[loopx-agent]

Goal And Delivered Outcome

• Outcome basis / optional anchor: Add a deterministic TypeScript control-plane wave reranking planner (material_lifecycle.plan_wave_rerank), Python wrapper (plan_material_wave_rerank), CLI subcommand (loopx material-lifecycle plan-wave-rerank), Antigravity project skill surface (antigravity -> .agent/skills), and multi-host native caller provenance documentation in loopx-material.
• Goal/source and gap: material_lifecycle previously provided single-item movement (plan_material_single_move) and generic permutation proposals (build_material_rerank_proposal), leaving multi-item Top-Window shortlist updates susceptible to three structural failure modes: (1) effort-blind multi-week monographs/courses blocking short high-VOI sprint slices when scoped_slice_ready is false, (2) missing/unranked load-bearing primary citations or secondary commentaries ranked ahead of their cited primary sources, and (3) orthogonal general-trend items crowding top_window_size. In addition, project_skill_delivery lacked the antigravity .agent/skills surface root and skills/loopx-material/SKILL.md did not explicitly clarify native caller provenance across verified local owner hosts.
• Observable before → after, with the validation row that proves it:
  • Before: callers assembling multi-item rerank proposals had no deterministic control-plane wave planner to stratify unsliced deep_monograph items behind quick_slice / focused_study waves, fail closed on unranked required_load_bearing_refs or primary-over-secondary inversions, or cap orthogonal_backlog / orthogonal_signal items inside top_window_size while keeping untouched backlog items in exact original relative order. PROJECT_SKILL_SURFACE_ROOTS also omitted antigravity.
  • After: plan_material_wave_rerank / loopx material-lifecycle plan-wave-rerank deterministically computes lossless wave-ordered permutations, protected-anchor placement, Top-Window orthogonal demotion into ranked_backlog, and build_material_rerank_proposal-compatible moves + constraints via the TypeScript control-plane owner loopx/control_plane/capabilities/material_wave_ranking.ts; project-skill supports --surface antigravity (.agent/skills). Proven by tests/control_plane_ts/material_wave_ranking.test.ts, tests/capabilities/test_material_wave_ranking.py, and tests/capabilities/test_material_project_skill_delivery.py.
• Issue/task and intended base: Standalone capability & documentation slice targeting main.

Author Declaration

• Written by: model_agent (Gemini, Google)

Implemented against

• Specification and revision: no written specification; the request in this PR is the basis (built on origin/main at beff0f3e55d1f70eceebbd8e78c11f1cff34643b).
• Criteria:

Criterion (spec clause)	Disposition	Symbol / path	Test or command
Horizon-wave & effort-band cost-adjusted ordering (stratify unsliced deep_monograph to foundational_track unless scoped_slice_ready=True)	implemented	planMaterialWaveRerank in loopx/control_plane/capabilities/material_wave_ranking.ts; plan_material_wave_rerank in loopx/capabilities/material_lifecycle/ranking.py	node --no-warnings --experimental-sqlite --experimental-strip-types --test tests/control_plane_ts/material_wave_ranking.test.ts
Load-bearing primary-over-secondary lineage & unranked required_load_bearing_refs fail-closed gate + Top-Window blocker check	implemented	planMaterialWaveRerank in loopx/control_plane/capabilities/material_wave_ranking.ts	uv run --extra test python -m pytest -q tests/capabilities/test_material_wave_ranking.py
Orthogonal signal Top-Window cap (max_orthogonal_in_top_window) shifting excess orthogonal items to ranked_backlog (> top_window_size)	implemented	planMaterialWaveRerank in loopx/control_plane/capabilities/material_wave_ranking.ts	uv run --extra test python -m pytest -q tests/capabilities/test_material_wave_ranking.py
Lossless set membership, protected 1-based anchors, untouched 300-item backlog relative order retention, and build_material_rerank_proposal compatibility + CLI plan-wave-rerank	implemented	loopx/capabilities/material_lifecycle/ranking.py, loopx/capabilities/material_lifecycle/cli.py, loopx/control_plane/effect_runtime_handlers.ts	uv run --extra test python -m pytest -q tests/capabilities/test_material_wave_ranking.py tests/capabilities/test_material_single_move.py
antigravity project skill delivery surface (.agent/skills) & multi-host native caller provenance + wave quality rules documentation	implemented	PROJECT_SKILL_SURFACE_ROOTS in loopx/capabilities/project_skill_delivery/core.py; skills/loopx-material/SKILL.md; loopx/capabilities/material_lifecycle/README.md & README.zh-CN.md	uv run --extra test python -m pytest -q tests/capabilities/test_material_project_skill_delivery.py tests/test_project_skill_cli.py

