Epic: qualify a stable v1.0 capability-review release from the exact shipped distribution

[pengfei-threemoonslab]

Release decision — 2026-09-08

No-go for v1.0 at main 452bdeb80. This issue owns the release decision and sequencing, not a second scanner verdict or an instruction to close every open epic. The audit covers all 45 open issues (the three open Dependabot PRs are separate), recent implementation, stability/distribution contracts, release workflows and the recorded user/corpus evidence.

What v1.0 promises

A developer or platform reviewer can review a supported repository-declared agent capability change, see exact base/head evidence and coverage limits, and follow a truthful next action. Local/static operation and advisory-first CI remain the defaults. The stable CLI, reports and control contracts must mean the same thing in the downloaded build and its generated CI.

Do not claim runtime-effective permission, universal MCP/tool coverage, proven organizational adoption, or authenticated GitHub continuation without the separate evidence those claims require. Existing human-owned stops remain human-owned.

Blocking work

Evidence and authority

• current-control can return merge authority after the workspace or comparison base changes during the read #567 — reconfirm full live workspace/base identity before returning current-control authority.
• Qualification: hash and parse the same artifact bytes before scoring release evidence #559 — hash and parse the same offline qualification bytes.
• Fixed-history release regression: establish reviewed scope for refused monorepo cases #563 — establish reviewed historical scope and restore the fixed-history catch bars on unchanged SHAs/labels. Miner: preserve new and renamed project directories when building the base verifier input #564's typed unsupported-input branch is delivered in Miner: retain original scoped base input obligations #582; the split engine gap is Verifier: bind absent and relocated scoped base inputs without inventing an empty agent surface #580 and no missing answer becomes a catch. The last full W37 run has 0/19 verifier results, 0/2 must-block catches and 1/3 needs-human catches using legacy scan fallback.

The selected product contract

• The first discovery decision rejects supported host-only repositories before they reach Route H #568 — a supported host-only repository reaches Route H rather than being rejected at discovery.
• Diff attribution: establish dependency coverage before excluding a standing finding #557 → Scope the verdict to the diff by default, with an opt-in whole-tree mode #515 — dependency evidence before default diff scope; unproven standing findings cannot disappear by fingerprint equality.
• Separate prose-only instruction touches from the generic trust-root review route #545 → Deprecate instruction-weakening judgments for prose edits; retain structured permission and execution changes #516 — remove unsupported prose judgments while retaining structured grants, hooks, CI and malformed-input protection.
• Coverage recovery: distinguish reader repair from missing input without inferring ownership from prose #561 — parser-backed coverage recovery distinguishes unavailable input, a bounded reader limitation and an unresolved expression without changing permissions; delivered in Explain coverage recovery from typed SDK source evidence #587.
• Expose the actionable host-inventory failure behind check's incomplete-input route #547 — host-boundary failures name the failed read phase/source and configured limits while denying authority; delivered in Name the failed host input and truthful recovery route #589.
• Relevant Preserve why agent.name is unresolved through init and doctor recovery #543 slice — test/template-only product identity remains an explicit setup decision after unrelated placeholders are resolved; product-backed recovery and actual manifest scope are preserved (Preserve product identity recovery from discovery evidence #594).
• Keep discovery's integration-surface enumeration in parity with the runtime contract #553 and Reconcile live qualification instructions and delivery contracts with the approved 38/80 policy #566 — current distribution/qualification instructions agree with the implemented contract; delivered in Discovery: match the same-build integration surface enumeration #583 and docs(release): align qualification instructions with approved policy #579.
• Diagnose intermittent authorization-context rejection in the missing-trust regression #548 receives a recorded reliability diagnosis before RC sign-off; Bound FastMCP injection claims to the SDK's whole-signature and generic-type rules #542 needs its explicit supported-boundary decision. Make the current-control boundary explicit for receipt-only optional artifacts #552's compact/closure decision is delivered in Define the compact-control and receipt-closure validation boundary #600; no bypass was established.

