From b9fc98d6e4a049fdd02fe599f22a00676e829f71 Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 10:58:57 +0800
Subject: [PATCH 1/6] feat(progress-review): bind observations to current task
 criteria

Signed-off-by: song <liusongstep@gmail.com>
---
 apps/presentation/dashboard/src/data/chat.ts  | 15 +++++
 .../goal-capability-settings.tsx              | 42 ++++++++++--
 .../personal-workspace/personal-workspace.css |  6 ++
 loopx/capabilities/progress_review/context.py | 46 +++++++++++--
 loopx/capabilities/progress_review/receipt.py | 11 +++
 loopx/chat_goal_configuration_api.py          |  8 +++
 .../control_plane/effect_runtime_handlers.ts  |  2 +
 .../work_items/external_progress_review.py    |  8 +--
 .../work_items/progress_review_evidence.ts    | 67 +++++++++++++++++++
 packages/loopx-jev/src/loopx_jev/drift.py     | 34 +++++++++-
 packages/loopx-jev/src/loopx_jev/runner.py    | 40 ++++++++++-
 11 files changed, 262 insertions(+), 17 deletions(-)
 create mode 100644 loopx/control_plane/work_items/progress_review_evidence.ts

diff --git a/apps/presentation/dashboard/src/data/chat.ts b/apps/presentation/dashboard/src/data/chat.ts
index 53ab43680f..3d42a62f0c 100644
--- a/apps/presentation/dashboard/src/data/chat.ts
+++ b/apps/presentation/dashboard/src/data/chat.ts
@@ -1650,6 +1650,20 @@ export const capabilityConfigurationEditorSchema = z.object({
   read_only_reason: z.string().optional(),
 });
 
+export const progressReviewObservationSchema = z.object({
+  mode: z.string(), read_state: z.enum(["observed", "missing", "unavailable"]), read_error: z.string().optional(),
+  receipt_count: z.number(), rejected_receipts: z.number(), stale_receipts: z.number(),
+  contract_revision: z.string().nullable(), authority: z.literal("none"),
+  latest: z.object({status: z.string(), event_id: z.string(), evidence_id: z.string(),
+    judgments: z.object({choice: z.object({relation: z.string().nullable(), increment: z.string().nullable()}).nullable(),
+      noul: z.object({serves_acceptance: z.number().nullable(), evidence_increment: z.number().nullable(), behavior_change: z.number().nullable()}).nullable()}),
+    criterion_current: z.boolean().optional(),
+    evidence_scope: z.object({coverage: z.literal("declared_file_net_change"), files: z.array(z.string()),
+      criterion_binding: z.object({origin: z.enum(["operator_study", "goal_acceptance"]), criteria_sha256: z.string(),
+        todo_id: z.string().optional(), contract_digest: z.string().optional(), contract_revision: z.number().optional(),
+        criterion_ids: z.array(z.string()).optional()})}).optional(),
+  }).nullable(),
+});
 export const capabilityConfigurationCatalogSchema = z.object({
   schema_version: z.literal("capability_configuration_catalog_v0"),
   capabilities: z.array(z.object({
@@ -1675,6 +1689,7 @@ export const capabilityConfigurationCatalogSchema = z.object({
       effective_revision: z.string(),
     }).optional(),
     documentation: z.record(z.string(), z.unknown()).optional(),
+    observation: progressReviewObservationSchema.optional(),
     context_contribution: z.object({
       supported_phases: z.array(z.enum(["before_plan", "before_delegate", "after_delegate_result"])),
       target: z.literal("coordinator"),
diff --git a/apps/presentation/dashboard/src/features/personal-workspace/goal-capability-settings.tsx b/apps/presentation/dashboard/src/features/personal-workspace/goal-capability-settings.tsx
index c246f65cbe..6339ba8475 100644
--- a/apps/presentation/dashboard/src/features/personal-workspace/goal-capability-settings.tsx
+++ b/apps/presentation/dashboard/src/features/personal-workspace/goal-capability-settings.tsx
@@ -1,4 +1,4 @@
-import { useEffect, useMemo, useState } from "react";
+import { useEffect, useMemo, useRef, useState } from "react";
 import { AlertTriangle, Code2, LoaderCircle, RefreshCw } from "lucide-react";
 
 import {
@@ -273,6 +273,33 @@ function CapabilityCatalog({ callbacks, catalog, goalId, notification, onApplied
         {localizedSelected.capability_id === "content_ops" ? <ContentReferenceWorkbench key={goalId} /> : <CapabilityEditorStatus available={editorAvailable} t={t}
           description={readOnlyReason} />}
 
