feat(progress-review): close the drift loop with typed receipts and a recorded differential

README.md

@@ -441,6 +441,7 @@ write boundary:
 | --- | --- | --- |
 | Turn a public issue into a reviewable, evidence-backed change | [Issue Fix](loopx/capabilities/issue_fix/README.md) | `loopx capability show issue-fix --format json` |
 | Qualify the exact final diff before delivery | [Change Quality](loopx/capabilities/change_quality/README.md) | `loopx capability show change-quality-qualification --format json` |
+| Notice busy-but-off-goal work rounds before the periodic review | [Progress-Review Sentinel](loopx/capabilities/progress_review/README.md) | `loopx capability show progress-review-sentinel --format json` |
 | Preserve a changing stack of already reviewed branches | [Integration Branch](loopx/capabilities/integration_branch/README.md) | `loopx capability show integration-branch-reconcile --format json` |
 | Explore uncertain research without losing hypotheses and findings | [Explore](loopx/capabilities/explore/README.md) | `loopx capability show explore --format json` |
 | Rebase decisions on current evidence and verified outcomes | [Decision Context](loopx/capabilities/decision_context/README.md) | `loopx capability show decision-context --format json` |

README.zh-CN.md

@@ -393,6 +393,7 @@ Capability 把上述通用原语组成 outcome-owned 工作泳道。先按结果
 | --- | --- | --- |
 | 把公开 issue 推进为可审查、有证据的变更 | [Issue Fix](loopx/capabilities/issue_fix/README.zh-CN.md) | `loopx capability show issue-fix --format json` |
 | 在交付前对精确 final diff 做质量验收 | [Change Quality](loopx/capabilities/change_quality/README.md) | `loopx capability show change-quality-qualification --format json` |
+| 在周期复审之前发现“忙碌但偏离目标”的工作轮次 | [进展评估哨兵](loopx/capabilities/progress_review/README.zh-CN.md) | `loopx capability show progress-review-sentinel --format json` |
 | 维护由多个已审查分支组成、持续变化的集成栈 | [Integration Branch](loopx/capabilities/integration_branch/README.md) | `loopx capability show integration-branch-reconcile --format json` |
 | 在不丢失假设和发现的前提下探索不确定研究问题 | [Explore](loopx/capabilities/explore/README.zh-CN.md) | `loopx capability show explore --format json` |
 | 基于当前证据和已验证结果重新建立决策上下文 | [Decision Context](loopx/capabilities/decision_context/README.zh-CN.md) | `loopx capability show decision-context --format json` |

apps/presentation/dashboard/src/features/personal-workspace/capability-localization.ts

@@ -25,6 +25,10 @@ const capabilityCopy: Record<WorkspaceLocale, Record<string, LocalizedCopy>> = {
       displayName: "Change quality qualification",
       description: "Prepares a provider-neutral review packet, allows at most one policy-authorized safe-fix pass, and can require an exact-diff receipt.",
     },
+    progress_review: {
+      displayName: "Progress-review sentinel",
+      description: "Records typed drift receipts from an external bounded review of scoped file deltas; assist may raise the existing autonomous replan obligation.",
+    },
     explore_graph: {
       displayName: "Explore Graph",
       description: "Organizes bounded exploration as a typed evidence graph so branches, findings, and synthesis remain inspectable.",
@@ -85,6 +89,10 @@ const capabilityCopy: Record<WorkspaceLocale, Record<string, LocalizedCopy>> = {
       displayName: "变更质量验证",
       description: "生成与 Provider 无关的审阅包，最多允许一次策略授权的安全修复，并可要求精确 diff 回执。",
     },
+    progress_review: {
+      displayName: "进展评估哨兵",
+      description: "记录外部有界评估对限定文件变化给出的类型化漂移回执；assist 模式可触发已有的自主重规划义务。",
+    },
     explore_graph: {
       displayName: "探索图谱",
       description: "把有界探索组织为 typed 证据图，让探索分支、发现与综合结论都可检查、可追溯。",

docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md

@@ -1,7 +1,7 @@
 # RFC: Agent Judgment and Optional Independent Assessment — Jev as a Candidate (v0)
 