• Self-check before submission: Verified TypeScript control-plane typecheck, TS unit/handler tests, Python material_lifecycle + project_skill pytest suites, semantic coinage inventory, smoke scripts, ruff, and loopx check public-boundary scan. Wave, effort, and evidence-role vocabularies live in the TypeScript decision owner; the Python boundary delegates domain validation to that owner rather than maintaining duplicate closed sets. Left out the unrelated pre-existing origin/main baseline failures in examples/control_plane/control-plane-risk-characterization-smoke.py and the separate checkpoint content-digest issue to keep this PR diff strictly scoped.

Scope And Continuation

• Completed scope and remaining work: Complete within this scope. material_lifecycle now exposes wave-based reranking across TypeScript control-plane RPC, Python API, CLI, bilingual capability READMEs, and loopx-material skill guidance, alongside .agent/skills surface support for antigravity.
• Slice boundary / successor: N/A — the requested capability owner, wrapper, CLI subcommand, surface mapping, tests, and documentation are self-contained and complete.

Re-review Repair

• [orth,a,b,c] with orth=orthogonal_backlog, a/b/c=active_sprint, and top_window_size=2 now returns [a,b,c,orth]: active waves that overflow the window stay ahead of deferred orthogonal backlog.
• [primary,secondary,independent] with top_window_size=1, secondary.primary_source_refs=[primary] now returns [independent,primary,secondary]: Top-Window replacement evaluates the complete post-swap topology and skips a candidate that depends on the source being shifted out.
• Both reviewer counterexamples have TypeScript tests and real CLI regressions. The Python wrapper no longer duplicates the TS wave/effort/evidence-role closed sets, so semantic inventory remains multi_value_twins=8/8 without raising the budget.
• Every commit in the updated range has a DCO Signed-off-by trailer.

Validation

• Tested revision: 2119bfbd64fc4def3b13da9119168f8270ef8a11
• Run state: finished
• Input classes: synthetic

Check kind	Result	Public-safe evidence / limitation
static	passed	npm run -s typecheck:control-plane, uv run --extra test python -m ruff check loopx/capabilities/material_lifecycle loopx/capabilities/project_skill_delivery tests/capabilities/test_material_wave_ranking.py tests/capabilities/test_material_project_skill_delivery.py tests/test_project_skill_cli.py, and git diff --check origin/main...HEAD all exited 0.
static	passed	python3 examples/semantic-vocabulary-drift-smoke.py passed with multi_value_twins=8/8; the TS owner remains the single domain vocabulary source.
static	passed	loopx check --scan-path ... across all 16 changed paths reported ok: True, errors=0, and public boundary scan clean.
unit	passed	node --no-warnings --experimental-sqlite --experimental-strip-types --test tests/control_plane_ts/material_wave_ranking.test.ts tests/control_plane_ts/effect_runtime_handlers.test.ts tests/control_plane_ts/effect_runtime_handler_loading.test.ts passed (42/42 tests, including both re-review counterexamples).
integration	passed	uv run --extra test python -m pytest -q tests/capabilities/test_material_*.py tests/test_decision_context_material.py tests/test_material_candidate_revision.py tests/test_project_skill_cli.py tests/capabilities/test_capability_extension_registry.py tests/control_plane/test_compact_identifier_shape_owner.py tests/architecture/test_material_lifecycle_apply_status_owner.py passed (257/257 tests, exercising Python -> TS effect_runtime_result round-trip and compatibility with build_material_rerank_proposal).
real_entrypoint	passed	Direct loopx material-lifecycle plan-wave-rerank --input - --format json subprocess checks for both reviewer counterexamples returned [a,b,c,orth] and [independent,primary,secondary] with apply_authorized=false. The CLI file/stdin and fail-closed cases also pass in tests/capabilities/test_material_wave_ranking.py.