+        {selected.capability_id === "progress_review" && selected.observation ? <section className="personal-capability-observation" aria-label={locale === "zh-CN" ? "审查证据读回" : "Review evidence readback"}>
+          <h3>{locale === "zh-CN" ? "审查证据读回" : "Review evidence readback"}</h3>
+          <button type="button" onClick={onApplied}>{locale === "zh-CN" ? "重新读取审查证据" : "Recheck review evidence"}</button>
+          <p>{selected.observation.read_state === "unavailable" ? (locale === "zh-CN" ? "审查存储不可读，当前判断未知。" : "Review storage unavailable; current judgment unknown.")
+            : selected.observation.read_state === "missing" ? (locale === "zh-CN" ? "尚无审查回执。" : "No review receipts yet.")
+            : `${selected.observation.receipt_count} ${locale === "zh-CN" ? "条已读取回执" : "loaded receipts"}`}</p>
+          <p>{locale === "zh-CN" ? "过期 / 拒收" : "Stale / rejected"}: {selected.observation.stale_receipts} / {selected.observation.rejected_receipts}</p>
+          {selected.observation.latest ? <>
+            <p>{locale === "zh-CN" ? "最新审查" : "Latest review"}: {selected.observation.latest.status}</p>
+            <p>{locale === "zh-CN" ? "与目标的关系" : "Relation to goal"}: {selected.observation.latest.judgments.choice?.relation ?? "unknown"}</p>
+            <p>{locale === "zh-CN" ? "新证据增量" : "Evidence increment"}: {selected.observation.latest.judgments.choice?.increment ?? "unknown"}</p>
+            {selected.observation.latest.evidence_scope ? <>
+              <p>{selected.observation.latest.evidence_scope.criterion_binding.origin === "goal_acceptance"
+                ? (locale === "zh-CN" ? "绑定任务的规范验收条款" : "Bound to canonical task acceptance criteria")
+                : (locale === "zh-CN" ? "依据操作者研究条款；不证明规范验收关联" : "Operator study basis; canonical acceptance binding is unverified")}</p>
+              {selected.observation.latest.criterion_current === false ? <p role="alert">{locale === "zh-CN" ? "条款或任务关联已变化，旧判断不可作为当前依据。" : "Criteria or task binding changed; the old judgment is unavailable for current use."}</p> : null}
+              <p>{locale === "zh-CN" ? "仅观察声明文件在两个检查点之间的净变化" : "Observed only the net change of declared files between two checkpoints"}</p>
+              <details><summary>{locale === "zh-CN" ? "观察范围与条款版本" : "Observed scope and criterion version"}</summary>
+                <code>{selected.observation.latest.evidence_scope.criterion_binding.criteria_sha256}</code>
+                <p>{selected.observation.latest.evidence_scope.criterion_binding.criterion_ids?.join(", ")}</p>
+                <p>{selected.observation.latest.evidence_scope.files.join(", ")}</p>
+              </details>
+            </> : <p>{locale === "zh-CN" ? "旧回执未记录所选条款及文件覆盖范围。" : "Legacy receipt did not record selected criteria or file coverage."}</p>}
+          </> : null}
+          <p>{locale === "zh-CN" ? "此观察不证明任务完成；未命中漂移条件也不等于工作符合目标。" : "This observation does not prove completion; a drift condition not met does not establish on-goal work."}</p>
+        </section> : null}
+
         {localizedSelected.capability_id === "lark_event_inbox" ? (
           <section className="personal-capability-linked-setting">
             <div>
@@ -361,18 +388,23 @@ export function GoalCapabilitySettings({ callbacks, goalId, notification, onChan
   const [inspection, setInspection] = useState<GoalConfigurationInspection | null>(null);
   const [error, setError] = useState<string | null>(null);
   const [loading, setLoading] = useState(false);
+  const readGeneration = useRef(0);
 
   function load() {
     if (!goalId) return;
+    const generation = ++readGeneration.current;
     setLoading(true);
     setError(null);
+    // Withdraw the previous observation during a recheck; preserve navigation.
+    setInspection(current => current ? {...current, capability_catalog: {...current.capability_catalog,
+      capabilities: current.capability_catalog.capabilities.map(item => item.observation ? {...item, observation: undefined} : item)}} : null);
     void fetchGoalConfiguration(goalId)
-      .then(setInspection)
-      .catch((reason: unknown) => setError(reason instanceof Error ? reason.message : t("capabilities.loadFailed")))
-      .finally(() => setLoading(false));
+      .then(value => {if (generation === readGeneration.current) setInspection(value);})
+      .catch((reason: unknown) => {if (generation === readGeneration.current) setError(reason instanceof Error ? reason.message : t("capabilities.loadFailed"));})
+      .finally(() => {if (generation === readGeneration.current) setLoading(false);});
   }
 