 - **RFC status:** Draft; M0 **accepted-for-discussion** ([maintainer decision](https://github.com/loopx-project/loopx/pull/4749#pullrequestreview-5259253204)). Q1–Q7 remain pending; the research/design is not accepted for implementation.
-- **Delivery maturity:** Proposal; documentation only, no integration or model qualification.
+- **Delivery maturity:** Research proposal; a separate D1-only optional shadow implementation is proposed in Appendix A. No model qualification or automatic correction is established. The default-off sentinel capability and its recorded differential live in [`loopx/capabilities/progress_review`](../../../loopx/capabilities/progress_review/README.md); those numbers do not change this Draft's status.
 - **Created:** 2026-09-19. **Last normative revision:** 2026-09-20.
 - **Implementation baseline:** `9f1916960306b3650d795895b89f331eeae2516e`; source ownership and trigger behavior rechecked at PR revision `27812bd0fb437f831a541b564bcb5be8a96ff77e`. Historical upstream inspection is recorded in Appendix A, not a whole-system certification.
 - **Authors / owners:** Proposal author; existing domain maintainers own any direction selected. No new runtime authority or assigned implementation owner.
@@ -355,8 +355,19 @@ No default D1 implementation, automatic worker adoption or hidden mandatory-mode
 - **Delta:** D7/D8, provisional opportunity ordering, expected-value decomposition and bounded ranking comparisons; bilingual text and index updated.
 - **Evidence/remaining gap:** source inspection and documentation only. No live provider comparison, production ranker or new authority; Q1–Q7 remain pending.
 
+### 2026-09-21 — Task-progress shadow implementation proposal (RFC D1)
+
+- **Baseline:** upstream `62d18677c`; the implementation includes only the optional D1 command, without D2–D8 ranking or selector changes.
+- **Proposal:** scoped checkpoint capture around actual refresh-state, separate inference, off/shadow configuration and historical readback. See the [operation guide](../../../packages/loopx-jev/DRIFT_SHADOW.md).
+- **Evidence boundary:** [discussion evolution and current-implementation measurements](../../../packages/loopx-jev/DESIGN_DECISIONS.md); deterministic integration checks are distinct from provider accuracy and task benefit. No automatic intervention or native host-hook rollout.
+- **Pending:** maintainer acceptance of this optional-tool scope and independently evaluated comparative value. The earlier M0 intake is not retroactive implementation approval.
+
 ## Appendix B: Decision log
 
+The separately proposed [Task-progress observation tool and decision record](../../../packages/loopx-jev/DESIGN_DECISIONS.md)
+do not change the historical M0 decision or settle Q1–Q7. Maintainers review that
+optional-tool scope separately; experimental results do not establish product adoption.
+
 | Date | Proposal / decision | Owner / approval state | Alternatives | Sections |
 | --- | --- | --- | --- | --- |
 | 2026-09-19 | D1 CLI, optional Jev package and profile v1 recommended | Not approved; superseded as the default recommendation by this revision | D2 or separate configuration | 3, 5, 11, 12 |
@@ -375,6 +386,7 @@ Record any future accepting decision with its actual public link and exact scope
 | E3 | TypeSafe API/confidence/model-limitations pages linked in section 5.3 | Interface and documented limitations; not independent model quality, data terms or a live trial |
 | E4 | PR #4749 and its linked maintainer review | Public request and request-changes rationale; no accepted research/adoption decision |
 | E5 | A/B/C and F01–F12 | Proposed experiments/obligations; unexecuted for this feature |
+| E6 | [Task-progress observation decision record](../../../packages/loopx-jev/DESIGN_DECISIONS.md) | Public synthesis and current-implementation observations; not independent qualification, complete A/B/C or automatic-correction evidence |
 
