From b87caf2602b61772144ad251591258275433b96c Mon Sep 17 00:00:00 2001 From: Lihua <1017343802@qq.com> Date: Mon, 12 Oct 2026 13:22:45 -0700 Subject: [PATCH 1/2] docs(research): record merged M3 checkpoint and open gates Signed-off-by: Lihua <1017343802@qq.com> --- .../research-exploration-control-plane-v0.md | 22 +++++++++---------- 1 file changed, 10 insertions(+), 12 deletions(-) diff --git a/docs/architecture/rfcs/research-exploration-control-plane-v0.md b/docs/architecture/rfcs/research-exploration-control-plane-v0.md index df3d50a7e0..cbf090f681 100644 --- a/docs/architecture/rfcs/research-exploration-control-plane-v0.md +++ b/docs/architecture/rfcs/research-exploration-control-plane-v0.md @@ -143,6 +143,7 @@ milestone status. | Host-delivered context | Quota can project a compact coverage ledger and uncovered frontier without requiring a manual evidence-read ritual. | | Explore evidence | Explore owns append-only nodes, edges, findings, and bounded public-safe projections. | | Research evidence and cold shadow | [Research observation v0](../../reference/protocols/research-observation-v0.md) composes generic progress with typed closure and explicit binary candidates. CLI summary and Lark node summaries share the read-only projection; exact experiment input fingerprints guard stale results. | +| Explicit M3 composition gate | [PR #5280](https://github.com/loopx-project/loopx/pull/5280) merged as `1e6cae19f` on 2026-10-12. Opt-in Goal-frontier, Todo, experiment and input lineage share a typed write-time qualification across File and SQLite; stale or unrelated evidence cannot discharge the current obligation. The existing Goal capability editor and CLI own activation. This is an integrated rule, not evidence of accepted scientific results or natural-host use. | | Explore planning | Optional branch planners are read-only and return execution to quota, Todo, claim, and lease. | | Evidence in turn context | Planning-mode turn-start reads retain bounded explicit Todo/node evidence diagnostics and existing exploring frontier nodes. Linked evidence is classified before detail truncation: all findings and relevant edges in the supplied projection contribute to diagnostic counts and hazards, while row details remain bounded. Turn context reports omitted detail counts from those totals. This delivers decision input, not the §11.5 adoption gate or proof of model behavior. | | Model behavior qualification | Real function-tool conversations can test whether a model reads an actual packet and selects a real semantic writeback. | @@ -151,10 +152,8 @@ milestone status. | Gap | Consequence | |---|---| -| Mainline M3 write-time gate and exact lineage are not integrated | Open PR #5280 proposes an exact obligation/Todo/experiment/input gate. Its local receipt tests do not establish accepted Goal closure, and current-head CI, mergeability and maintainer effect acceptance remain open. | -| Cold shadow not adopted by hot status/frontier on main | The existing #3173 projection remains behavior-compatible; the proposed M3 consumer adoption still needs integrated entrypoint qualification. | -| Dismissal and deferral are not integrated | PR #5280 locally implements typed candidate dismissal, blocker waits and exact invalidated-duty retirement; review and mainline readback remain separate gates. | -| No experiment-result adoption gate | Negative evidence can be recorded without proving that the next bound decision used its scope, validity or uncertainty. Proposed single-experiment integration is defined in §11.5; current cold receipts and PR #5280 do not enforce it. | +| M3 integrated behavior lacks complete remote qualification | PR #5280 has an exact-head approved review and local File/SQLite, default-off and packaged-configuration evidence. Its final-head CI did not finish green: jobs were cancelled on merge and the Stage2c aggregate failed because its matrix result was `cancelled`. This is not a demonstrated Stage2c test failure or remote browser acceptance. The separate browser CI source-install and composer-focus PRs #6287/#6282 closed without merging; their unresolved checks remain