EPIC: PR merge sweep 2026-10-19/20 -- handoff to the Queen (state, requests out, how to resume TRI CLAW) #11259

Report channel for the pull-request merge sweep of 2026-10-19/20, handed to the Queen at 07:20Z because the working session ends. This epic deliberately carries none of the four dispatch headings, so it is never dispatched as bee work. Sub-issues are filed under it only if bee-shaped work appears.

Goal

Owner order (2026-10-18/19): merge every open pull request in gHashTag/t27 that passes the repository's rules, and let TRI CLAW do it around the clock.

The rules the sweep applied

• Required checks green: validate, check-linked-issue, parse-ratchet, own-language.
• Every other red must be inherited: the PR's failure set is a subset of master's newest finished run of the same check. A red master has since fixed is not inherited; update the branch and rerun.
• Closes #N present (L1), not a draft, no conflict, no no-auto-merge label. Bee PRs (queen-*) are merged only by auto-merge-ready-prs.yml.
• Foreign-line budget of the own-language gate respected.

State at 07:20Z on 2026-10-20

• Merged by the sweep in its last stretch: gen-verilog: do not wrap an assignment target in $signed -- fixes the systolic fpga-lint / fpga-conformance reds (Closes #10780) #11094 (gen-verilog emitter fix), chore(deps)(deps): Bump ed25519-dalek from 2.2.0 to 3.0.0 #10961, chore(deps)(deps): Bump dirs from 5.0.1 to 7.0.0 #10962 (dependency bumps), elab baseline: add systolic 0 so master fpga-conformance is green again (Closes #11099) #11100 (elab baseline systolic 0), spec(gen_w868): port scripts/gen_w868.py to specs/port/scripts/gen_w868.t27 #11085 (gen_w868 port), claw README: the */10 cron is best effort, measured gaps are hours (Closes #11132) #11134 (README cadence correction). Earlier in the sweep about 40 PRs were merged or prepared and 25 closed on the owner's word; the TRI CLAW merge lane itself is claw: merge_judge.t27 -- the 24/7 merger decided in a spec; claw-merge replaces claw-refresh (Closes #10973) #10978 (spec specs/claw/merge_judge.t27).
• Master, measured at 07:14Z over its last 8 commits: scan green; duplicate-bodies RED on every one of them (red since at least 04:46Z; the cause was not investigated, it was green on 2026-10-19 at 2e2ca5f55); spec-guards and coverage green at e10a1bb7c, red at ef2e7ee45 (07:04Z), still running on ae97ff06d. The two seal/baseline causes the sweep fixed (seals: ml_host_io_model gen_hash_verilog follows the gen-verilog assignment-target fix (Closes #11102) #11105, elab baseline: add systolic 0 so master fpga-conformance is green again (Closes #11099) #11100) are not what is red now.
• Open: 168 PRs at 07:2xZ. About 131 are bee PRs (left to the bee merger); the rest are drafts or held. All 20 non-bee, non-draft, unheld PRs were judged BLOCKED at 07:2xZ, each by its own failure set (not a subset of master's red set):
  • spec(gen_w867): port scripts/gen_w867.py to specs/port/scripts/gen_w867.t27 #11086 to spec(gen_w861): port scripts/gen_w861.py to specs/port/scripts/gen_w861.t27 #11093 and spec(gen_w869): port scripts/gen_w869.py to specs/port/scripts/gen_w869.t27 #11184 to spec(gen_w876): port scripts/gen_w876.py to specs/port/scripts/gen_w876.t27 #11191 (dmitrii-f-t27, gen_w861 to gen_w876 ports): duplicate-bodies; the new copies include build_tree_levels, total, wrap_matches_normal. Correction and options posted on spec(gen_w861): port scripts/gen_w861.py to specs/port/scripts/gen_w861.t27 #11093 (comment 6281839194).
  • spec(ternary): pipelined dot27 -- 5-stage on_clock adder tree (Closes #11237) #11244 and spec(ternary): pipelined dot27 -- five-stage on_clock pipeline for the 27-trit dot product (Closes #11242) #11245 (ternary pipelined dot27): the absurd-width ledger for specs/t27os/glass_window.t27 and specs/t27os/pty_server.t27; spec(ternary): pipelined dot27 -- five-stage on_clock pipeline for the 27-trit dot product (Closes #11242) #11245 also emit-bitexact.
  • spec(ternary): trit codon -- perfect [4,2,3] Hamming code over GF(3), base-3 and wire forms (Closes #10969) #10970 (a spec that no longer compiles), igla cell 25 published: the whole-model HSLM trained by t27c --backend ml is on Hugging Face (Closes #10833) #10836 and spec(port): land the CR-CHAT-03 wave 3 batch -- 23 targets (Refs #8883) #10802 (duplicate-bodies, unreachable source files).
  • feat(t27f): T27 OS home -- the 27-petal triangle menu as one spec #10783 (conflict, width ledger) was in the 04:12Z list and is not in the 07:2xZ one.

TRI CLAW lane: why it is not merging yet

The claw-merge job of gate-topology.yml is plan-only. Its App token cannot be minted: every run since 2026-10-19 shows "The level of access for permissions requested are not granted to this installation". The repository variable CLAW_MERGE_LIVE is unset (HTTP 404).

Requests out

How to check the lane once it is live

• The first claw-merge run after the grant has no "level of access" annotation; its step summary lists G lines (merged) and a reason for every other open PR.
• It merges squash, pinned to the judged head sha, at most 5 per run, and leaves PRs younger than 10 minutes alone.

What bees can and cannot do here

Bees run in containers without a GitHub credential by design, so they cannot merge a PR, update a branch or set a repository variable. The continuing worker for this epic is the claw-merge lane in GitHub Actions, once it is live. No bee-shaped work is open right now.

Traps learned (short)

• A watcher can judge a PR in the seconds before GitHub creates the update-branch commit; re-judge after the new head appears.
• A verdict "REVIEW" means a human check is pending: compare the failing seal or row sets against master's run by hand before merging.
• A PR that moves a Linux-only seal must not be merged before the Linux re-save; an owner-token merge bypasses that.
• tools/check_elab_ratchet.py --update-baseline run on macOS writes iverilog-version 13.0; CI measures 12.0, so keep the 12.0 line.
• The cron cadence claim in the docs was never measured; it is now corrected (claw README: the */10 cron is best effort, measured gaps are hours (Closes #11132) #11134).

Refs: #10978, #10973, #11094, #11100, #11105, #11132, #11134.