[loopx-agent]

Goal And Delivered Outcome

• Outcome basis / optional anchor: personal-loopx-maintenance repair for a live heartbeat blocked by a false prior-Turn closeout obligation.
• Goal/source and gap: owedPriorTurnFromSnapshot() checked only advancement writeback/spend receipts. A prior Turn with an exact committed quota_monitor_poll therefore remained a permanent owed closeout even though monitor polls are durable no-spend closeouts. Every later Turn could be refused as receipt_unbound, inviting a duplicate writeback/spend solely to escape the loop.
• Observable before → after: a later unbound Turn was told that the earlier monitor Turn still owed durable_writeback_receipt and quota_spend_receipt. It now recognizes the exact committed monitor-poll identity and reports no prior-Turn debt, while unrelated or malformed poll identities still fail closed.
• Issue/task and intended base: personal-loopx-maintenance; target loopx-project/loopx:main.

Author Declaration

• Written by: GPT-5 model agent in OpenAI Codex.

Implemented against

• Specification and revision: current control-plane settlement contract and monitor closeout semantics at f8238e6f26e801a65c8c6a16bc9aea1659bdb690.
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
A committed monitor poll closes its exact Turn without writeback/spend	implemented	committedMonitorPollFromSnapshot() reuse in owedPriorTurnFromSnapshot()	new TypeScript prior-Turn regression
Wrong Turn/Todo/agent/goal/effect remain rejected	retained	existing exact identity checks	quota_settlement_readback.test.ts
No fake receipt or spend is introduced	implemented	no mutation added to readback	monitor-followthrough suite; source CLI live readback

• Self-check before submission: reproduced the live stuck guard, verified the source fix against the same persisted state, and confirmed unrelated tests that fail do so identically on untouched origin/main.

Scope And Continuation

• Completed scope and remaining work: complete within this repair. The PR changes one settlement readback reducer and adds one focused regression test.
• Slice boundary / successor: N/A; this removes the blocking false debt without changing monitor cadence, Todo state, authority or spend.

Validation

• Tested revision: 3d62dd985a56ae89c0aa745c2950aeaaa39b37de
• Run state: finished
• Input classes: synthetic, public_fixture

Check kind	Result	Public-safe evidence / limitation
unit	passed	node --no-warnings --experimental-strip-types --test tests/control_plane_ts/quota_settlement_readback.test.ts: 135 passed.
integration	passed	pytest -q tests/control_plane/test_monitor_followthrough_contract.py: 22 passed.
real_entrypoint	passed	The patched source CLI re-read the live stuck runtime and returned a bound normal run with no unsettled_host_turn_recovery.
manual	not_applicable	No UI or user-visible presentation changed.
manual	blocked	tests/control_plane/test_interrupted_turn_continuation.py has two pre-existing fixture failures on untouched origin/main; not used as proof for this change.

The standard premerge gate passed all 13 selected checks with zero failures and zero manual holds.

• Coverage and gaps: the regression reproduces the exact prior monitor-poll plus later unbound Turn shape. Existing malformed/preview/wrong-identity cases continue to return open/poll-due behavior. No PostgreSQL-specific path is involved in this read-only reducer.

Type of Change

• Bug fix
• New feature
• Breaking change
• Refactoring (no functional changes)
• Documentation update
• Test update

LoopX Area

• Control plane (goals, todos, quota, scheduler, registry, runtime)
• Benchmark boundary (adapters, runners, verifiers, scoring, evidence)
• Capability or extension (providers, adapters, skills)
• Public docs or presentation surface (README, protocols, dashboard)
• Build, packaging, installer, or CI
• Host or runtime integration

Technical Direction

• Direction / acceptance reference, when applicable: Core control-plane hardening; exact receipt identity and no-spend monitor settlement.

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A
• Semantic dimensions changed, or reviewed no-impact rationale: N/A; no vocabulary or persisted state shape changed.
• Provider conformance arms run: N/A
• Read-only legacy/file/PostgreSQL three-arm rehearsal: N/A; file/SQLite existing settlement tests remain green.

