Follow-ups to the one-job gate: the manual heal route is frozen and declared; two latent witness reds repaired

[gunbai-bot]

Follow-ups to the one-job gate (#13658): the manual heal route is frozen and declared, and two latent witness reds it left behind are repaired.

Base: main, after the v1 closeout (#13641) lands. The two commits were written on the approved head of #13658 (75bd66f) and carry only what the operator's review of that PR deferred; they are cherry-picked onto the closeout's final composed head 2bf8735 (one generated-file conflict, docs/design-rung-drops.md, regenerated) so the PR diff against main after landing is exactly these changes.

What changes

1. The manual heal route is frozen, not left as a green-looking seal path (review 78398 on Required CI: withdraw the four v1-judged lanes; the gate is emit-build alone (operator ruling 2026-10-09) #13658). heal.yml (gunbc.heal_workflow) still sealed and uploaded the repair candidate that the deleted heal publisher consumed, so a dispatch of it produced an artifact nobody publishes. The authority now carries a FROZEN header annotation stating that the route has no consumer on the required path; gunbc.rung_drop v1_required_lanes_withdrawn names the frozen route as a member of the drop with the parenthetical that nothing on the required path seals a candidate any more; gunbc.heal_publication names its producer path as HealWorkflowYamlArtifact instead of the deleted publisher; and the CI row of gunbc.design_document says the same in one sentence. The operator's standing ruling on this (2026-10-10, chat): keep the heal publication concept, build a proper v2 regen later; this change freezes the v1 route without deleting the concept.
2. Two latent reds repaired. test.claim.fabric.fci1_bounded_execution_context_witness_test expected the envelope basis in gunbc.runner_slot_allocation while gunbc.fabric.fci1_bounded_execution_context declares it in gunbc.runner_slot_desired (2/2 after). test.claim.heal_publication_boundary_witness_test fixtured the candidate producer path as the deleted publisher's workflow; it is now .github/workflows/heal.yml, in production and in the fixture (34/34 after).

Evidence

• Both witness files were executed on srv1 with the seed built from 75bd66f under systemd-run --user --scope -p MemoryMax=28G -p MemorySwapMax=0; counts above are the runs' own partition summaries.
• The required gate (witnesses, one job) does not evaluate .dag witnesses, so these runs are the only execution the two files get; that gap is the declared drop v1_required_lanes_withdrawn, whose trigger is the native compiler judging the population.
• Generated projections: main_wet over the composed head plus these commits (srv1, the closeout's fixed-point seed) exit 0 and changed only docs/design-rung-drops.md; claim_executor --required-regen on that tree reports first_generation_equal=true (no stage0 mirror moves), so the follow-ups are projection-only on the generated side.

Not in this PR

• The gunbc.witness_floor_lanes required_lanes_roster fork (12 consumers still read the deleted lane roster) is a runtime task, filed separately.
• No change to the gate's authority (gunbc.compiler_gate_workflow, gunbc.emitted_subject_build_gate).

🤖 Generated with Claude Code

[gunbai-bot]

Closing as superseded, at the operator's direction (2026-10-11). This PR froze and declared the manual heal route under gunbc.rung_drop v1_required_lanes_withdrawn. The operator has since directed a non-required, scoped regeneration and push-back job instead: Tier S stage0 mirrors plus seed-build artifacts, and a scheduled read-only drift check for operational configs. That lane re-derives the heal and publication route and amends the same rung-drop row. Its two latent witness-red repairs (fci1_bounded_execution_context_witness_test, heal_publication_boundary_witness_test) are handed to that lane to carry if they still apply after its rework. The branch is kept for history.
— sent from smart-gull-336