Milestones
List view
Primary product track under #778. Latest recorded baseline (2026-09-30, main 6ced6f70): 1 Q2 / 49 development PRs, 9 with rows, zero false-complete answers in that sample. #908 preserves development evidence and freezes the untouched holdout before gated reader merges; record the post-Oct-1 sampling constraint. Sequence #909 + #874, #910 (including #866), #913 + #914, then #911/#912. #915 can deliver current truthful text before #914; #916 is research, not a reader commitment. #580/#864/#865/#872/#875/#876 are delivered; #655/#867 need acceptance reconciliation. Exit: #868 ten qualifying Q2 cases with development/holdout counts on the released build. Oct 14 remains an evidence checkpoint. Owner pengfei-threemoonslab; name implementation owners at pickup. No adoption, qualification or outreach claim follows from engineering delivery.
Due by October 14, 2026•8/20 issues closedOutcome: establish automatic value, then earn actual reviewer value and voluntary use on the next eligible change. Sequence: exercised #796 route -> #830 V1 on the released Phase 2 build -> explicit owner decision -> only then #653/#571 external execution within the existing per-note/consent rules. Readiness and gate completion do not themselves establish adoption. The phase's external-value outcome follows the gate; it is not a circular prerequisite to running it. Keep #830's strata, novelty, yield, honesty, precision/noise and release requirements unchanged. Keep #571's separate first-value definition and full invited/attempted/valid/value/opportunity/use denominators. Record real decisions, correction interpretation, prompting and actual reviewer consumption; automatic CI and constructed recurrence are not voluntary use. The individual four-week window starts at first value. No next eligible change means unobserved. Existing same-route ladder in order: Continue >=2 unaided first values and >=1 observed second eligible change; Stop first_value=0 with first_valid_result>=1; Narrow otherwise. Original checkpoint 2026-12-13 is a dated evidence review, not a promise of participants. If gated or incomplete, report the shortfall without changing thresholds or resetting clocks. App/new hosts remain conditional.
Due by December 13, 2026•0/3 issues closed2026-09-24 selected execution: application-agent PR proof (#868) is primary; host work is bounded maintenance. Days 1–5 evidence/design: https://github.com/ThreeMoonsLab/agents-shipgate/pull/869. Accountable product owner: pengfei-threemoonslab; technical executor: Codex current task. Oct 14 / Nov 13 / Dec 13 remain evidence checkpoints. #830 and external outreach holds are unchanged. #780 now has ten published v1.1.0 hosted runs; its remaining hosted cases are a no-change update to an existing comment, Not compared, and setup/execution failure. #853 merged. #819/#821/#823 are closed; preserve their delivered work. Historical scope (superseded where the dated update above differs): Outcome: one supported permission change supports an explicit advisory decision and verifiable correction. #811 owns the experience; #839 supplies evidence-bound questions/conditional next steps; #840 verifies the selected static target against the same original B. Reuse CLI/JSON/PR surfaces and preserve B/C/F/R, actual engine/config identities, coverage, remaining/new deltas and negative controls. Intentional retention is valid; lost coverage or a moved base cannot prove a fix. Existing engine order and dependencies remain in #778: #819/#820/#821, #808/#822, #809/#827, bounded readers/notes and #702, then #811. #780/#570 hosted replay remains owner-deferred. No new verdict, approval ledger or automatic policy. #504/#337/#555 retain authenticated continuation. Original checkpoint 2026-11-13. Internal independent interpretation and installed replay establish readiness, not external value. Under the Sept 16 owner sequence, external recruitment waits for released Phase 2 evidence, #830 and explicit owner approval. Record shortfalls separately from engineering completion; do not require a deferred participant as a prerequisite to running the gate.
Due by November 13, 2026•19/26 issues closed2026-09-24 selected execution: application-agent PR proof (#868) is primary; host work is bounded maintenance. Days 1–5 evidence/design: https://github.com/ThreeMoonsLab/agents-shipgate/pull/869. Accountable product owner: pengfei-threemoonslab; technical executor: Codex current task. Oct 14 / Nov 13 / Dec 13 remain evidence checkpoints. #830 and external outreach holds are unchanged. Historical scope (superseded where the dated update above differs): Outcome: a concrete, correctly scoped permission change is understandable without reconstructing source facts. #795 slice 1/#837 and #812 slice 1/#838 are merged; finish remaining source/package evidence, #792 entry and #781 released-kit closeout. #802/#810/#804/#813/#816/#817/#818 are delivered dependencies. #798 supplies the first supported case. Internal cold-reader exercises establish readiness only. Program #778; original checkpoint 2026-10-14. #653 remains here for preparation and the checkpoint, but external recruitment is deferred under the Sept 16 owner decision until Phase 3 #830 passes and the owner approves. Record a dated shortfall; do not extend the program silently or count internal readers as adoption. The phase does not require outreach-dependent observations before its engineering prerequisites can finish. Product/report/release roles; name accountable owners at pickup. No account, new manifest or authored policy on the first route. Source, released wheel and hosted evidence are distinct. Dates are evidence reviews, not publication or participant promises.
Due by October 14, 2026•10/18 issues closed2026-09-24 selected execution: application-agent PR proof (#868) is primary; host work is bounded maintenance. Days 1–5 evidence/design: https://github.com/ThreeMoonsLab/agents-shipgate/pull/869. Accountable product owner: pengfei-threemoonslab; technical executor: Codex current task. Oct 14 / Nov 13 / Dec 13 remain evidence checkpoints. #830 and external outreach holds are unchanged. Historical scope (superseded where the dated update above differs): Program container for #778 and discovery workstream #791. Phase 1 (milestone 10) establishes truthful facts; Phase 2 (11) completes decision guidance and original-base correction verification through #811, #839 and #840; Phase 3 (12) runs #830's released-build value gate before owner-approved external observations. The Sept 16 owner sequence supersedes earlier parallel-outreach wording. Original Sept 14 clock and Oct 14 / Nov 13 / Dec 13 checkpoints remain; participant windows start at first value. #571's definitions/ladder and #830's independent novelty/yield thresholds are unchanged. Remaining items are conditional, not promised App/host/runtime expansion. Issue assignment and code delivery establish no adoption.
No due date•5/20 issues closedAdvisory v1.0.0 published with exact wheel, explicit advisory claims and current pins. #646/#570 stay open only for the literal post-publication hosted tag/Action replay and linkage of its evidence. #572 retains separate qualified obligations; external research #653/#571 can start without waiting on this follow-up.
Due by December 9, 2026•3/3 issues closedDeferred qualification work, distinct from the published advisory product. #572 coordinates remaining evidence and authority obligations; #456/#512 corpus, #509 promotion/signing, #510 rehearsal, #511 validation, #520 labeling/coverage, #515/#557/#563/#640 scope/attribution and #548 diagnosis. No delivery date or qualification clearance is implied. Current advisory application proof is #868 / milestone 13; the product program is #778. Historical publication record: Advisory v1.0.0 is published; #643 records the accepted narrow scope and below-bar recall/benign measurements. #646/#570 retain the literal post-publication hosted Action replay. Completed install/workflow/publication work is recorded in #644/#645/#648. Remaining qualified claims/evidence are distinct in #572; post-release adoption execution is #778 in milestone 9.
No due date•24/36 issues closedRelease integrity and gate correctness. The final tag requires an approved qualification policy, tagged-source/qualified-wheel binding, deterministic rehearsal and test evidence, a recoverable publication transaction, and a wheel-scoped signed SBOM.
No due date•13/15 issues closed