Freeze and actual release

• Freeze the report 1.0 contract and prove migration before calling the product v1.0 #569 — report 1.0 compatibility freeze and migration, after behavior convergence.
• A released wheel permanently generates CI pinned to the previous release #570 — final wheel and generated Action use the intended release engine; candidate smoke before the tag, real-channel smoke after publication.
• Deliverable 2: the 80-case human-labeled beta corpus required for 1.0 (#456) #512 — the current 80-case beta corpus, human primary labels, holdout/origin coverage, and terminal receipts on the exact final wheel.
• Cut D: exact-candidate evidence, independent qualification promotion and publication handoff (#456) #509 — exact-candidate artifact handoff, independent signer/trust root, signed qualification and location configuration.
• Rehearse substantive qualification rejection and the exact final candidate before publication (#456) #510 — substantive negative policy rehearsal and a positive rehearsal of the exact final candidate, with no publication from rehearsal.
• Enforce the repository and publication controls the release runbook already requires #573 — effective tag/asset/workflow/environment protection meets the existing release runbook, with actual independent review and dated read-back evidence.
• Define the v1.0 maintenance, security response and release recovery contract #494 — release/security ownership, support scope and recoverable publication. The accepted architecture record contradicts the product entering the v1.0 freeze #493's minimal accepted-decision corrections are delivered in Reconcile the accepted decisions before the v1.0 contract freeze #602; reconcile any later behavior change before Freeze the report 1.0 contract and prove migration before calling the product v1.0 #569. Broad ADR/refactor work is not a release gate.
• Run the existing Route H pilot on an installable candidate and decide the v1.0 adoption claim #571 — record the existing pilot's product decision and align launch claims with what was actually observed.

Dependency order and parallel work

1. Now: close the contract/document contradictions (Reconcile live qualification instructions and delivery contracts with the approved 38/80 policy #566); name owners for release, benchmark signing, human labeling, security/support and pilot execution. A role is an obligation, not an invented assignee.
2. Parallel engineering: current-control can return merge authority after the workspace or comparison base changes during the read #567; Qualification: hash and parse the same artifact bytes before scoring release evidence #559; Fixed-history release regression: establish reviewed scope for refused monorepo cases #563 and Miner: preserve new and renamed project directories when building the base verifier input #564; Diff attribution: establish dependency coverage before excluding a standing finding #557→Scope the verdict to the diff by default, with an opt-in whole-tree mode #515; Separate prose-only instruction touches from the generic trust-root review route #545→Deprecate instruction-weakening judgments for prose edits; retain structured permission and execution changes #516; The first discovery decision rejects supported host-only repositories before they reach Route H #568/A released wheel permanently generates CI pinned to the previous release #570 with bounded recovery. Keep discovery's integration-surface enumeration in parity with the runtime contract #553, minimum The accepted architecture record contradicts the product entering the v1.0 freeze #493/Define the v1.0 maintenance, security response and release recovery contract #494 and Enforce the repository and publication controls the release runbook already requires #573's effective publication controls and Pilot recovery: use a baseline path that satisfies the non-symlink identity contract on macOS #550's macOS pilot recipe can progress alongside them. Partition shared files when implementing.
3. Parallel evidence: start Deliverable 2: the 80-case human-labeled beta corpus required for 1.0 (#456) #512's beta sourcing and blind human labeling immediately. A 0.16 tag or completed 38-case pre-1.0 corpus is not a prerequisite for 1.0. An explicitly unqualified preview can support Run the existing Route H pilot on an installable candidate and decide the v1.0 adoption claim #571; its four-week repeat-change observation overlaps engineering/corpus work.
4. Converge: rerun the fixed history and supported cold-user/CI paths; finish Freeze the report 1.0 contract and prove migration before calling the product v1.0 #569, including active pre-1.0 tier retirement with historical readability before final wheel freeze. An unpublished RC can supply the compatibility exercise. Freeze the exact source, workflows, version, policy and wheel only after the selected behavior is settled.
5. Qualify and rehearse: collect final-wheel receipts, score the unchanged beta policy, independently sign/promote through Cut D: exact-candidate evidence, independent qualification promotion and publication handoff (#456) #509, and rehearse Rehearse substantive qualification rejection and the exact final candidate before publication (#456) #510. Negative fixture preparation can happen earlier. Any final candidate change invalidates affected receipts/rehearsal evidence.
6. Go/no-go and rollout: all pre-publication criteria met, product claims decided, release/support owner records the decision. Publish only the qualified bytes through the existing protected workflow. Complete A released wheel permanently generates CI pinned to the previous release #570's real tag/PyPI smoke and recovery checks before announcing rollout complete.