Boundary Checklist

• Neither the diff nor this PR body/comments/attachments disclose private state, credentials, raw traces or verifier output, internal links, or local machine paths.
• I did not duplicate maintainer-owned benchmark work.
• I kept the change scoped to the closeout reducer.
• No English/Chinese RFC pair changed.
• UI impact is none.
• The commit includes a DCO Signed-off-by trailer.

[loopx-agent]

Reviewer: model_agent · GPT-5 · OpenAI · self_reported

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

审查精确 head：d18e81bc11bde151fa0e1bc3193ec8fcc98f36d8。

动机

monitor_only 或 due monitor 的 Agent lane，以及依赖 heartbeat 连续性的维护者。

上一 Turn 提交精确 monitor poll 后，旧代码把该 Turn 报告为仍欠 durable_writeback_receipt 和 quota_spend_receipt；新代码在同一 snapshot 上先按 Goal、Agent、Turn、Todo 和 effect 验证 committed monitor poll，验证通过即视为已关闭，不再把假债务带到当前 Turn。

新的回归用例覆盖“上一 Turn 精确 committed monitor + 当前 Turn 尚未绑定”，当前 Turn 仍正确报告 receipt_unbound，但不再携带 owed_prior_turn；既有 wrong Turn/Todo/agent/goal/effect 负例继续失败关闭。

不修改 monitor cadence、Todo 生命周期、quota 扣费、CLI 参数或持久化 schema；不把 preview、缺失 commit 或错误 identity 的 observation 当成关闭凭据。

改动思路

读路径从 readQuotaSettlement 进入 resolveIdentity。当当前 Turn 的 receipt 尚未绑定时，owedPriorTurnFromSnapshot 会遍历更早、且明确要求 closeout 的候选 Turn。旧实现只向 writeback/spend 链寻找结算；新实现先复用 committedMonitorPollFromSnapshot，用同一份 snapshot 验证 monitor run 的 classification、owner、Goal、Agent、Turn、Todo 与 committed effect。精确匹配后，该候选按现有合同视为 no-spend closeout；否则继续走原来的 fail-closed 读回。

这个边界是可逆且比例合适的：生产代码只改一个 reducer 分支，没有新增状态、调度、权限、CLI 参数或 schema。修复属于 settlement invariant，放在既有 readback owner；没有新增机制或默认策略。当前 PR 边界：只修正 prior-Turn closeout 消费，不扩展 monitor 产品流程。monitor 的产生、到期、材料变化和 Todo 生命周期仍由原 owner 管理。

具体改动

关键代码讲解

• owedPriorTurnFromSnapshot（loopx/control_plane/quota/settlement_readback.ts:1071）在 Todo-bound 候选上调用现有 committed-monitor helper。匹配时不再查询不适用于 monitor 的 writeback/spend；不匹配时保留原 missingCloseoutReceipts 行为。
• committedMonitorPollFromSnapshot（同文件 :420）没有扩张语义；它继续要求 quota_monitor_poll classification、owner、Goal、Agent、Turn、Todo 和精确 effect 全部一致。这里的关键不是新建“monitor 已读”状态，而是让另一个既有消费者复用同一 authoritative fact。
• 新回归 a prior committed monitor poll never blocks a later unbound Turn rebind（tests/control_plane_ts/quota_settlement_readback.test.ts:979）构造“上一 monitor Turn 已提交、当前 Turn 仍未绑定”的交叉快照，并断言当前 Turn 仍为 receipt_unbound，但不再附带 owed_prior_turn。

Spec ref: docs/quota-allocation.md @ revision 630c68a1cb871996fbe491292e95681c1c9b87c3。heartbeat settlement 条款位于 docs/heartbeat-automation-prompt.md，但本条 immutable review basis 固定到前述 quota contract revision。

• Q-MONITOR-CLOSEOUT implemented：精确 committed monitor poll 作为 no-spend closeout。
• Q-EXACT-EFFECT implemented：preview、缺失或错配 metadata 不能关闭 Turn。
• Q-PRIOR-RECOVERY implemented：same-Turn 与 prior-Turn 读取复用同一 exact effect 身份。

对主干的风险

最强反例是匹配过宽：只要同 Turn 存在类似 monitor 的行就错误关闭真实缺口。当前 predicate 对 owner、Goal、Agent、Turn、Todo、classification 与 effect 逐项匹配；既有 negative table 还覆盖 wrong Agent/Goal/Todo/Turn/effect、preview、空 commit 与 malformed commit，均要求 Turn 保持 open。新增用例同时证明修复没有伪造当前 Turn 的 binding。

