docs(getff-any-stack-trace): S4 dispatch input — acceptance mechanisms + closure

[artyhoo]

Dispatch input for getff-any-stack-trace S4 — the umbrella's last stage. Staging placement is the precondition for the aif dispatch that follows (kickoff-staging-placement.md §1).

Its gate is S3 merged — gh pr list --search "is:merged head:feature/getff-any-stack-trace-s3 base:staging" returns #1253 (mergedAt 2026-08-07T09:51:31Z), so this is the first moment S4 could be planned against a tree where the F-A decision exists.

What the stage owns (spec §9.1 + §9.3)

W6 cell v1 — scripted fresh python project on a master default branch → install.sh python → agent-surface asserts → committed Tier-1 fixture → generation → RED/GREEN firing + the §4.4 reject fixture — plus the named one-beat cold-run protocol (agents/ class, session-read, $0-in-CI), run at closure with its verdict quoted, plus the umbrella done.md that gates getff-freshness-widening.

Anchors re-verified live at ea5f7468d5, not carried from the spec

Anchor	Live state
.github/workflows/audit-self.yml:1461	consumer-matrix-start-cell: — the spec §9.1 and the umbrella kickoff both cite :1319; stale (file is 1546 lines). Recorded in the kickoff as a standing T3 proof.
.github/workflows/audit-self.yml:1523	- consumer-matrix-start-cell in the ci-success needs: list — named load-bearing, with the :1524-1529 alwayson-budget incident comment as the precedent for why an unlisted cell is a gate nobody reads
tests/consumer-matrix/pnpm-monorepo-cell.sh	267 lines, the only file in the directory; its :20-46 header enumerates asserts and deliberate deferrals
setup.d/45-python.sh:896	for _py_agent in rule-researcher rule-test-author — the lane's curated 2-agent set, so the cell's surface assert matches python, not npm
install.sh:606-618	the sub-agent copy glob with its explicit authoring-only skip-list; install.sh:204-212 is the shipped-header-verify set
packages/core/principles/09-doc-authority-hierarchy.ts:252	auto-covers every *.md directly under agents/ — the header is owed whether or not the protocol ships
agents/rule-researcher.md:204-211, :212-221	the S3 F-A DECLARE binding site — Node at generation time, Node-free install

Three decisions handed to the stage with criteria, not left implicit

1. Ship-or-not for the protocol artefact — default framework-only (a continue at install.sh:606-618, matching shipped-agent-liveness-prober.md, because the protocol is run by the framework against a consumer), with the baseline consequence of the other branch stated.
2. The cell's determinism boundary — a park trigger rather than a guess, with || true / SKIP / continue-on-error named as the tempting non-answers. Fail-closed: a missing tool is RED, never SKIP (S2b shipped a container-green suite whose load-bearing arm had silently SKIPped for want of ast-grep/ruff).
3. Running the protocol from a container — park the RUN, not the authoring, if a genuinely cold agent + fresh consumer cannot be produced. A warm run reported as cold is T-AST-B, the domain trap this umbrella named against itself.

Source-before-shape obligation carried into the kickoff

The protocol artefact is capability-shaped, so §1 names the two in-repo probers — agents/shipped-agent-liveness-prober.md and agents/manual-rule-liveness-prober.md (SSOT #115 is the ADAPT source of the RED→GREEN methodology both use) — as a mandatory overlap consult with an explicit T16 problem-class statement required before the body is written (source-before-shape.md §1).

R1 (#1254) routed as a read, not a fix

R1 §5.1 parked whether the lane CI templates should substitute the consumer's real default branch. The cell's master-default fixture settles that question for the python lane mechanically — honest-signals S4 (#1167) fixed it at delivery via deliver_getff_workflow (setup.d/lib.sh:194, called at setup.d/45-python.sh:384), leaving [main] in the template as the literal being substituted. The kickoff asks for the assertion and the result either way, and forbids widening to cargo/go (R1's routing, owned by widening).

Domain traps added (principle 12 §3 requires ≥1; there are three)

T-S4-A the cell that asserts the install instead of the loop — a cell stopping at «agent surface present» re-implements the entry-lane test and proves nothing about W1-W3. T-S4-B done.md as a success narrative — it gates the next umbrella, so omitted descopes hand it a false baseline. T-S4-C editing the protocol until it passes — the first run is evidence about the shipped docs, and a failing cold agent yields a finding or a park, never a quiet loosening of the cold-start conditions.

Local validation

npx vitest run --root packages/core principles/12-ai-laziness-traps.test.ts   7/7 pass
bash scripts/host-verify.sh --list getff-any-stack-trace-s4                   rc=0, resolves the 4-command contract
markdownlint-cli2 (pre-commit)                                                0 errors
pre-push: link check 8 OK / 0 errors; always-on 48671B within ceiling 54000B

Fidelity verdict

FIDELITY: skipped — dispatch-input authoring only; the diff is one kickoff markdown file under .claude/orchestrator-prompts/, no implementation surface and no Provenance substrate declared. It was validated against principle 12 and the host-verify contract grammar before push, and will itself be the Basis of a cold fidelity audit when S4 lands.

§1.7 Skipped: the diff touches only .claude/orchestrator-prompts/**, which is not a §4b trigger path in either channel (the union is .claude/rules/**, packages/core/principles/**, EXECUTION-PLAN.md, prior-art-evaluations.md, CLAUDE.md, packages/core/templates/**, .claude/skills/**, agents/**). No discipline is introduced or extended here — this is a dispatch input that cites existing rules rather than changing any.