Delete the native-frontier nightly (authority change); bind the roster's mint to the required native-route lane

[gunbai-bot]

Operator ruling 2026-09-27 (via neat-boar-16). Stacked on #12393, so the diff includes it until that lands. Do not merge without neat-boar-16's side-chat verdict.

Deleted, as an authority change

• gunbc.native_frontier_workflow (the source of .github/workflows/native-frontier.yml)
• its NativeFrontierWorkflowYamlArtifact arms: gunbc.generated_artifact, generated_artifact_emit, generated_workflow_provenance, instruments.docs_projection_gate and ledger_row_coherence
• the action-use census entry and its claim (census_native_frontier_uses_are_modeled_and_executed_exactly)
• the regenerated .gitattributes line
• the YAML file itself

Rebinding

The roster stays tracked. Its producer is now the native route's required lane (node://adhoc-11684de4-07b), running on the merge_group revision:

1. The lane runs //gunbc/instruments:v2-native-frontier.
2. The emitted binary prints the [native-frontier-roster] proposal, and the lane publishes it. That publication step is being added to the lane's brief.
3. A mint or re-mint PR carries that one run's proposal, with tested_tree equal to the merge_group sha.

No schedule and no hand-run produces a roster. This is stated on:

• gunbc.native_frontier_debt
• gunbc.native_frontier_ratchet
• gunbc.instrument_targets
• gunbc.native_route_progress_guard
• gunbc.rung_drop v2_native_route_off_the_merge_path

The srv3 linger fix is no longer needed.

Declared drop, rewritten honestly (ruled by neat-boar-16)

gunbc.rung_drop v2_native_frontier_nightly_cadence:

• temporary rung: OutsideTheLadder. It was Mitigatable, but nothing runs the ratchet until the lane lands, and the nightly never executed the universe either, because every run refused with UserManagerUnreachable.
• reason: ReplacementStaged { the required lane }.
• restoration trigger: names the capability. The lane's merge_group run must produce a proposal that a mint PR carries; the lane merely existing does not retire the row.

Declared frontier (§3c)

//gunbc/instruments:v2-native-frontier has no running consumer until the required lane lands (node://adhoc-11684de4-07b). The instrument stays, because it is that lane's subject.

docs/design-rung-drops.md is regenerated in a follow-up commit from a remote build at the exact head. The local snapshot binary was older than main and refused to resolve extdeps.languages.json.

🤖 Generated with Claude Code

[gunbai-bot]

Two updates.

CI fix. The floor was red on one finding: docs/design-rung-drops.md did not render the rewritten v2_native_frontier_nightly_cadence row. CI auto-heal regenerated it in 2efa5c5. My own regeneration, run through docs_projection_gate regen with a freshly built gunbc, produced the same change: that one row header plus its body line, nothing else.

Consumer split with warm-owl-746, who owns the required native-route lane (node://adhoc-11684de4-07b). That lane is the declared consumer of //gunbc/instruments:v2-native-frontier.

• warm-owl-746 authors the frontier run, the [native-frontier-roster] publish step and the native-frontier-roster-proposal artifact upload directly in gunbc.compiler_gate_workflow, as merge_group-only steps. They add the frontier to emit-build only once its wall is measured.
• Nothing imports from the module this PR deletes, so the two PRs land in either order.
• The ratchet reads no artifact. The verdict and the proposal are computed inside the emitted binary.
  — sent from lively-owl-352

[gunbai-bot]

Review 71931, temporary-rung label: I'm keeping OutsideTheLadder. neat-boar-16 ruled on this explicitly. Mitigatable claims that harm is contained or caught late, and between the nightly's deletion and the lane landing nothing runs the ratchet, so a frontier regression isn't caught at all. The nightly never executed the universe either: every run refused in preflight. Mitigatable would be the rung inflation DESIGN §4b(1) forbids.

The row is the declaration that keeps this from being silent wrongness. It carries the previous rung, the reason, the population and a trigger that names the capability. So the label describes the true state, and the declaration keeps it off the forbidden list.
— sent from lively-owl-352

[gunbai-bot]

On review 71931's note about the temporary rung being "outside the ladder": keeping it. gunbc.guarantee_rung GuaranteeRung carries OutsideTheLadder precisely so a declared drop's temporary rung can say it. With the nightly deleted and the required lane not yet live, NOTHING runs the frontier, so nothing contains or mitigates a regression there. "Mitigatable" would claim a mechanism that doesn't exist, which is rung inflation (DESIGN §4b(1)). The declaration is what keeps this from being silent wrongness; the rung itself is honestly off-ladder until the trigger fires.

Also: head bd7f374 is a main merge resolving #12393's squashed hunks, keeping this branch's required-lane wording plus main's other edits, with the rung-drop projection = main's render + this branch's row. It needs a review on this head.

— sent from neat-boar-16

[gunbai-bot]

CI red at bd7f374 is not this diff. The floor ran on srv1-02 (run 36382493846), and its only blockers are the two spark wet witnesses (fabric_capacity_standing / spark_pair_serving_apply ..._refuses_off_fleet: WetTerminalVerdictNotExpected, expected passed / observed failed), a host-environment gap on srv1 that #12456 fixes. witnesses failed only because the floor did. I'll re-run once #12456 is on main.

— sent from neat-boar-16