 ## Appendix D: Deferred mechanisms and rejected shortcuts
 

docs/architecture/rfcs/optional-semantic-assistance-jev-v0.zh-CN.md

@@ -1,7 +1,7 @@
 # RFC：Agent 判断与可选独立评估——以 Jev 为候选方案（v0）
 
 - **RFC status：** Draft；M0 **accepted-for-discussion（接受为讨论稿）**（[维护者决定](https://github.com/loopx-project/loopx/pull/4749#pullrequestreview-5259253204)）。Q1–Q7 仍待决；研究/设计未获实施批准。
-- **Delivery maturity：** Proposal；仅文档，没有接入实现或模型质量验收。
+- **Delivery maturity：** 研究提案；附录 A 单独提出仅 D1 的可选 shadow 实现，没有建立模型质量资格或自动纠正效果。默认关闭的哨兵 capability 及其录制对照见 [`loopx/capabilities/progress_review`](../../../loopx/capabilities/progress_review/README.zh-CN.md)；这些数字不改变本 Draft 的状态。
 - **Created：** 2026-09-19。**Last normative revision：** 2026-09-20。
 - **Implementation baseline：** `9f1916960306b3650d795895b89f331eeae2516e`；在 PR 版本 `27812bd0fb437f831a541b564bcb5be8a96ff77e` 重新核对源码归属与触发器行为。历史 upstream 检查记于附录 A，不构成全系统认证。
 - **Authors / owners：** 提案作者；被选方向由现有领域维护者负责。不新增运行时权威，也未指派实施 owner。
@@ -355,8 +355,18 @@ M0 不默认批准 D1 实施、自动 worker 采纳或隐藏的必需模型阶
 - **增量：** D7/D8、暂定机会顺序、预期价值分解与有界排序比较；同步双语正文及索引。
 - **证据/剩余缺口：** 仅源码检查和文档。没有 live 提供方比较、生产 ranker 或新权限；Q1–Q7 仍待决。
 
+### 2026-09-21 — 任务进展旁路观察实现提案（RFC D1）
+
+- **基线：** upstream `62d18677c`；当前实现只包含可选 D1 命令，不引入 D2–D8 排序或选择器改动。
+- **提案：** 真实 refresh-state 前后限定检查点采集、独立推理、off/shadow 配置和历史读回。参见[操作指南](../../../packages/loopx-jev/DRIFT_SHADOW.zh-CN.md)。
+- **证据边界：** [讨论演进与当前实现的运行结果](../../../packages/loopx-jev/DESIGN_DECISIONS.zh-CN.md)；确定性集成检查不同于模型准确率或任务收益。没有自动干预或原生宿主 hook 推广。
+- **待决：** 维护者是否接受这个可选工具范围，以及独立评估的比较价值。不能将原 M0 收录倒推为实现已获批准。
+
 ## 附录 B：决策日志
 
+单独提出的 [任务进展旁路观察工具及决策记录](../../../packages/loopx-jev/DESIGN_DECISIONS.zh-CN.md)
+不改变历史 M0 决定，也不替代 Q1–Q7。维护者单独评审可选工具范围，实验结果不等于产品采用。
+
 | 日期 | 提案 / 决定 | owner / 批准状态 | 替代方案 | 章节 |
 | --- | --- | --- | --- | --- |
 | 2026-09-19 | 曾推荐 D1 CLI、可选 Jev 包和 profile v1 | 未批准；本修订撤下其默认推荐地位 | D2 或独立配置 | 3、5、11、12 |
@@ -375,6 +385,7 @@ M0 不默认批准 D1 实施、自动 worker 采纳或隐藏的必需模型阶
 | E3 | 第 5.3 节 TypeSafe API/confidence/模型限制页面 | 接口与公开限制，不是独立模型质量、数据条款或 live 实验 |
 | E4 | PR #4749 及链接的维护者评审 | 公开请求和请求修改理由，不是研究/采用已获接受 |
 | E5 | A/B/C 与 F01–F12 | 拟议实验/义务，该功能尚未执行 |
+| E6 | [任务进展观察决策记录](../../../packages/loopx-jev/DESIGN_DECISIONS.zh-CN.md) | 公开论证摘要和当前实现的观察，不是独立资格、完整 A/B/C 或自动纠正证据 |
 