The 80-case beta policy retains 30 passed / 20 review-required / 30 blocked, at least 32 qualifying origins, κ≥0.80, per-stratum holdout≥20%, zero unsafe auto-passes, blocked exact30/30, safe≥27/30 and review≥19/20. Expected-IE is not a ground-truth class; actual IE remains a coverage/exact-score miss. The required report schema must deliberately follow #569's reviewed freeze. Neither a favorable pilot nor the 19-case history substitutes for beta qualification.

The existing external beta rollout condition — three distinct design partners and four weeks for affected profiles — is separate from the pilot decision ladder and machine qualification. A finite pilot shortfall or narrower launch language does not satisfy or waive it. Record evidence and applicability, or obtain an explicit reviewed policy decision. #570's pre-publication implementation/candidate smoke is a tag prerequisite; its real-channel smoke completes rollout after publication, so whole-issue closure is not required before the tag.

Conditional and deferred

#555→#337→#504 is mandatory only if authenticated GitHub recording/continuation is included in v1.0. The recommended first release withholds that claim; current request/evaluator/presentation do not establish independent human approval. #293 requires an actual host attestor. Keep #338 open for the broader autonomous workflow.

#511 remains non-gating participant validation. Do not block the release on all dependency bumps, general module decomposition, type/complexity budgets, schema relocation, a benchmark database, OWASP mapping, new adapters, committed organization state or a hosted control plane. Address a concrete candidate failure where demonstrated.

Done when

A linked release record identifies the exact candidate, unchanged approved beta evidence, named owners, compatibility/migration validation, both rehearsal outcomes, publication/channel verification and the explicit product claim. Every unresolved issue has an accurate hard/conditional/deferred classification. A milestone deadline, closed measurement issue, passing ordinary CI or missing private artifact in Git is never substituted for that record.

Release planning record

The full evidence audit and disposition of the original 45 open issues are in docs/engineering/v1-release-readiness.md. ROADMAP.md carries the 0–30 / 30–60 / 60–90-day parallel work and exact handoff order. Delivery PR: #574 (merged at 0aac28671e43586eec04eee14ddb9c6b04006bfd; one GitHub coding-agent review/address round and final CI, including combined coverage, passed). These capacity windows do not override the exit gates.

Implementation progress — 2026-09-08

• current-control can return merge authority after the workspace or comparison base changes during the read #567 delivered in fix(control): reject live input drift during the final authority check #576, merged 8e8d5530d: same-path and resolved-base drift now revoke stale current-control authority. One independent coding-agent GitHub review/address round, local full suite and all applicable CI including combined coverage passed.

• Qualification: hash and parse the same artifact bytes before scoring release evidence #559 delivered in fix(qualification): score the bytes bound by input digests #578, merged ded4be37f: qualification hashes and parses the same captured input bytes while preserving the detached archive format and approved scoring bars. One independent coding-agent GitHub review/address round, 151 targeted tests and all applicable CI including combined coverage passed.

• Newly reproduced Control refresh outside the workspace defaults to the caller's reports directory #575 (default report location) and Evidence readers can block on a FIFO before rejecting non-regular inputs #577 (FIFO open can block before type rejection) are separately tracked and deferred. These repairs do not establish beta labels, independent signing, rehearsal, pilot adoption or v1.0 readiness.

• Reconcile live qualification instructions and delivery contracts with the approved 38/80 policy #566 delivered in docs(release): align qualification instructions with approved policy #579, merged 6df9a871c: current qualification counts/schema/ordering and final-wheel instructions now agree with the approved policy; one independent GitHub review/address round, 102 targeted tests and all applicable CI passed.