independent of M3's local evidence. | +| No experiment-result adoption gate | Negative evidence can be recorded without proving that the next bound decision used its scope, validity or uncertainty. Proposed single-experiment integration is defined in §11.5; merged PR #5280 does not enforce it. | | Execution attribution is not effect authority | An exact Todo/experiment/input receipt does not grant a task lease, effect permission or accepted Goal closure. | | Live qualification incomplete | Deterministic and real CLI/file-log tests establish state semantics, not model selection quality or scientific truth; no live Lark sync is qualified by projection tests. | | No promotion evidence for inferred combinations | Shared constraints are not known to be precise enough to trigger obligations. | @@ -954,8 +953,8 @@ control-plane failures. |---|---|---|---| | M0 | RFC, current-state inventory, and explicit ownership decision | Maintainer review; no runtime behavior | Accepted design | | M1 | Characterization fixtures plus typed research observation and closure contract in Explore | Deterministic normalization, privacy, compatibility, and negative tests | Implemented evidence/CLI slice; live research qualification remains separate | -| M2 | Explicit-only composition candidate, canonical gap projection, and read-only status shadow | No pairwise inference; bounded packet; projection parity | Partial: #3173 legacy quota/successor; canonical binary cold shadow in CLI/Lark projection; hot status adoption and live Lark qualification remain | -| M3 | Goal-frontier obligation, exact Todo/experiment lineage, shared write-time gate; proposed §11.5 single-experiment result/adoption prerequisite | State/replay matrix, real provider/state readback, default-off parity and packaged journey before enforcement | Proposed in open PR #5280, not integrated: local state/replay and premerge checks exist, while current-head CI, mainline conflict, public effect case and maintainer acceptance remain open. §11.5 is a design proposal, not delivered runtime | +| M2 | Explicit-only composition candidate, canonical gap projection, and read-only status shadow | No pairwise inference; bounded packet; projection parity | #3173 legacy quota/successor and canonical binary cold shadow in CLI/Lark projection; hot status/frontier adoption shipped with M3, while live Lark qualification remains | +| M3 | Goal-frontier obligation, exact Todo/experiment lineage, shared write-time gate; proposed §11.5 single-experiment result/adoption prerequisite | State/replay matrix, real provider/state readback, default-off parity and packaged journey before enforcement | Explicit-only gate merged in #5280 at `1e6cae19f` with exact-head approved review. Integrated-source local checks passed: 67 focused Python, 37 focused TypeScript, control-plane typecheck, packaged Explore configuration journey and 19/19 risk canaries. The final-head CI was cancelled/failed as described in §5.2, so remote and installed-host acceptance remain unproven. §11.5 result adoption is still a proposal, not delivered runtime. | | M4 | Bounded multi-candidate cards, `composition_selection_v0`, real model-tool behavior qualification, and repeated live shadow | Model autonomously selects a legal semantic action from the delivered candidate set; selection quality is no worse than the declared fallback; compact receipts only | Not started | | M5 | Shared-constraint candidate ranking in shadow mode | Precision and cost evidence; no automatic trigger | Not started | | M6 | Optional inferred trigger | Explicit maintainer decision and measured promotion thresholds | Deferred | @@ -971,8 +970,8 @@ The M1/M2 evidence slice delivers: - characterize the current Explore projection before changing it; - prove public safety, canonical identity, and bounded projection; - show candidates and gaps in Explore summary and existing Lark node projections; - preserve #3173 quota behavior. Hot status adoption remains part of M3's common - frontier integration, with independent qualification before any promotion. + preserve #3173 quota behavior. The merged M3 gate now shares the live frontier + with status; remote and natural-host readback remain separate qualification. Validation uses `tests/control_plane_ts/explore_research.test.ts`, `tests/capabilities/test_explore_research_evidence.py` and the existing @@ -981,14 +980,13 @@ terminal coverage, attribution, replay, invalidated inputs, stale experiment lineage and real CLI/file-log readback. Projection tests do not establish live remote effects or model-selected research behavior. Delivery is tracked in [#5214](https://github.com/loopx-project/loopx/issues/5214), now closed. Its closure -does not supply missing live/M3 qualification. Remaining research direction is +does not supply missing live M3 qualification. Remaining research direction is tracked by [#4391](https://github.com/loopx-project/loopx/issues/4391) and [#3243](https://github.com/loopx-project/loopx/issues/3243); preserve the explicit milestone gaps here rather than create a parallel task tree. -M3 is the first behavior-changing slice. It should be a separate PR so the -obligation and write gate can be reviewed and reverted independently from the -evidence schema. +M3 was the first behavior-changing slice. PR #5280 kept the obligation and +write gate separately reviewable and reversible from the evidence schema. The M3 development boundary joins current same-agent Todo, experiment and input facts in the typed Explore owner. Goal policy is explicitly scoped and disabled From 9fb297bdce561135084738fca9df87773ce9493b Mon Sep 17 00:00:00 2001 From: Lihua <1017343802@qq.com> Date: Mon, 12 Oct 2026 18:43:55 -0700 Subject: [PATCH 2/2] docs(research): mirror merged M3 checkpoint in Chinese Signed-off-by: Lihua <1017343802@qq.com> --- ...arch-exploration-control-plane-v0.zh-CN.md | 21 +++++++++---------- 1 file changed, 10 insertions(+), 11 deletions(-) diff --git a/docs/architecture/rfcs/research-exploration-control-plane-v0.zh-CN.md b/docs/architecture/rfcs/research-exploration-control-plane-v0.zh-CN.md index e6b5536767..78707dcbc4 100644 --- a/docs/architecture/rfcs/research-exploration-control-plane-v0.zh-CN.md +++ b/docs/architecture/rfcs/research-exploration-control-plane-v0.zh-CN.md @@ -124,18 +124,17 @@ hypothesis”或“新 probe family”。它无法持久表达：A 和 B 都已 | Host-delivered context | Quota 可以投影 compact coverage ledger 与 uncovered frontier，不要求手动 evidence-read 仪式。 | | Explore evidence | Explore 拥有 append-only node、edge、finding 与有界 public-safe projection。 | | Research evidence 与 cold shadow | [Research observation v0](../../reference/protocols/research-observation-v0.zh-CN.md) 组合 generic progress、typed closure 和 explicit binary candidate。CLI summary 与 Lark node summary 共用只读投影，精确 experiment input fingerprint 防止过期结果继续有效。 | +| 显式 M3 组合门禁 | [PR #5280](https://github.com/loopx-project/loopx/pull/5280) 已于 2026-10-12 作为 `1e6cae19f` 合入。可选启用的 Goal-frontier、Todo、experiment 和 input lineage 在 File/SQLite 中共用 typed write-time qualification；陈旧或无关证据不能解除当前义务。仍由既有 Goal 能力编辑器和 CLI 负责启用。规则已集成，不等于科学结果获接受或自然宿主已验收。 | | Explore planning | 可选 branch planner 保持只读，并把执行交还 quota、Todo、claim 和 lease。 | -| Turn context 中的证据 | Planning 模式的 turn-start 读取保留有界的显式 Todo/节点证据诊断与既有 exploring 前沿节点；先解析 finding 再裁剪近期历史。这交付决策输入，不代表 §11.5 adoption gate 或模型行为已经验证。 | +| Turn context 中的证据 | Planning 模式的 turn-start 读取保留有界的显式 Todo/节点证据诊断与既有 exploring 前沿节点。关联证据先分类再裁剪详情：输入投影中的全部 finding 和相关边参与诊断计数与风险判断，逐行详情仍保持有界。Turn context 同时报告被省略的详情数。这交付决策输入，不代表 §11.5 adoption gate 或模型行为已经验证。 | | Model behavior qualification | 真实 function-tool 对话可以验证模型是否读取真实 packet 并选择真实 semantic writeback。 | ### 5.2 还缺什么 | 缺口 | 后果 | |---|---| -| Mainline 尚未集成 M3 写入门禁与精确 lineage | 开放的 PR #5280 提议精确 obligation/Todo/experiment/input 门禁；本地 receipt 测试不证明 Goal 已接受关闭，当前 head 的 CI、可合并性与维护者效果验收仍未完成。 | -| Cold shadow 尚未在 main 接入 hot status/frontier | 既有 #3173 投影保持行为兼容；提议中的 M3 消费端仍需集成后的入口资格验证。 | -| Dismissal 与 deferral 尚未集成 | PR #5280 在本地实现 typed candidate dismissal、blocker wait 与精确 invalidated-duty retirement；review 和主干回读仍是独立门槛。 | -| 尚无 experiment-result adoption gate | 负证据可被记录，但不能证明下一绑定决策使用了其 scope、validity 与 uncertainty。§11.5 提议 single-experiment 集成；当前 cold receipt 与 PR #5280 均未强制执行。 | +| 已集成的 M3 行为尚缺完整远端资格验证 | PR #5280 有精确 head 的批准评审，以及本地 File/SQLite、默认关闭和打包配置旅程证据。