• Coverage and gaps: Covers pure TypeScript wave planner logic, lazy RPC registration, Python validation/delegation, proposal builder round-trip, CLI --input JSON execution, and antigravity project-skill installation/removal. The unrelated pre-existing origin/main control-plane risk-characterization smoke was reproduced on the immutable base and latest main and left untouched; the separate checkpoint content-digest issue is not changed by this PR. Native Antigravity host loading and model adoption remain unverified; the installer path and managed lifecycle are covered.

See validation disclosure guidance.

Frontend / Visual Evidence

• UI impact: none
• Before: N/A
• After: N/A
• First viewport and preview: N/A
• States and viewports shown: N/A
• Source data: none
• Attention review: N/A (CLI, Python/TS capability, and Markdown documentation only; no dashboard/web/desktop UI changes).
• Craft review: N/A

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: Capability hardening (material_lifecycle deterministic wave reranking owner in TypeScript control plane + multi-host project skill delivery).

Shared-authority RFC fixture impact

• Production-scale fixture schema: N/A
• Semantic dimensions changed, or reviewed no-impact rationale: N/A (pure capability planning RPC material_lifecycle.plan_wave_rerank; does not alter Goal Authority persistence or shared-authority state schemas).
• Provider conformance arms run: N/A
• Read-only legacy/file/PostgreSQL three-arm rehearsal (required for promotion, runtime-routing, or compatibility-projection changes): N/A

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 (including .loopx/, .codex/goals/, and live ACTIVE_GOAL_STATE.md).
• I did not duplicate maintainer-owned benchmark work unless a maintainer split out a public issue for it.
• I kept the change scoped to the linked issue/task.
• Existing English/Chinese RFC pairs (including ledger entries) both have a diff, and the local premerge check python3 scripts/check_rfc_bilingual.py --base origin/main --head HEAD passes, or no paired RFC changed.
• I completed the visual evidence section for UI changes, or marked UI impact none.
• Every commit includes a DCO Signed-off-by trailer (git commit -s).

[songoow]

Reviewer: model_agent · GPT-6 Astra Pro · OpenAI · self_reported

Reviewed exact head: 9744784ce315e7eaf9ce6c762f627ca772c4377b；base：beff0f3e55d1f70eceebbd8e78c11f1cff34643b。结论基于完整 16 文件差异、现有调用链和本地真实 CLI，当前有两个需要修复的 P2。

动机 / Motivation

受影响的是已接入素材来源、需要一次调整多条学习或研究材料优先级的项目用户及其本地 Agent。

原有单条移动预览只能表达一个条目的目标名次；例如把短期交付材料、一手来源和长期阅读同时调序，需要调用方自行处理相互依赖、固定名次和整条队列，重复操作容易给出互相冲突的计划。

本次希望让调用方提交完整队列和从材料中核实的标注，一次取得可审计的多条重排预览，再复用现有提案和授权写入流程。实测入口、提案兼容性和无修改预览成立，但主动波次溢出窗口、正交来源依赖这两个组合尚未得到正确结果。

边界是显式调用的排序预览与项目级 Skill 交付；本 PR 不负责生成材料标注、自动执行重排、创建来源权限或实现新的多主机身份系统。

改动思路 / Approach

排序决策放在单一 TypeScript owner，Python 只做输入边界、调用与结果投影，复用现有提案、receipt 和项目安装器，这个边界适合本次显式预览需求；无需另建调度器、持久化排序 authority 或通用 Agent 框架。

当前 PR 应保留可直接调用的预览、已有授权边界及 Antigravity 安装映射，在同一 owner 内修复波次拼接和依赖保持的配额调整；正确性检查应验证候选结果，不能把算法自己造成的冲突归因给用户输入。

实际路径为 loopx.entrypoint.main → CLI dispatcher → _load_wave_rerank_cli_payload → plan_material_wave_rerank → lazy effect-runtime handler → planMaterialWaveRerank。新 owner 接收完整 ranked set，先规范化标注，再排波次、处理来源拓扑和固定位置，最后生成 moves。它没有 provider IO；build_material_rerank_proposal 继续要求调用方提供 inventory 与 Decision Context 引用，写入仍属于既有 source owner。新生成的约束是预览边界，不是 apply 授权。保留旧 single-move API 也合理：一次精确移动与整列按波次排序并非同一语义。