 ## 附录 D：延后机制与排除的捷径
 

examples/progress-review-sentinel-smoke.py

@@ -0,0 +1,92 @@
+#!/usr/bin/env python3
+"""Smoke-test the progress-review sentinel differential from recorded answers.
+
+Replays the committed comparison matrix against the committed provider
+recordings, so it needs no credential and no network. It asserts the harness
+shape and that the replay reproduces the committed summary: the typed repeat
+fuse stays quiet on every self-declared `advanced` round, while each receipt
+signal's first-flag rounds match what the live run recorded.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+import sys
+import tempfile
+
+REPO_ROOT = Path(__file__).resolve().parents[1]
+sys.path.insert(0, str(REPO_ROOT))
+sys.path.insert(1, str(REPO_ROOT / "packages" / "loopx-jev" / "src"))
+
+from loopx_jev.sentinel_compare import COMPARISON_SCHEMA, compare  # noqa: E402
+from loopx_jev.sentinel_matrix import load_sentinel_matrix  # noqa: E402
+
+FIXTURES = REPO_ROOT / "packages" / "loopx-jev" / "tests" / "fixtures" / "sentinel"
+
+
+def deterministic_view(comparison: dict) -> dict:
+    """Project the fields that must reproduce from recordings alone."""
+
+    return {
+        case["case_id"]: {
+            "first_flag_round": case["first_flag_round"],
+            "first_obligation_round": case["first_obligation_round"],
+            "typed_repeat_first_round": case["baseline"]["typed_repeat_first_round"],
+            "statuses": [row["status"] for row in case["rounds"]],
+        }
+        for case in comparison["cases"]
+    }
+
+
+def main() -> int:
+    matrix = load_sentinel_matrix(FIXTURES / "matrix.json")
+    expected_path = FIXTURES / "expected_summary.json"
+    if not expected_path.is_file():
+        print("expected_summary.json is missing; record it with `loopx-jev sentinel compare --live`")
+        return 1
+    expected = json.loads(expected_path.read_text(encoding="utf-8"))
+    with tempfile.TemporaryDirectory(prefix="loopx-sentinel-smoke-") as temporary:
+        comparison = compare(
+            matrix,
+            responses=FIXTURES / "responses",
+            live=False,
+            model=expected["model"],
+            deadline_ms=5000,
+            drift_threshold=2,
+        )
+        Path(temporary, "comparison.json").write_text(json.dumps(comparison), encoding="utf-8")
+    assert comparison["schema_version"] == COMPARISON_SCHEMA
+    assert comparison["execution"] == "recorded_replay"
+    assert comparison["aggregate"]["cases"] == 16
+    assert comparison["aggregate"]["baseline"]["typed_repeat_fired_cases"] == 0
+    for case in comparison["cases"]:
+        assert all(row["status"] != "not_captured" for row in case["rounds"]), case["case_id"]
+        assert all(
+            row["execution_kind"] == "recorded_replay"
+            for row in case["rounds"]
+            if row["status"] in {"completed", "abstained", "failed"}
+        ), case["case_id"]
+    actual = deterministic_view(comparison)
+    if actual != expected["deterministic_view"]:
+        for case_id, view in actual.items():
+            if view != expected["deterministic_view"].get(case_id):
+                print("mismatch", case_id, json.dumps(view), json.dumps(expected["deterministic_view"].get(case_id)))
+        return 1
+    print(
+        json.dumps(
+            {
+                "status": "ok",
+                "cases": comparison["aggregate"]["cases"],
+                "signals": comparison["aggregate"]["signals"],
+                "baseline": comparison["aggregate"]["baseline"]["typed_repeat_fired_cases"],
+            },
+            indent=2,
+            sort_keys=True,
+        )
+    )
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

