[trissim]

P440: delete the dead delivery CLI; keep the unit tool

TIME-1: CommsOrderDeliverTool owns holder delivery. DeliverVerb was a forwarding copy behind a commander-fork-only preflight, so ordinary holders never reached its gate. Delete the copy rather than widen --fork or add another caller family.

• Delete DeliverVerb, its help and now-unused tool import (22 production lines removed).
• Move both delivery call sites in test_order_claims.py to ThreadCaller("worker").invoke(CommsOrderDeliverTool(...), comms), reusing the first test's existing request.
• Import the existing proof_build_checkouts fixture into the test module; its held-commander fixture needs it when this file runs alone.
• Canonical skill names comms_order_deliver as the unit's own tool; no replacement verb or alias.
• No changes to CommanderVerb.apply, CommanderForkCaller.commanding, register_commander_child, CommsOrderDeliverTool, or P440's merged identity work.

Source evidence

Fresh main: d683dffd0a8aee75a40acb1a6da2136320fa6a7a.
Before/after existing refactor-audit Package inventory: 440 production, 471 test, 14 tools Python modules; zero parse omissions. Paired Toad source at de66b5120c2cc9ed29d86b84aa0e5a3bc22c32d9: 283 modules, zero omissions, no bindings to either class or literal order deliver consumers. After: no DeliverVerb declaration or dead CLI calls; three references to the unchanged tool in the test family. The tool is declared in tools.py; command-family membership derives from class declarations. Two dead CLI call sites, in the first test and its shared helper, are the complete tracked Python consumer set. Skill instruction is migrated with them. Dynamic external clients/MRO are not resolved by this AST inventory; the registration and tool boundary were read semantically. NRA was not run (last known shared environment lacks tree_sitter).

All 32 assertions are AST-identical to main. Protected guard methods are AST-identical. Ordered whole-src check git grep -n 'class DeliverVerb' -- src | wc -l: 0 (main: 1). git diff --check: clean.

Verification

Only tests/test_order_claims.py is run locally, under timeout 300, -n 0, and faulthandler timeout 60; normal conftest retained.

• Initial exact-main standalone invocation: five fixture setup errors, missing proof_build_checkouts; zero cases executed. Not a behavioral baseline.
• Main with that existing fixture registered via -p test_proof_build, PYTHONPATH=src:tests: 5 executed, 0 passed, 5 failed in 5.53s, each RelationViolationError: This thread is not a declared commander fork.
• Branch 7b68ae2312622ec40db9165b7fb60e3e004d4d95, normal standalone invocation with the imported shared fixture: 5 executed, 5 passed, 0 failed in 15.68s. Both runs emitted the existing goal_owner_fixture pytest assertion-rewrite warning.
• Hosted checks on 7b68ae2312622ec40db9165b7fb60e3e004d4d95: all 6 passed (smoke, native-scripts, install-rehearsal, review-verdict, debt-src, debt-tools), run 39696185747. No local ratchets or full suite.
• Core holder claim recorded successfully through native comms_order_deliver, on order f6f92e024e5c, revision 82. The current installed commander's real order board exited 0 and shows P440 held by unit-mi-1b, PR P440: delete dead delivery verb and use the unit tool #2584 at this head, Claimed by the unit, waiting on the commander's verdict. This demonstrates the real holder tool/board path on the current installation, not that the new build is installed.

Remaining application acceptance

Not installed or live-qualified by this PR report. After installation: original P440 fork order/check and superseded verdict work under their commander, sibling competition is refused, and an actual holder uses native comms_order_deliver and appears claimed by that unit on the board. No source-only native run or hand-built runtime is attempted.

Assumed

• Tool-catalog consumers continue using the unchanged declaration; dynamic external automation is not statically resolved.
• No assumption of installed success: the application run remains explicit unfinished acceptance.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

An organization owner can choose the repository organization as the billing source for reviews requested by members with a paid Copilot license. Learn about Copilot code review billing.

[trissim]

P440 installed command acceptance — 2026-10-21

Selected runtime: core-afb5909c8-lock-df5099ac8eb01d85; selected pointer unchanged at retirement. Git ancestry confirms it contains #2555 (4e6038dd7), #2576 (e0c4ff12c), #2580 (cd7233737) and #2584 (6b2b32ad1). Production imports came from that installed runtime, not a source checkout.

Results

p440-installed-afb-complete.log records 7 executed command checks, 7 expected outcomes:

1. order send ... --fork p440-installed-one: exit 0; registered child, commander remains creator, fork owns assignment and send attribution.
2. order check P2 ... --fork p440-installed-one: exit 0; check changed, order identity retained.
3. Competing order send ... --fork p440-installed-two: exit 1, expected refusal naming the original fork; assignment unchanged.
4. judge show --fork p440-installed-one: exit 0; reads the commander's external unit's edit.
5. judge superseded ... --by P2 --fork p440-installed-one: exit 0; verdict accepted, zero remaining unjudged edits.
6. Ordinary unit-mi-1b invokes declared comms_order_deliver through the installed CLI's production-owned InvokeCliCommand.argv: exit 0; claim sender is that unit and recipient is its commander.
7. Installed order board: exit 0; shows that unit's claim.

This is a continuous workflow on a fresh CheckRoot copy of the configured saved history, using a tiny generated Git repository and an external-only fixture registration for the review event. No agent processes were launched; all fixture process identities were checked. This does not claim a separate provider/Toad/UI qualification. No pytest tests ran this turn.

The installed command's live board also exited 0 and displays the original P440 claim (p440-installed-afb-board.log). Live P440 remained revision 88 before/after the copied workflow; its original claim and inputs were not replayed. Core still calls it claimed, awaiting the commander's delivery/closure record.

Failed verification setup, not product failures

• First copy: send, check correction and competition refusal passed. The synthetic external transcript was incorrectly written for a native unit; judge show returned zero edits and the driver failed before any verdict or delivery attempt. Log: p440-installed-afb-journey.log.
• A continuation tried opening that already-retired CheckRoot and failed before changing anything. Log: p440-installed-afb-continuation.log.
• Final copy used an explicitly external fixture participant and completed all seven checks. CheckRoot retired both disposable roots through its existing uncertainty-aware owner. No UNKNOWN attempt was replayed.

Retirement

p440-installed-afb-retire.log: no borrowers; generated fixture /home/ts/wt/unit-mi-1b-installed-afb-fixture removed. P440's development checkout remains absent; P428 remains preserved. Retained output consists of these small logs and this receipt, owned by unit-mi-1b for P440 closure.

Hosted-check wording correction

The earlier #2584 report's “all six passed” described PR job conclusions, not six executed substantive checks. review-verdict skipped its real review step on PR events; the commander later established that its merge-group refusal was not required by the ruleset. That was corrected by the commander at 20:18 (messages 17797/17805). The five substantive hosted checks and fork-review-t's separate accepted edit review are distinct evidence. The installed workflow above supplies the application evidence; the green skipped job does not.