具体改动 / Changes

关键代码讲解

• planMaterialWaveRerank 是新的纯计算 owner。固定 anchor、完整成员集合、来源拓扑和 Top Window 校验都在这里，但两个调整阶段的组合存在下述反例。
• plan_material_wave_rerank 复用 effect runtime，把结果转换为既有 builder 可消费的 moves/constraints；runtime 拒绝会变为调用方可见的 ValueError，无静默降级。
• _load_wave_rerank_cli_payload 提供文件/stdin 的真实入口，成功和拒绝都保留 apply_authorized=false；失败退出码为 2。
• PROJECT_SKILL_SURFACE_ROOTS 增加 antigravity → .agent/skills，复用现有显式项目安装、managed marker 和卸载逻辑；原有 pi 等 surface 未改路径。

独立接受边界来自 docs/reference/protocols/material-lifecycle-architecture-v0.md，spec_revision=beff0f3e55d1f70eceebbd8e78c11f1cff34643b。该文档没有条款编号，以下沿用其章节名作为 criterion_id：Position（默认关闭、安装不授予来源权限）implemented；Stage-0 Contracts（有界 proposal、显式 no-change 与独立 receipt）implemented；Derived ranked state and lifecycle lineage（保留完整成员集合与 backlog）implemented；Decision-Driven Ranking and Exploration（排序提案与写入 owner 分离）implemented；Migration Boundary（本次不扩展迁移路径）out_of_scope。新增四波次具体策略没有已合入的独立规范；本次另对照它新增 README 中的公开排序承诺验证，发现 F1/F2，未把新文档当作已经验收的证据。

差异为 16 文件 +2212/-34：生产、配置及 Agent 指令 9 文件 +1510/-31，测试 4 文件 +604/-0，公开文档 3 文件 +98/-3。主体为 1092 行 TS owner；既有 receipt/schema 和权限路径没有迁移需求。

对主干的风险 / Risks

F1 — [P2] orthogonal backlog 不能插入尚未排完的主动波次之前。 initialMovable 拼接 将 orthogonalWaveItems 放在所有 remainingMovable 前。输入 [orth,a,b,c]，后三项均为 active_sprint、orth 为 orthogonal_backlog、窗口 2，真实 CLI 返回成功和 [a,b,orth,c]；既有 proposal builder 也接受其三条 moves。无 anchor、来源依赖或配额冲突，正确的波次顺序可以是 [a,b,c,orth]。这违背新增公开承诺的 max(top_window_size + 1, after_active_waves)，会把当前工作材料压到正交热点之后。最小修复是把溢出窗口的主动波次与被保序的普通 backlog 分开，在全部主动波次之后插入正交材料；补充主动波次数量大于窗口的回归用例。

F2 — [P2] 配额替换项必须在移出旧来源后仍满足依赖。 allPredsInTop 只检查交换前的排名。输入 [primary,secondary,independent]，primary 为正交 backlog，另外两项为 active sprint，secondary.primary_source_refs=[primary]，窗口 1。来源提升后配额调整选择 secondary 替换正要移出的 primary，最终校验报 primary must precede secondary，CLI 退出 2。合法结果 [independent,primary,secondary] 同时满足窗口零正交、来源在前和完整成员集合。这不是不可满足的用户约束；应排除依赖被移出来源的 replacement，并验证交换后的完整拓扑，找下一合法候选。补充这个三条目真实 CLI 用例即可区分修复。

语义与 CI 对齐

这里扩展的是 material-lifecycle 本地排序词汇，不修改通用工作 lane 或权限语义。现有公开预览承诺被上述两个边界组合破坏；修复应在同一 TS owner 中完成，并复跑 node --no-warnings --experimental-sqlite --experimental-strip-types --test tests/control_plane_ts/material_wave_ranking.test.ts 及 python -m pytest -q tests/capabilities/test_material_wave_ranking.py，增加对应反例。不能只降低 README 的承诺来把错误顺序或错误拒绝算作成功。