• Miner: preserve new and renamed project directories when building the base verifier input #564 delivered in Miner: retain original scoped base input obligations #582, merged e99e5dc76: original Git base absence and incoming rename evidence are typed obligations before scoped manifest injection. Both fixed ADK cases reproduced on the exact PR head; one independent GitHub review/address round, 96 targeted tests and all applicable CI passed. Verifier: bind absent and relocated scoped base inputs without inventing an empty agent surface #580 (independent scoped base/head engine input) and Verifier: a scoped manifest under a directory containing spaces can crash adoption-path checks #581 (space-containing scoped path crash) remain deferred; Fixed-history release regression: establish reviewed scope for refused monorepo cases #563 remains unmet.

• The first discovery decision rejects supported host-only repositories before they reach Route H #568 is not implemented: the concrete required instruction/stop-rule change set was refused by the repository's proactive preflight as human_review_required for AGENTS.md. The dependency and bounded design are recorded in The first discovery decision rejects supported host-only repositories before they reach Route H #568; removing those consumers from acceptance would leave the original false-negative in place. Other independent roadmap implementation continues.

• Keep discovery's integration-surface enumeration in parity with the runtime contract #553 delivered in Discovery: match the same-build integration surface enumeration #583, merged c079864e8: discovery enumerates exactly the integration surfaces of the same running source contract. One independent GitHub review/address round, 831 focused tests and all applicable CI including combined coverage passed. The released site and Action remain pinned to their actual published version.

• Coverage recovery: distinguish reader repair from missing input without inferring ownership from prose #561 delivered in Explain coverage recovery from typed SDK source evidence #587, merged 1d4c0f841: missing SDK entrypoints, unsupported literal tool-list concatenation and unresolved expressions retain typed recovery evidence through CLI/report/PR and qualification. One independent GitHub coding-agent review/address round and all applicable CI including aggregate coverage passed. New OpenAI Agents SDK: resolve bounded literal tool-list concatenation in static binding evidence #584/Reconcile required missing SDK entrypoints with the documented scan input-error contract #585/Include source-omission redactions in the final privacy audit snapshot #586 are separately deferred; Expose the actionable host-inventory failure behind check's incomplete-input route #547 continues next.

• Expose the actionable host-inventory failure behind check's incomplete-input route #547 delivered in Name the failed host input and truthful recovery route #589, merged e3cf199ad: typed read facts reach existing inventory, boundary and control output with bounded/redacted recovery, stable failure identity and unchanged denial. One independent GitHub coding-agent review/address round, 9,087 full local tests (5 skipped), focused follow-up regressions and all applicable CI including aggregate coverage passed. New Identity reader: preserve the known failed entry and final-validation reason #588 (deeper reader entry/reason facts) and Redact sensitive host artifact paths without changing source identity or policy scope #590 (inherited artifact-path privacy) remain separately deferred. Diagnose intermittent authorization-context rejection in the missing-trust regression #548's original intermittent cause was not reproduced by the independent probe; bounded prerequisite diagnostics are the next implementation slice, not a claimed race fix.

• Diagnose intermittent authorization-context rejection in the missing-trust regression #548 diagnostic slice delivered in Identify the rejected authorization prerequisite #591, merged ad2055917: eleven fixed prerequisite codes, unchanged exact rejection assertions and no exception leakage; one independent GitHub coding-agent review/address round, 9,099 local tests (5 skipped), all applicable CI and coverage passed. The original intermittent cause remains unattributed, so Diagnose intermittent authorization-context rejection in the missing-trust regression #548 and the RC reliability obligation remain open. Preserve why agent.name is unresolved through init and doctor recovery #543 is now in implementation; new Define recovery for an agent name whose discovery coverage remains unresolved #592 owns unresolved/capped name recovery and Correct stale placeholder path examples in the diagnostics guide #593 owns a stale diagnostics example.

• Preserve why agent.name is unresolved through init and doctor recovery #543 delivered in Preserve product identity recovery from discovery evidence #594, merged 15b317a3: current discovery distinguishes product evidence from test/template-only identity, survives purpose-only edits and source failures, and resolves linked manifests to their actual source directory. Two independent GitHub review/address rounds, 9,130 full local tests (5 skipped), final-head CI and aggregate coverage passed. Define recovery for an agent name whose discovery coverage remains unresolved #592/Correct stale placeholder path examples in the diagnostics guide #593 remain deferred. Pilot recovery: use a baseline path that satisfies the non-symlink identity contract on macOS #550's pilot-path repair is under review in Use a canonical private path for pilot baselines #595; A released wheel permanently generates CI pinned to the previous release #570's reproducible installed-wheel/Action pin rule is the next release-provenance workstream.