验证为 135/135 TypeScript settlement tests、22/22 Python monitor follow-through tests，以及 canary premerge 的 13/13 项检查，零失败、零 manual hold。剩余未验证项是本机私有 live runtime 的再次回放；这不影响此次只读 reducer 的合成与仓库原生门禁证据，也不应外推为模型遵循或 UI 证明。

我的整体评价

Long-horizon impact 为 improved：已关闭的 monitor 不再让后续 heartbeat 反复承担假债务；当前未绑定 Turn 仍会明确要求自己的重绑定，恢复语义没有被删除。User experience 也为 improved，因为 Agent 和维护者不再被不存在的历史欠账引向重复写回或扣额。没有 blocking finding；主要残余风险是未来有人削弱 exact-effect predicate，现有 identity 负例应继续保留。

批准该精确 head。合并仍需要独立的 exact-head readiness 与仓库 merge authority；本评论不授予合并权限。

English verdict: APPROVE at d18e81bc11bde151fa0e1bc3193ec8fcc98f36d8; exact committed monitor polls now close their prior Turn without suppressing real identity failures. 135 TypeScript tests, 22 monitor integration tests, and 13/13 premerge checks passed. No blocking finding; remaining risk is limited to the private live-runtime replay and future weakening of the exact-effect guard.

[loopx-agent]

Reviewer: model_agent · GPT-5 · OpenAI · self_reported

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

审查精确 head：3d62dd985a56ae89c0aa745c2950aeaaa39b37de。

动机

monitor_only 或 due monitor 的 Agent lane，以及依赖 heartbeat 连续性的维护者。

上一 Turn 提交精确 monitor poll 后，旧代码把该 Turn 报告为仍欠 durable_writeback_receipt 和 quota_spend_receipt；新代码在同一 snapshot 上先按 Goal、Agent、Turn、Todo 和 effect 验证 committed monitor poll，验证通过即视为已关闭，不再把假债务带到当前 Turn。

新的回归用例覆盖“上一 Turn 精确 committed monitor + 当前 Turn 尚未绑定”，当前 Turn 仍正确报告 receipt_unbound，但不再携带 owed_prior_turn；既有 wrong Turn/Todo/agent/goal/effect 负例继续失败关闭。

不修改 monitor cadence、Todo 生命周期、quota 扣费、CLI 参数或持久化 schema；不把 preview、缺失 commit 或错误 identity 的 observation 当成关闭凭据。

改动思路

读路径从 readQuotaSettlement 进入 resolveIdentity。当当前 Turn 的 receipt 尚未绑定时，owedPriorTurnFromSnapshot 会遍历更早、且明确要求 closeout 的候选 Turn。旧实现只向 writeback/spend 链寻找结算；新实现先复用 committedMonitorPollFromSnapshot，用同一份 snapshot 验证 monitor run 的 classification、owner、Goal、Agent、Turn、Todo 与 committed effect。精确匹配后，该候选按现有合同视为 no-spend closeout；否则继续走原来的 fail-closed 读回。

这个边界是可逆且比例合适的：生产代码只改一个 reducer 分支，没有新增状态、调度、权限、CLI 参数或 schema。修复属于 settlement invariant，放在既有 readback owner；没有新增机制或默认策略。当前 PR 边界：只修正 prior-Turn closeout 消费，不扩展 monitor 产品流程。monitor 的产生、到期、材料变化和 Todo 生命周期仍由原 owner 管理。

具体改动

关键代码讲解

• owedPriorTurnFromSnapshot（loopx/control_plane/quota/settlement_readback.ts:1071）在 Todo-bound 候选上调用现有 committed-monitor helper。匹配时不再查询不适用于 monitor 的 writeback/spend；不匹配时保留原 missingCloseoutReceipts 行为。
• committedMonitorPollFromSnapshot（同文件 :420）没有扩张语义；它继续要求 quota_monitor_poll classification、owner、Goal、Agent、Turn、Todo 和精确 effect 全部一致。这里的关键不是新建“monitor 已读”状态，而是让另一个既有消费者复用同一 authoritative fact。
• 新回归 a prior committed monitor poll never blocks a later unbound Turn rebind（tests/control_plane_ts/quota_settlement_readback.test.ts:979）构造“上一 monitor Turn 已提交、当前 Turn 仍未绑定”的交叉快照，并断言当前 Turn 仍为 receipt_unbound，但不再附带 owed_prior_turn。

