docs(backlog): re-verify the #1085 gate fix at HEAD and propose closure (BACKLOG #1085)

[wshallwshall]

The fix is on main, it is complete, and the archive pass is the only thing outstanding

This is a verification pass, not a build. #1085 has read FIXED 2026-08-26 with its banner held
open for the archive pass. I checked every claim it makes against main at fd44b0f17 by reading
the code and by exercising the gate, and all of them hold. The one change in this PR is a
verification record appended to #1085's banner block, proposing closure. I did not take the
closure, per the brief.

No code changed. scripts/hooks/worktree_gate.ps1 is byte-identical to main.

What I verified, and how

Brief item	Result
1. Composition present, relative vs absolute distinguished	Yes. Get-GitTargetCandidatesRaw joins a relative -C onto $cd, leaves an absolute one alone.
2. Not unconditional (the fail-open the item warns about)	Confirmed. The join is guarded on [System.IO.Path]::IsPathRooted.
3. Three bail-outs guard both branches	Yes. $cd is computed above the -C branch; popd, cd - and a (/{ subshell leave it null, and so does more than one chdir ($cds.Count -eq 1).
4. The negative control the item describes	It exists. No remaining work here.
5. Gate suites and the mutation	201 pass; the mutation reds only the composition test.

The guarded join, at scripts/hooks/worktree_gate.ps1:

if ($cd -and -not [System.IO.Path]::IsPathRooted($dashC)) {
    $dashCOut += (Join-Path $cd $dashC)
} else {
    $dashCOut += $dashC
}

On point 4 specifically, since it was the one flagged as possibly real remaining work:
test_an_absolute_dash_C_still_ignores_a_preceding_cd is shipped in
tests/test_worktree_gate_shell_semantics.py, and its docstring names the
broader-than-its-evidence problem in test_a_dash_C_beats_a_preceding_cd in as many words. Two
further controls sit beside it: test_a_relative_dash_C_with_no_cd_still_resolves_against_the_session_cwd
and test_a_cd_inside_a_subshell_does_not_compose. Nothing was missing, so I added nothing.

I also confirmed the gate's own defect inventory was updated rather than deleted: the
COMPOSE vs PREFER entry now reads FIXED 2026-08-26 (BACKLOG #1085), and listed here rather than deleted so this inventory is not read as still complete.

The mutation reproduces

I reverted the join alone to $dashCOut += $dashC, which reverts #1085's composition and leaves
$postC, the bail-outs and everything else untouched.

1 failed, 34 passed in 83.74s
FAILED tests/test_worktree_gate_shell_semantics.py::test_a_relative_dash_C_composes_with_a_preceding_cd

Exactly one red, the composition test, with all four controls green. Its failure prints the
original defect verbatim: a deny naming the primary for a write that lands in the ungoverned
../Unrelated. The file was restored and verified byte-identical with diff -q before I staged
anything.

The tests exercise this worktree's copy of the gate (GATE = parents[1] / "scripts" / "hooks" / "worktree_gate.ps1"), not the installed hook, so nothing about the live gate was touched or
worked around at any point.

Checks run

Interpreter check, as required, printing a path inside this worktree:

$ /c/Users/Scott/Code/MessageFoundry/.venv/Scripts/python.exe -c "import messagefoundry; print(messagefoundry.__file__)"
C:\Users\Scott\Code\MessageFoundry\.claude\worktrees\agent-a4cf8ff83fe9871cd\messagefoundry\__init__.py

Check	Result
test_worktree_gate_shell_semantics.py, test_worktree_gate.py, test_worktree_gate_git.py, test_worktree_gate_hijack.py	201 passed in 609.78s
Mutated-resolver re-run of the shell-semantics suite	1 failed, 34 passed, as intended
test_backlog_status_check.py, test_backlog_citation_check.py, test_dangling_citation_check.py, test_citation_line_check.py, test_ledger_check.py, test_backlog_hygiene_claim_extraction.py	155 passed
scripts/docs/backlog_status_check.py	OK, 657 items, each declaring exactly one status
scripts/docs/backlog_citation_check.py	OK, 231 citations in scope, 2 pre-existing advisory warnings
pre-commit (ledger gate, leak guard, conflict-marker parse, secrets, control chars)	Passed

Skipped, with CI as the authority: ruff, mypy and the rest of pytest tests/. This PR
changes one Markdown file and no Python, so ruff and mypy have nothing to check here; the
pre-commit run reports both as no files to check. The full suite was skipped deliberately per the
brief, which names tests/test_connscale_smoke.py as failing under worktree contention from the
fixed-port collision in #1014. Hosted-runner-only legs were never visible to this process.

Two judgment calls, stated rather than asked

The 222 does not reproduce, and I recorded that rather than silently matching it. The item says
"222 tests pass across the four closest gate suites" but never names which four. The four closest
by subject collect 201, all passing. Two counts over unnamed sets are not comparable, so I read
the recorded number as dated rather than wrong and named the suites in the new record so the next
reader can re-run the same measurement. The resolver has been changed three times since the fix
landed in a490993b4 (PR #628) -- 7599ad190 (#1065, #1072), 2b9f5b3c4 (#1304, #1071, #1039)
and 85516d100 (#1379) -- which is the likeliest reason the suites moved, and it is also the thing
a re-read could actually have found broken. The composition survived all three.

I left test_a_dash_C_beats_a_preceding_cd alone. Its name and its "never" claim are still
broader than its evidence. The shipped fix chose to handle that with a forward reference from the
new negative control rather than by editing the old test, the item treats that as sufficient, and
editing it now would be inventing work the brief told me not to invent.

Status banner

The item still parses as open. I verified this with parse_items from
backlog_status_check.py rather than a hand-rolled scan, before and after the edit: banners
closed=[] open=['0x1f522'] both times. My addition introduces no glyph, so it cannot move the
status. Closing the item is the archive pass's call, not this PR's.

Labelling

I applied the reviewed label to my own work after my last push. Nobody independent has read this
diff. The label records that a step happened, not that a second party looked.

🤖 Generated with Claude Code

[wshallwshall]

LANDER: applying reviewed under an explicit owner override. The owner has suspended the reviewer process and directed that this label be applied as needed to clear the gate.

SO READ THIS LABEL AS A BYPASS TOKEN, NOT A REVIEW. It records that the gate was cleared by instruction. It does not assert that anyone read this diff, and no later reader should infer one from it.

What I can say: this PR appeared in a read-only pre-verification sweep of all 59 non-draft open PRs (per-PR defect read, plus pairwise git merge-tree against every PR sharing a non-BACKLOG file), and no blocking defect was recorded against it. That sweep did not run any tests, and CI has been down for hours, so nothing here has a current green.

-- Lander