验证：PASS TS owner/handler/loading 40/40、Python wave/single-move/contracts/project-scope/project-skill/CLI 70/70、全 control-plane typecheck、git diff --check。FAIL 两个真实 CLI 的预期排序/可行性条件，现有测试全绿仍未覆盖。PASS base/head 的 material-lifecycle architecture --format json 完全相等，无标注预览保持原队列，输出明确不授权 apply；项目安装正反路径覆盖 Antigravity。native host 实际加载和模型遵循指令未验证，未将路径测试冒称为主机集成通过。GitHub 队列快照无 status-check rollup，远端 CI 记 UNVERIFIED，与本地已复现缺陷分开判断；两次因测试文件名选错而未收集测试的探索命令不计入通过数。

我的整体评价 / Verdict

架构归属、默认关闭和已有写入权限的分工可以保留；本次没有创建与 Core 并行的权限来源。long_horizon 判断为 preserved：新增预览无持久化队列、重试调度或外部写入，旧路径和成员集合守卫仍在。user_experience 判断为 regression：显式使用新排序入口时会拿到错误 backlog 顺序，或在存在合法解时被要求处理并不存在的输入冲突，尚不能认定目标已交付。视觉审查不适用：本次只触及 CLI、纯计算和项目 Skill，没有渲染页面变化；未验证范围为真实 Antigravity host 加载及模型采纳。

不因代码规模要求重建框架，也不把等待全部远端 CI 作为这次判断的代替品。最小重审条件是修复 F1/F2，在上述实际入口证明主动波次溢出和依赖来源配额组合，同时保持当前成功/拒绝、no-change、固定 anchor、授权隔离测试。此结论仅覆盖本次固定 head；发布前仍需核对远端 head。

English verdict: REQUEST_CHANGES — two reproducible ranking defects remain at the reviewed head.

[huangruiteng]

Reviewer: model_agent · GPT-5 · OpenAI · self_reported

Reviewed exact head: 2119bfbd64fc4def3b13da9119168f8270ef8a11；base：bcf1b6363f70487ab21dfaec34c5eda14b9db226。上一轮两个 P2 已在当前 head 修复。

动机 / Motivation

已启用 Material Lifecycle 的项目用户及其本地 Agent，在需要同时调整多条学习或研究材料优先级时使用该预览。

此前只有单条移动和不带波次语义的 proposal builder；调用方需要自行协调一手来源先于二手解读、固定名次、Top Window 正交热点配额以及大型 backlog 的相对顺序，容易生成相互冲突或压低当前工作的计划。具体到本轮反例，[orth,a,b,c] 曾返回 [a,b,orth,c]，而合法的 secondary.primary_source_refs=[primary] 场景又会错误拒绝存在合法解的计划。

调用一次 CLI 或 Python API 即可得到完整、可审计且保持成员集合的重排预览，并可直接交给既有 proposal builder；写入仍需原有 owner gate。

本 PR 不生成材料标注、不读取私有正文、不新增 source authority、不自动 apply，也不替代 Antigravity 等宿主自身的 turn 与权限系统。

改动思路 / Approach

排序规则与不变量归 TypeScript owner，Python 只做边界校验、RPC 调用和结果投影，CLI 保持只读预览，这与现有 material-lifecycle 的 decision/effect 分离一致。

实际路径是 loopx.entrypoint.main → _load_wave_rerank_cli_payload → plan_material_wave_rerank → lazy effect-runtime handler → planMaterialWaveRerank。TypeScript owner 接收完整 ranked set 和公开安全的标注，先建立 primary_source_refs 拓扑，再按波次、投入成本、证据角色和原始稳定名次排序，随后处理 protected anchors 与 Top Window 正交配额，最后逐项验证成员集合、拓扑和 fixed-rank 后置条件。输出只包含 opaque refs、moves 和 constraints，不携带正文、URL、凭据或 provider state。

flowchart LR
    INPUT["完整 ranked refs + 公开安全标注"]
    TS["TypeScript owner<br/>拓扑 / 波次 / anchor / Top Window"]
    CHECK["后置条件<br/>lossless membership / precedence / cap"]
    OUT["preview<br/>reordered refs + moves + bounds"]
    PROP["既有 proposal builder<br/>仍需 inventory / Decision Context / owner gate"]
    WRITE["source owner apply<br/>本 PR 不执行"]