Spec ref: docs/quota-allocation.md @ revision 2ad0e88be2859e3a53836c32e9f467b2b40ad377。heartbeat settlement 条款位于 docs/heartbeat-automation-prompt.md，但本条 immutable review basis 固定到前述 quota contract revision。

• Q-MONITOR-CLOSEOUT implemented：精确 committed monitor poll 作为 no-spend closeout。
• Q-EXACT-EFFECT implemented：preview、缺失或错配 metadata 不能关闭 Turn。
• Q-PRIOR-RECOVERY implemented：same-Turn 与 prior-Turn 读取复用同一 exact effect 身份。

对主干的风险

最强反例是匹配过宽：只要同 Turn 存在类似 monitor 的行就错误关闭真实缺口。当前 predicate 对 owner、Goal、Agent、Turn、Todo、classification 与 effect 逐项匹配；既有 negative table 还覆盖 wrong Agent/Goal/Todo/Turn/effect、preview、空 commit 与 malformed commit，均要求 Turn 保持 open。新增用例同时证明修复没有伪造当前 Turn 的 binding。

rebase 到最新 main 后重新执行：135/135 TypeScript settlement tests、22/22 Python monitor follow-through tests，以及 canary premerge 的 13/13 项检查，零失败、零 manual hold。剩余未验证项是本机私有 live runtime 的再次回放；这不影响此次只读 reducer 的合成与仓库原生门禁证据，也不应外推为模型遵循或 UI 证明。

我的整体评价

Long-horizon impact 为 improved：已关闭的 monitor 不再让后续 heartbeat 反复承担假债务；当前未绑定 Turn 仍会明确要求自己的重绑定，恢复语义没有被删除。User experience 也为 improved，因为 Agent 和维护者不再被不存在的历史欠账引向重复写回或扣额。没有 blocking finding；主要残余风险是未来有人削弱 exact-effect predicate，现有 identity 负例应继续保留。

批准该精确 head。合并仍需要独立的 exact-head readiness 与仓库 merge authority；本评论不授予合并权限。

English verdict: APPROVE at 3d62dd985a56ae89c0aa745c2950aeaaa39b37de; exact committed monitor polls now close their prior Turn without suppressing real identity failures. 135 TypeScript tests, 22 monitor integration tests, and 13/13 premerge checks passed. No blocking finding; remaining risk is limited to the private live-runtime replay and future weakening of the exact-effect guard.

[loopx-agent]

Reviewer: model_agent · GPT-5 · OpenAI · self_reported

Approval conclusion (author-owned PR; GitHub blocks formal self-approval)

审查精确 head：d3a21cc20ee159aaa18c3cd9c628a7f8b3f88c7e。

动机

monitor_only 或 due monitor 的 Agent lane，以及依赖 heartbeat 连续性的维护者。

上一 Turn 提交精确 monitor poll 后，旧代码把该 Turn 报告为仍欠 durable_writeback_receipt 和 quota_spend_receipt；新代码在同一 snapshot 上先按 Goal、Agent、Turn、Todo 和 effect 验证 committed monitor poll，验证通过即视为已关闭，不再把假债务带到当前 Turn。

新的回归用例覆盖“上一 Turn 精确 committed monitor + 当前 Turn 尚未绑定”，当前 Turn 仍正确报告 receipt_unbound，但不再携带 owed_prior_turn；既有 wrong Turn/Todo/agent/goal/effect 负例继续失败关闭。

不修改 monitor cadence、Todo 生命周期、quota 扣费、CLI 参数或持久化 schema；不把 preview、缺失 commit 或错误 identity 的 observation 当成关闭凭据。