• Pilot recovery: use a baseline path that satisfies the non-symlink identity contract on macOS #550 delivered in Use a canonical private path for pilot baselines #595, merged 47de3703: canonical private baseline paths in both pilot recipes, unchanged symlink refusals and base/change separation; one independent GitHub review/address round, 9,101 full local tests (5 skipped), all applicable CI and coverage passed. No adopter/pilot counts changed. A released wheel permanently generates CI pinned to the previous release #570's complete implementation plan is now human-routed by preflight for the required protected release-verification workflow (issue comment 5593846472); no implementation files were edited. Diff attribution: establish dependency coverage before excluding a standing finding #557's dependency-attribution prerequisite is the next independent implementation workstream. The v1.0 no-go and exact-wheel/external evidence obligations remain unchanged.

• Diff attribution: establish dependency coverage before excluding a standing finding #557's first bounded SDK guard evidence profile delivered in Compare imported SDK guard predicates with bound input evidence #599, merged 5c65f19c6: actual dependency bytes/candidates are identity-bound, predicates are compared, and incomplete coverage never excludes a finding. Two independent GitHub review/address rounds, 9,193 full local tests (5 skipped), final CI and aggregate coverage passed. Diff attribution: establish dependency coverage before excluding a standing finding #557/Scope the verdict to the diff by default, with an opt-in whole-tree mode #515 and benchmark: re-run the labeled corpus on the v0.18 candidate and compare with W27 #312's historical measurement remain open. New Base-scan cache does not distinguish reader changes within the same package version #596/Reconfirm refused dependency inputs without losing the exact refusal evidence #597/Local check routes an incomplete generated-cache inventory to human review during concurrent tests #598 are deferred.

• Separate prose-only instruction touches from the generic trust-root review route #545 has an isolated unpublished prototype only. Automatic approval review rejected the proposed removal of agent_instructions from the unconditional protected-edit deny list as an unauthorized persistent weakening; that operation was not applied, no PR/review loop is claimed, and the required deny-list/hook reconciliation is paused (issue comment 5594662077). Independent Make the current-control boundary explicit for receipt-only optional artifacts #552/Bound FastMCP injection claims to the SDK's whole-signature and generic-type rules #542/The accepted architecture record contradicts the product entering the v1.0 freeze #493 work continues. The first discovery decision rejects supported host-only repositories before they reach Route H #568/A released wheel permanently generates CI pinned to the previous release #570 retain their existing human routes; no release stop is cleared by review-subagent permission.

• Make the current-control boundary explicit for receipt-only optional artifacts #552 delivered in Define the compact-control and receipt-closure validation boundary #600, merged e6f754512: compact current-control validates its explicit pointer set, optional consumers validate the full receipt closure and join captured identity. No behavior/wire/authority change; real damage and I/O-boundary tests, one independent GitHub review/address round, all final CI and coverage passed. Bound FastMCP injection claims to the SDK's whole-signature and generic-type rules #542's whole-signature repair is in implementation; newly reproduced mixed-framework/exact-export provenance is separately deferred in Bind injected Context evidence to the selected MCP framework and exact export #601. The accepted architecture record contradicts the product entering the v1.0 freeze #493's accepted-decision correction is in implementation. Report freeze, exact-wheel qualification and external obligations remain unmet.

• The accepted architecture record contradicts the product entering the v1.0 freeze #493 delivered in Reconcile the accepted decisions before the v1.0 contract freeze #602, merged 0d5d339df: the current accepted-decision index replaces contradictory v0.1 guidance and traces shipped versus pending contracts. One independent GitHub coding-agent review/address round, 9,193 full local tests (5 skipped), final CI and aggregate coverage passed. Bound FastMCP injection claims to the SDK's whole-signature and generic-type rules #542 is under review in Bound FastMCP injection claims to whole-signature evidence #603; round 1 is checking an invalid-typing-arity signature gap before merge.