    INPUT --> TS --> CHECK --> OUT --> PROP -. "外置授权边界" .-> WRITE

Loading

上图是当前 head 的源码示意，不代替执行证据。正的公开调用路径走真实 CLI 与 Python wrapper；失败路径在 TypeScript owner 内以 typed request error 拒绝，再由 Python 转为 ValueError，不会静默降级或部分写入。

本 PR 交付显式调用的 wave-rerank 预览、项目级 Antigravity Skill 映射和相关文档；不新增 source authority、持久化队列、自动 apply 或通用 Agent 框架。

复用判断明确：plan_material_single_move 只表达一次精确移动，build_material_rerank_proposal 负责既有 proposal 格式和授权边界，二者都不拥有多条目波次排序与来源拓扑求解；新增 TS owner 因此是本问题的最小决策边界，而不是平行 authority。Python 侧最初重复的 wave/effort/evidence-role 闭集已删除，域词汇现在只由 TS owner 定义。

具体改动 / Changes

完整 diff 为 16 文件，+2402/-34。主体是 1126 行 TypeScript owner、413 行真实 CLI/Python 回归、366 行 TS 回归，以及双语 README、Skill 指引和现有 installer 映射的小幅扩展；既有 receipt/schema、Core authority 和 source apply 路径没有迁移。

关键代码讲解

• planMaterialWaveRerank 是纯决策 owner。它校验紧凑 refs、annotation 字段和数量、来源 DAG、anchor 可达区间及 load-bearing 依赖，并返回不可变的顺序、moves、bounds 和审计列表。
• initialMovable 与 Stage B 把溢出窗口的主动波次置于 deferred orthogonal backlog 之前；配额替换现在先构造完整交换后顺序，再用 preservesPrimaryPrecedence 检查整个拓扑并继续寻找合法候选。
• plan_material_wave_rerank 只做 Python 边界规范化、effect-runtime 调用和 typed result 投影；它不复制 TS 的领域闭集，invalid wave/effort/role 仍由 TS owner 拒绝。
• _load_wave_rerank_cli_payload 支持文件与 stdin，严格限制顶层字段，成功和失败都显式保留 apply_authorized=false。
• PROJECT_SKILL_SURFACE_ROOTS 增加 antigravity → .agent/skills，复用现有 managed marker、预览、安装和卸载事务；effect_runtime_handlers 只注册 lazy handler，不扩大默认加载面。

契约映射

审查基准为 docs/reference/protocols/material-lifecycle-architecture-v0.md，spec_revision=bcf1b6363f70487ab21dfaec34c5eda14b9db226。

criterion_id	requirement	disposition
POSITION	capability 默认关闭、owner-scoped，不拥有原文或 Core authority	implemented：catalog 与 architecture packet 仍报告 default_enabled=false、creates_authority=false、mutates_core_state=false
STAGE0	输出 bounded rerank delta，并支持明确 no-change	implemented：planMaterialWaveRerank 生成 moves/constraints，空移动保持既有 proposal 可消费
DERIVED-STATE	保留完整 ranked set 与 backlog 成员关系	implemented：lossless membership、protected rank 与 unannotated backlog 后置检查及 TS/CLI 回归
RANKING	排序决策与写入分离，写入依赖 owner gate	implemented：新路径只返回 apply_authorized=false 的 preview，既有 proposal/receipt 继续独占写入边界
MIGRATION	本次不扩展 migration provider 或 cutover	out_of_scope：diff 未修改 migration plan/apply/rollback 路径

文档与宿主面

双语 README 和 loopx-material Skill 指引增加了三组排序规则、真实 CLI 示例、引用顺序和 Top Window 配额说明；Antigravity、Kiro CLI、Pi 的已注册 surface 也在文档表中对齐。项目级安装仍只影响显式目标目录，不激活材料写入。原生 Antigravity 实际加载 Skill 与模型遵循仍未验证，不能由路径测试推出。

对主干的风险 / Risks

当前 head 没有 blocking finding。最强回归风险仍是“约束求解在后续阶段改变前一步判断”：F1 会让正交 backlog 插到尚未排完的 active wave 前，F2 会让配额替换破坏来源顺序。两项都已在完整 owner 上复现并修复；当前实现通过真实 CLI 分别得到 [a,b,c,orth] 和 [independent,primary,secondary]，同时保留 anchor、lineage、membership 和 authorization 检查。

