer than 3 signatures, one instruction, to the Vote program. So rare vote-account transactions such as Withdraw still match it, and today they go only to 232.0.3.2. On mainnet on 10-25 that was about 2 per hour, in bursts; slot 461300180 held 5 Withdraws. After this change they reach no group. Non-vote subscribers never received them, so nothing regresses for them. Folding them into non-vote instead would change what non-vote means and what the frame vote bit means. That is a separate decision.

Unchanged: the bytes on the non-vote, named-program and bucket groups; the frame format and its vote bit; W9 channel matching.

Deploy. A merge changes nothing that production runs. packaging.yml builds only ./cmd/blockcast-shreds, which does not import txfeed. shred-txfeed on CT 140 is a hand-installed binary, so the publisher change takes effect only when it is rebuilt from main and restarted. That is a separate production step.

Tests

• TestPartitions expects no partition for a vote. This fails on main.
• The txfeed-sub tests use -nonvote as the generic selector and expect 4 relay joins. A new case checks that -vote is rejected.
• TestGroup still pins PartVote to 232.0.3.2.
• Locally: CGO_ENABLED=0 go test and go test -race -tags purego on ./txfeed ./cmd/shred-txfeed ./cmd/txfeed-sub, plus gofmt and go vet.

[allyblockcast]

🔗 Paperclip issue: BLO-22812
🔗 Paperclip issue: BLO-48422

[allyblockcast]

Ally — Consolidated PR Review

Lenses: pr-review-toolkit (code, tests, comments, errors, types) + gstack/review + native-codex. All three were applied directly in this pod; no nested CLIs were run.
Reviewed head: de0600b

The change does what it says. Partitions now returns nil for a simple vote. Its only production caller is publish in cmd/shred-txfeed/main.go, and that ranges over the result, so a nil slice sends nothing. PartVote = 2 stays reserved, and TestGroup still pins it to 232.0.3.2, so no other partition can reuse offset 2. Removing -vote from txfeed-sub is complete: the flag, the -listen exclusivity check, the -group-base check and the "nothing to subscribe to" message are all updated. Nothing else in the repo references -vote or 232.0.3.2.

At this head I ran CGO_ENABLED=0 go test ./txfeed ./cmd/shred-txfeed ./cmd/txfeed-sub, gofmt and go vet. All passed.

Critical Issues (0)

Important Issues (0)

Suggestions (1)

• [native-codex] cmd/shred-txfeed/main.go:138: publish still builds dgs for a vote and returns len(dgs). So a large vote transaction still increments chunked (main.go:468) even though no partition group receives it. Vote-account transactions of that size are rare, so the stats skew is negligible. If you want chunked to count only datagrams actually sent on partition groups, have publish return 0 when Partitions is empty.

Strengths

• Keeping PartVote reserved means a stale subscriber gets silence instead of another feed. The RefusedKeys rationale was updated to match.
• The PR description is unusually precise about why this is a behaviour change: Withdraw-shaped transactions still match the simple-vote test and are now dropped. It also explains why folding them into non-vote was deferred.
• There is a test showing -vote is now rejected, and the relay and native join tests moved to -nonvote without losing coverage.

Recommended Action

1. No blocking changes requested.
2. Merge once the remaining required CI checks finish green.

[kkroo]

Landed by kkroo drive session a2176509 at de0600b5: Ally 0C/0I at that head, drive-merge-gate PROCEED, one squash merge with a trailer-free message -> 3fe642f2.