改动思路

读路径从 readQuotaSettlement 进入 resolveIdentity。当当前 Turn 的 receipt 尚未绑定时，owedPriorTurnFromSnapshot 会遍历更早、且明确要求 closeout 的候选 Turn。旧实现只向 writeback/spend 链寻找结算；新实现先复用 committedMonitorPollFromSnapshot，用同一份 snapshot 验证 monitor run 的 classification、owner、Goal、Agent、Turn、Todo 与 committed effect。精确匹配后，该候选按现有合同视为 no-spend closeout；否则继续走原来的 fail-closed 读回。

这个边界是可逆且比例合适的：生产代码只改一个 reducer 分支，没有新增状态、调度、权限、CLI 参数或 schema。修复属于 settlement invariant，放在既有 readback owner；没有新增机制或默认策略。当前 PR 边界：只修正 prior-Turn closeout 消费，不扩展 monitor 产品流程。monitor 的产生、到期、材料变化和 Todo 生命周期仍由原 owner 管理。

具体改动

关键代码讲解

• owedPriorTurnFromSnapshot（loopx/control_plane/quota/settlement_readback.ts:1071）在 Todo-bound 候选上调用现有 committed-monitor helper。匹配时不再查询不适用于 monitor 的 writeback/spend；不匹配时保留原 missingCloseoutReceipts 行为。
• committedMonitorPollFromSnapshot（同文件 :420）没有扩张语义；它继续要求 quota_monitor_poll classification、owner、Goal、Agent、Turn、Todo 和精确 effect 全部一致。这里的关键不是新建“monitor 已读”状态，而是让另一个既有消费者复用同一 authoritative fact。
• 新回归 a prior committed monitor poll never blocks a later unbound Turn rebind（tests/control_plane_ts/quota_settlement_readback.test.ts:979）构造“上一 monitor Turn 已提交、当前 Turn 仍未绑定”的交叉快照，并断言当前 Turn 仍为 receipt_unbound，但不再附带 owed_prior_turn。

Spec ref: docs/quota-allocation.md @ revision 770b11c9512eb7f9097329417072a4ad72d0af78。heartbeat settlement 条款位于 docs/heartbeat-automation-prompt.md，但本条 immutable review basis 固定到前述 quota contract revision。

• Q-MONITOR-CLOSEOUT implemented：精确 committed monitor poll 作为 no-spend closeout。
• Q-EXACT-EFFECT implemented：preview、缺失或错配 metadata 不能关闭 Turn。
• Q-PRIOR-RECOVERY implemented：same-Turn 与 prior-Turn 读取复用同一 exact effect 身份。

对主干的风险

最强反例是匹配过宽：只要同 Turn 存在类似 monitor 的行就错误关闭真实缺口。当前 predicate 对 owner、Goal、Agent、Turn、Todo、classification 与 effect 逐项匹配；既有 negative table 还覆盖 wrong Agent/Goal/Todo/Turn/effect、preview、空 commit 与 malformed commit，均要求 Turn 保持 open。新增用例同时证明修复没有伪造当前 Turn 的 binding。

再次 rebase 到最新 main 后重新执行：135/135 TypeScript settlement tests、22/22 Python monitor follow-through tests，以及 canary premerge 的 13/13 项检查，零失败、零 manual hold。剩余未验证项是本机私有 live runtime 的再次回放；这不影响此次只读 reducer 的合成与仓库原生门禁证据，也不应外推为模型遵循或 UI 证明。

我的整体评价

Long-horizon impact 为 improved：已关闭的 monitor 不再让后续 heartbeat 反复承担假债务；当前未绑定 Turn 仍会明确要求自己的重绑定，恢复语义没有被删除。User experience 也为 improved，因为 Agent 和维护者不再被不存在的历史欠账引向重复写回或扣额。没有 blocking finding；主要残余风险是未来有人削弱 exact-effect predicate，现有 identity 负例应继续保留。

批准该精确 head。合并仍需要独立的 exact-head readiness 与仓库 merge authority；本评论不授予合并权限。

English verdict: APPROVE at d3a21cc20ee159aaa18c3cd9c628a7f8b3f88c7e; exact committed monitor polls now close their prior Turn without suppressing real identity failures. 135 TypeScript tests, 22 monitor integration tests, and 13/13 premerge checks passed. No blocking finding; remaining risk is limited to the private live-runtime replay and future weakening of the exact-effect guard.