最终 head 的 CI 未全绿：合入时多项任务被取消，Stage2c 汇总因矩阵结果为 `cancelled` 而失败。这不证明 Stage2c 测试本身失败，也不构成远端浏览器验收。独立的浏览器 CI 源码安装和输入框焦点 PR #6287/#6282 均关闭且未合入；其未完成的检查与 M3 本地证据分开。 | +| 尚无 experiment-result adoption gate | 负证据可被记录，但不能证明下一绑定决策使用了其 scope、validity 与 uncertainty。§11.5 提议 single-experiment 集成；已合入的 PR #5280 仍未强制执行结果采用。 | | 执行 attribution 不是 effect 权限 | 精确 Todo/experiment/input receipt 不授予 task lease、effect 权限或已接受的 Goal closure。 | | Live qualification 不完整 | Deterministic 与真实 CLI/file-log 测试证明状态语义，不证明 model selection 质量或科学结论；projection 测试不构成 live Lark sync 资格。 | | inferred combination 没有 promotion evidence | 共享 constraint 的精度还不足以直接触发 obligation。 | @@ -865,8 +864,8 @@ rule，以及 model variance 与 control-plane failure 的分离。 |---|---|---|---| | M0 | RFC、current-state inventory 与显式 ownership decision | Maintainer review；无 runtime behavior | 已接受的设计 | | M1 | Characterization fixture，以及 Explore 中的 typed research observation 与 closure contract | Deterministic normalization、privacy、compatibility 与 negative test | Evidence/CLI 切片已实现；真实研究 qualification 独立保留 | -| M2 | Explicit-only composition candidate、canonical gap projection 与 read-only status shadow | 不做 pairwise inference；packet 有界；projection parity | 部分实现：#3173 legacy quota/successor；CLI/Lark projection 的 canonical binary cold shadow；hot status adoption 与 live Lark qualification 仍未完成 | -| M3 | Goal-frontier obligation、精确 Todo/experiment lineage、共享 write-time gate；§11.5 提议 single-experiment result/adoption prerequisite | State/replay matrix、真实 provider/state readback、default-off parity；enforcement 前验证 packaged journey | 开放 PR #5280 提案，尚未集成：已有本地 state/replay 与 premerge 检查，但当前 head 的 CI、主干冲突、公开效果案例与维护者验收仍未闭合。§11.5 是设计提案，不是已交付 runtime | +| M2 | Explicit-only composition candidate、canonical gap projection 与 read-only status shadow | 不做 pairwise inference；packet 有界；projection parity | #3173 legacy quota/successor 与 CLI/Lark projection 的 canonical binary cold shadow 已交付；hot status/frontier adoption 随 M3 合入，live Lark qualification 仍未完成 | +| M3 | Goal-frontier obligation、精确 Todo/experiment lineage、共享 write-time gate；§11.5 提议 single-experiment result/adoption prerequisite | State/replay matrix、真实 provider/state readback、default-off parity；enforcement 前验证 packaged journey | 显式门禁已在 #5280 作为 `1e6cae19f` 合入，并有精确 head 的批准评审。集成源码的本地验证通过：67 项聚焦 Python、37 项聚焦 TypeScript、control-plane typecheck、打包 Explore 配置旅程及 19/19 风险 canary。最终 head 的 CI 如 §5.2 所述被取消/汇总失败，远端和已安装宿主验收仍未证明。§11.5 结果采用仍是设计提案，不是已交付 runtime。 | | M4 | 有界 multi-candidate card、`composition_selection_v0`、真实 model-tool behavior qualification 与重复 live shadow | 模型从交付 candidate set 中自主选择合法 semantic action；选择质量不劣于 declared fallback；只保留 compact receipt | 未开始 | | M5 | Shared-constraint candidate 在 shadow mode 中排序 | 有 precision/cost evidence；不自动触发 | 未开始 | | M6 | 可选 inferred trigger | 显式 maintainer decision 与量化 promotion threshold | 延后 | @@ -882,8 +881,8 @@ M1/M2 evidence 切片交付： - 修改 Explore projection 前先 characterization； - 证明 public safety、canonical identity 与 bounded projection； - 在 Explore summary 与既有 Lark node projection 展示 candidate/gap，保持 #3173 - quota 行为。Hot status adoption 留在 M3 common frontier 集成，任何 promotion - 之前仍需独立 qualification。 + quota 行为。已合入的 M3 门禁与 status 共用 live frontier；远端和自然宿主 + 读回仍需独立 qualification。 验证入口为 `tests/control_plane_ts/explore_research.test.ts`、 `tests/capabilities/test_explore_research_evidence.py` 与既有 composition-frontier/ @@ -891,12 +890,12 @@ result-layer 检查。覆盖反向配对 canonical identity、terminal coverage replay、input invalidation、stale experiment lineage 和真实 CLI/file-log 读回。 Projection 测试不证明 live remote effect 或模型自主研究行为。交付继续由 [#5214](https://github.com/loopx-project/loopx/issues/5214) 记录，该 issue 已关闭。 -关闭不补齐缺失的 live/M3 qualification。后续研究方向由 +关闭不补齐缺失的 live M3 qualification。后续研究方向由 [#4391](https://github.com/loopx-project/loopx/issues/4391) 与 [#3243](https://github.com/loopx-project/loopx/issues/3243) 跟踪；本 RFC 保留明确 milestone gap，不另建平行 task tree。 -M3 是第一个 behavior-changing slice。它应单独成 PR，使 obligation 与 write gate +M3 是第一个 behavior-changing slice。PR #5280 使 obligation 与 write gate 能够独立于 evidence schema 评审和回滚。 M3 开发边界在 typed Explore owner 中连接当前同一 Agent 的 Todo、experiment