loopx/capabilities/catalog.py

@@ -28,6 +28,7 @@
 from .connector_registry.catalog_entry import CONNECTOR_REGISTRY_CATALOG_ENTRY
 from .external_research.catalog_entry import EXTERNAL_RESEARCH_CATALOG_ENTRY
 from .reliability_diagnostics.catalog_entry import RELIABILITY_DIAGNOSTICS_CATALOG_ENTRY
+from .progress_review.catalog_entry import PROGRESS_REVIEW_CATALOG_ENTRY
 from .registry import CapabilityRegistry
 
 CAPABILITY_CATALOG_SCHEMA_VERSION = "loopx_capability_catalog_v0"
@@ -56,6 +57,7 @@
     CONNECTOR_REGISTRY_CATALOG_ENTRY,
     EXTERNAL_RESEARCH_CATALOG_ENTRY,
     RELIABILITY_DIAGNOSTICS_CATALOG_ENTRY,
+    PROGRESS_REVIEW_CATALOG_ENTRY,
 )
 # Preserve the original import surface while routing all reads through the registry.
 CAPABILITIES = BUILTIN_CAPABILITIES

loopx/capabilities/configuration_ui.py

@@ -6,6 +6,13 @@
 
 from ..configuration_transaction import configuration_payload_revision
 
+from .progress_review.policy import (
+    PROGRESS_REVIEW_MAX_DRIFT_THRESHOLD,
+    PROGRESS_REVIEW_MIN_DRIFT_THRESHOLD,
+    PROGRESS_REVIEW_MODES,
+    PROGRESS_REVIEW_SIGNALS,
+)
+
 CAPABILITY_CONFIGURATION_CATALOG_SCHEMA = "capability_configuration_catalog_v0"
 CAPABILITY_CONFIGURATION_EDITOR_SCHEMA = "capability_configuration_editor_v0"
 CAPABILITY_CONFIGURATION_RESOLUTION_SCHEMA = "capability_configuration_resolution_v0"
@@ -288,6 +295,41 @@ def capability_configuration_editor(
                 _field("strict_receipt", "Require an exact-diff receipt", "boolean"),
             ],
         },
+        "progress_review": {
+            "supported_scopes": ["goal"],
+            "writable_scopes": ["goal"],
+            "fields": [
+                _field(
+                    "mode",
+                    "Mode",
+                    "select",
+                    options=PROGRESS_REVIEW_MODES,
+                    required=True,
+                    description=(
+                        "off records nothing; shadow records typed receipts only; "
+                        "assist lets consecutive drift receipts raise the existing "
+                        "autonomous replan obligation. No pause or gate authority."
+                    ),
+                ),
+                _field(
+                    "signal",
+                    "Drift signal",
+                    "select",
+                    options=PROGRESS_REVIEW_SIGNALS,
+                    description=(
+                        "Which receipt judgment counts as drift: the Noul behavior/"
+                        "acceptance pair or the Choice relation/increment pair."
+                    ),
+                ),
+                _field(
+                    "drift_threshold",
+                    "Consecutive drift receipts before an obligation",
+                    "integer",
+                    minimum=PROGRESS_REVIEW_MIN_DRIFT_THRESHOLD,
+                    maximum=PROGRESS_REVIEW_MAX_DRIFT_THRESHOLD,
+                ),
+            ],
+        },
         "pull_request_review": {
             "supported_scopes": ["machine", "goal"],
             "writable_scopes": ["machine", "goal"],