其他风险按证据处理如下：

• 性能与规模：正式 TS 用例覆盖 300 条队列，额外随机检查覆盖 4000 个 1–40 条队列的成员集合与 Top Window 配额；尚未证明 10,000 条上限下的交互延迟，但实现为纯计算预览，不影响普通调用路径。
• 默认关闭隔离：base 与 head 的 loopx material-lifecycle architecture --format json 输出字节一致；新增 CLI 只有显式调用时出现，没有加入 heartbeat、scheduler、自动 apply 或默认 Skill 安装。
• 权限语义：material_lifecycle.plan_wave_rerank、material_wave_rerank_plan_v0 和 CLI planned 都只表示预览，不承诺写入；现有 proposal builder 继续要求 revision-bound inventory、Decision Context 和 owner gate。
• 宿主集成：antigravity 安装目标与 managed lifecycle 有正反路径测试，但真实宿主发现与模型采纳未执行，保留为 unverified，不影响本次排序 owner 的正确性判断。
• 既有基线红灯：control-plane-risk-characterization-smoke.py 在 immutable main 与当前 head 都因同一 safe_bypass_allowed=False 断言失败；本 PR 不修改该因果路径，且变更不变量有独立通过的测试。GitHub CI 在评审时仍有 queued job，当前记作诊断项，不冒充成功。

语义与 CI 对齐

新增词表只属于 material-lifecycle 本地排序契约：MATERIAL_HORIZON_WAVES、MATERIAL_EFFORT_BANDS、MATERIAL_EVIDENCE_ROLES 和 effective wave 由 TS owner 单点定义。Python 仅验证 opaque token，实际域校验通过真实 runtime 返回；semantic-vocabulary-drift-smoke.py 从初始的 multi_value_twins=11/8 修复为 8/8，没有提高预算或注册无关词汇。

验证结果：PASS TypeScript owner/handler/loading 42/42；PASS Python 关联套件 257/257；PASS control-plane typecheck、Ruff、Rfc bilingual check、git diff --check、semantic vocabulary smoke；PASS 两个真实 CLI 反例。canary premerge 的 diff、编译、risk-profile 与 public-boundary 部分通过，唯一失败已按上面同一 base/head 签名归因为既有风险 smoke。

我的整体评价 / Verdict

long_horizon 判断为 preserved：新增能力是显式预览，不创建持久队列、自动重试、静默副作用或第二套写入 authority，完整成员集合与既有 proposal/receipt 链路仍受保护。user_experience 判断为 improved：用户和本地 Agent 不再需要手工拼接多条相互约束的 moves，CLI 与 Python API 返回同一可审计结果，失败信息指出真正缺失的 intake、anchor 或 source dependency；本轮两个错误顺序/错误拒绝已消除。

代码体积较大，但主要来自可执行的纯 TS 约束 owner 和对应测试；复用单条移动或通用 proposal builder 无法表达整列波次与拓扑求解，因此没有发现可删除的平行状态机。最有价值的简化已经执行：移除 Python 的一份领域词汇副本，恢复单一 TS 词表 owner。spec 映射的五个 criterion 均 implemented 或按既有边界 out_of_scope，没有 not_met。

视觉审查不适用：本 PR 没有页面或渲染布局变化，只触及 CLI、Python/TypeScript capability、installer 与 Markdown；最需要补足但仍不阻塞本次合并的验证是真实 Antigravity 宿主加载与模型采纳。排除该未验证宿主维度后，当前 exact head 的功能不变量、默认关闭、授权分离、旧 API 兼容和 reviewer 反例都有独立证据。

English verdict: APPROVE — both reported ranking defects are fixed and validated at exact head 2119bfbd64fc4def3b13da9119168f8270ef8a11; no blocking finding remains.

Both requested changes are resolved at exact head 2119bfb. F1 now places all overflowing active waves before deferred orthogonal backlog and the real CLI returns [a,b,c,orth]. F2 now validates the complete post-swap primary_source_refs topology, skips dependent replacements, and the real CLI returns [independent,primary,secondary]. TypeScript and Python CLI regressions pass, all review threads are resolved, and the current exact head has a fresh approval.