[mkreyman]

US-47.11, part of #986 (items 2 and 4).

Gate A: chg-loop tickets measured apart from human intake

• Measurement.Ticket gains chg_loop?, true exactly when the labels include chg-loop; the title plays no part.
• Report.gate_a_strata/1 builds every stratum once, and both the main report and the sensitivity run use it. It now returns:
  • intake and all, unchanged;
  • chg_loop, the chg-loop tickets;
  • human, every other ticket.
    chg_loop and human partition all.
• The summary prints all four strata.
• Escalation rows carry a chg_loop flag.
• Gate A itself is not touched.
• New artifact pair docs/measurements/gate_a_2026-10-19.{json,md}, made by mix loopctl.gates.measure_a from home_care_billing's 536 issues. It is redacted the same way as the 2026-09-14 pair, which stays as it was.
  • chg_loop is empty: chg has filed no chg-loop issue yet on home_care_billing, claude-config or chg.
  • human 20.7% escalated, intake 61.0%.

Gate B: self-modification

• A repository absent from DELIVERY_GATES_CONFIG escalates every change with {:unknown_repo, repo}. A new test pins this for both repos, at the merge gate and at the triage screen.
• claude-config and chg are absent today, so Take chg-loop issues as delivery-loop intake: enroll claude-config and chg, measure Gate A on machine intake #986 item 4 already holds.
• docs/agent-delivery-loop.md states the rule: when either repo gets an entry, its human_paths must carry its self-modification code.
• The pasteable glob lists are in KB article 4a86157b, not in this public doc, because both repos are private. Every glob was checked against git ls-files.

Item 1 (operational, done outside this PR)

claude-config and chg are enrolled as intake sources, each with its own work project and target epic, and an Issues-only webhook. Both pings returned 200.

• The work projects have no repo_url, deliberately: the existing KB scopes already carry it, and a second project on the same URL makes resolve_project ambiguous.
• required_checks is empty on both. That matches the loopctl source.

Tests

• Ticket flag: chg-loop label, no label or a lookalike, title-only.
• Strata counts on a mixed set.
• The partition.
• intake and all unchanged.
• An empty chg_loop has a nil rate.
• Sensitivity run and summary.
• Escalation-row flag.
• Gate B unknown-repo escalation.
• Mutation-checked: splitting on intake? fails 4 tests, hard-coding the flag fails 5, and letting chg through Gate B fails the pin.

Defaults Mark may want to overrule

1. Gate B globs for the private repos are in the KB, not in the public doc. To publish them instead, copy article 4a86157b into the doc.
2. required_checks is empty on both new intake sources, so story verification records no_required_checks there. To opt in, set it with intake_source_update, using job names from a PUSH run.

[mkreyman]

Review round 1 (code) counted at 262a261e2c1f via Skill lean-review. The review-round counter's merge gate reads this marker on any machine.