• Rehearse substantive qualification rejection and the exact final candidate before publication (#456) #510's complete negative-only rehearsal design was preflighted and routed to a human for .github/workflows/release-rehearsal.yml (issue comment 5595144224). No files were edited and no run was dispatched. It joins the existing The first discovery decision rejects supported host-only repositories before they reach Route H #568/A released wheel permanently generates CI pinned to the previous release #570 protected-surface dependencies; synthetic negative evidence cannot replace the actual final-candidate rehearsal or independent signing.

[pengfei-threemoonslab]

Delivery checkpoint — 2026-09-09

#542 is complete in merged PR #603 (5c5a4b94118c93e2493812689d3920212451f1e4). The bounded whole-signature/first-match injection repair and installed/standalone parity passed two independent coding-agent GitHub review/address rounds. The typing-arity defect found in round 1 was fixed; final full suite: 9,203 passed, 5 skipped; independent round-2 checks: 690 passed. Final CI including aggregate coverage and refreshed verifier merge authorization passed. GitHub read-back confirms the merge. #601 stays deferred; #548's original intermittent cause remains unresolved.

The existing ROADMAP.md is receiving a separate tested delivery-status PR under this epic. It retains all remaining release gates and the original audit. A fresh read-only check still reports zero qualification-location Actions variables and zero Release Rehearsal runs; latest published tag is v0.15.0. None of the implementation PRs supplies human labels, independent signing, pilot observations or v1.0 qualification.

This is appended to preserve the original release audit and issue history. An attempted full-body status update to this epic was rejected by automatic approval review as an apparent overwrite of the unrelated #542 issue; no such overwrite was performed. #542's focused content and all unresolved release obligations are preserved.

[pengfei-threemoonslab]

Roadmap delivery checkpoint — 2026-09-09

#604 is merged at 54b231ce00e11ecf62e16cdc075bb3bac40264f6. ROADMAP.md now links the 14 delivered implementation PRs, preserves the partial obligations in #564/#548/#557, names the protected/human routes in #545/#568/#570/#510 and lists the 15 separately deferred discoveries. The original audit, release bars and dependency order remain intact; this is delivery accounting, not qualification or a go decision.

Independent coding-agent review/address round: review 5149485646, response 5595431451. Full local suite: 9,203 passed, 5 skipped; final head 9cad7a23245a33ef910b6d2366a01aae01c63721 passed every applicable CI check, including aggregate coverage. Current-base verifier control was refreshed and complete before merging.

#572 remains open and v1.0 remains no-go. Actual owners, reviewed historical scope, completed attribution/convergence, report freeze, blind human labels, independent signing, effective platform protections, exact-candidate rehearsals and external product observations remain obligations. No release/tag was created.

[pengfei-threemoonslab]

Delivery checkpoint — 2026-09-09: #499 is complete in #605 (826557678271b5fdabb0727f2103b5ed1a00ce0c). All 24 current sample goldens now have a checked, reproducible source-tree recipe; contributor/schema/contract guidance uses it. Two independent coding-agent GitHub review/address rounds; the repeated-CRLF test defect was fixed, final full suite 9,291 passed / 5 skipped, final-head CI including aggregate coverage and current-base verification passed. This brings this implementation pass to 15 merged implementation PRs, plus roadmap checkpoints #574/#604.

The release decision remains NO-GO. Reproducible sample fixtures are not qualification. #569 still owns the report 1.0 freeze and migration fixtures. #557's complete dependency/finding proof, #515, the paused #545→#516 route, human-routed #568/#570/#510, independent labeling/signing, release controls and actual pilot evidence remain open. The existing roadmap sequence and safety bars are unchanged.

[pengfei-threemoonslab]

Delivery checkpoint — #606 is merged as 7a5abcdd897b71205d6d53e111b98b6b3e4e96ac. It adds whole finite-Boolean-function and literal source Agent membership comparisons beside #599's first-guard evidence. Final full suite: 9,347 passed / 5 skipped; final-head CI including aggregate coverage and current-main verification passed. Review 5149940982 / address 5596207129 resolved a boundedness defect, with independent confirmation of all final blobs. This is one loop in #606 and three for #557 across #599/#606, respecting the requested cap.

The implementation pass now has 16 merged implementation PRs. #557 and #515 remain open. A true return is not an action effect or approval; new deferred #607 is formally a sub-issue of #557 and owns operation-to-finding-predicate attribution. It is one of 16 newly deferred issues, not a delivered exclusion feature or a weakened release bar. The ROADMAP checkpoint is being updated to show this dependency explicitly.

v1.0 remains NO-GO: source comparison does not supply reviewed historical scope, exact-beta qualification, report 1.0 freeze, independent human signing, actual release controls/rehearsals or product observation. The protected-edit human routes already recorded on #545/#568/#570/#510 are unchanged.

[pengfei-threemoonslab]

Delivery checkpoint — 2026-09-09, after #608 merged as 42818ec. The release decision remains NO-GO and #572 remains open.

The current ROADMAP.md now records the delivered repairs, incomplete obligations and actual dependency order. #608 received independent COMMENT review 5150151743 and author response 5596393907 at its final head 62b9298. All applicable final-head CI, including aggregate coverage, passed; release-tag-consistency was the expected PR-only skip. Current-control was freshly validated as complete before merge. Local validation was 9,347 passed / 5 skipped before the final documentation wording clarification, followed by 576 passing documentation/public-contract checks on that wording.

This implementation pass merged 16 implementation PRs: #576, #578, #579, #582, #583, #587, #589, #591, #594, #595, #599, #600, #602, #603, #605 and #606. Their linked reviews and author responses record the required 1–3 coding-agent review/address loops per issue. In particular, #557 has three total loops across #599 and #606; the final fix recheck in #606 belongs to its existing address loop, not a fourth formal review. The roadmap checkpoints are separate documentation changes.

Sixteen newly discovered problems remain filed and deferred as requested: #575, #577, #580, #581, #584, #585, #586, #588, #590, #592, #593, #596, #597, #598, #601 and #607. #607 is explicitly connected beneath #557. Deferring it from this implementation pass does not satisfy the operation-to-finding-predicate prerequisite or make #557/#515 release-complete.

Remaining routes:

• Diff attribution: establish dependency coverage before excluding a standing finding #557, including Connect modeled operations to finding predicates before excluding standing findings #607, precedes Scope the verdict to the diff by default, with an opt-in whole-tree mode #515; the finite source comparison still cannot exclude a standing finding or satisfy the TypeScript/historical acceptance cases. Diagnose intermittent authorization-context rejection in the missing-trust regression #548's original intermittent cause also remains unresolved.
• Separate prose-only instruction touches from the generic trust-root review route #545's proposed persistent protected-edit adjustment was rejected by automatic approval review and not applied. Its conditional instruction-structure boundary requires an explicit decision before dependent Deprecate instruction-weakening judgments for prose edits; retain structured permission and execution changes #516 work.
• The complete necessary plans for The first discovery decision rejects supported host-only repositories before they reach Route H #568, A released wheel permanently generates CI pinned to the previous release #570 and Rehearse substantive qualification rejection and the exact final candidate before publication (#456) #510 reached protected-surface human-review stops. Their issue comments preserve the exact plans and refusals; there was no alternate-route bypass, protected edit, release dispatch or publication.
• Fixed-history release regression: establish reviewed scope for refused monorepo cases #563 requires actual reviewed scope/declarations. Deliverable 2: the 80-case human-labeled beta corpus required for 1.0 (#456) #512/Cut D: exact-candidate evidence, independent qualification promotion and publication handoff (#456) #509/Define the v1.0 maintenance, security response and release recovery contract #494/Enforce the repository and publication controls the release runbook already requires #573 and the pilot Run the existing Route H pilot on an installable candidate and decide the v1.0 adoption claim #571 require real independent people, evidence, ownership or platform controls, not coding-agent review identities.
• Converged behavior and candidate provenance precede Freeze the report 1.0 contract and prove migration before calling the product v1.0 #569's report 1.0 freeze/migration work, then exact-wheel qualification, independent signing and the actual final-candidate rehearsal. The post-publication smoke remains a separate A released wheel permanently generates CI pinned to the previous release #570 obligation.

The v1.0 milestone still has 16 open issues. No acceptance bar, gate, qualification label or release claim was relaxed to reduce that count. This checkpoint appends delivery evidence to the existing epic; it does not replace its original audit or assert v1.0 readiness.