-  useEffect(load, [goalId]);
+  useEffect(() => {setInspection(null); load(); return () => {readGeneration.current++;};}, [goalId]);
 
   if (!goalId) {
     return <p className="personal-capability-empty">{t("capabilities.chooseGoal")}</p>;
diff --git a/apps/presentation/dashboard/src/features/personal-workspace/personal-workspace.css b/apps/presentation/dashboard/src/features/personal-workspace/personal-workspace.css
index 32040d72f4..95ad18a5cc 100644
--- a/apps/presentation/dashboard/src/features/personal-workspace/personal-workspace.css
+++ b/apps/presentation/dashboard/src/features/personal-workspace/personal-workspace.css
@@ -902,6 +902,12 @@ button.personal-execution-chip:focus-visible { outline: 2px solid #0070f3; outli
 .personal-capability-editor-status p { margin-top: 3px; }
 .personal-capability-editor-status.is-preview { background: var(--pw-blue-soft); color: var(--pw-blue-ink); }
 .personal-capability-editor-status.is-read-only { background: #fff5e8; color: #9a5a10; }
+.personal-capability-observation { display: grid; gap: 10px; margin: 20px 24px; min-width: 0; overflow-wrap: anywhere; }
+.personal-capability-observation h3 { margin: 0; font-size: 14px; font-weight: 600; }
+.personal-capability-observation button { justify-self: start; min-height: 44px; padding: 8px 12px; border: 1px solid var(--pw-line-strong); border-radius: 6px; background: var(--pw-card); color: var(--pw-text); font: inherit; font-size: 12px; cursor: pointer; }
+.personal-capability-observation details { font-size: 12px; }
+.personal-capability-observation summary { min-height: 32px; cursor: pointer; }
+.personal-capability-observation code { display: block; font-size: 11px; overflow-wrap: anywhere; }
 .personal-capability-linked-setting { display: grid; grid-template-columns: minmax(0, 1fr) auto; align-items: center; gap: 16px; margin: 0 24px 20px; padding: 14px; border: 1px solid var(--pw-line); border-radius: 10px; background: var(--pw-bg); }
 .personal-capability-linked-setting > div { display: grid; gap: 4px; }
 .personal-capability-linked-controls { min-width: 0; justify-items: start; }
diff --git a/loopx/capabilities/progress_review/context.py b/loopx/capabilities/progress_review/context.py
index 2fa8e24d81..8f8eed1530 100644
--- a/loopx/capabilities/progress_review/context.py
+++ b/loopx/capabilities/progress_review/context.py
@@ -33,10 +33,44 @@ def external_progress_review_context(
     goal_id = str(goal.get("id") or "").strip()
     if policy["mode"] == "off" or runtime_root is None or not goal_id:
         return None
+    from ...control_plane.effect_runtime import effect_runtime_result, EffectRuntimeRemoteError
+    read_error = None
     try:
         loaded, rejected = load_progress_review_receipts(Path(runtime_root), goal_id)
-    except (OSError, ValueError):
+    except (OSError, ValueError, EffectRuntimeRemoteError):
         loaded, rejected = [], 0
+        read_error = "receipt_store_unavailable"
+    # Only newly scoped canonical observations need this current-owner check.
+    # Old receipts and manual studies retain their existing policy semantics.
+    current_by_todo = {}
+    qualified = []
+    for receipt in loaded:
+        binding = receipt.get("evidence_scope", {}).get("criterion_binding", {})
+        if binding.get("origin") == "goal_acceptance":
+            todo_id = binding["todo_id"]
+            if todo_id not in current_by_todo:
+                try:
+                    observation = effect_runtime_result("goal.acceptance.inspect", {
+                        "goal_id": goal_id, "runtime_root": str(runtime_root), "todo_id": todo_id,
+                    })
+                    current_by_todo[todo_id] = observation.get("completion_requirements")
+                except (OSError, ValueError, EffectRuntimeRemoteError):
+                    current_by_todo[todo_id] = None
+            try:
+                requirements = current_by_todo[todo_id]
+                selected = effect_runtime_result("progress_review.criterion_basis", {
+                    "requirements": requirements, "criterion_ids": binding["criterion_ids"], "acceptance": [],
+                    "goal_id": goal_id, "agent_id": receipt["run"]["agent_id"],
+                })["binding"] if requirements is not None else None
+                current = selected == binding
+            except (ValueError, EffectRuntimeRemoteError):
+                current = False
+            receipt = {**receipt, "criterion_current": current}
+            if not current:
+                receipt.update(status="stale", reason="criterion_basis_unavailable",
+                               drift_signal={"noul": None, "choice": None})
+        qualified.append(receipt)
+    loaded = qualified
     pinned = policy.get("contract_revision")
     # Newest by run order, never by the observer's local sequence counter.
     newest = max(loaded, key=progress_review_receipt_order_key, default=None)
@@ -46,10 +80,10 @@ def external_progress_review_context(
         # pin is manual, so a newer receipt under another revision means the
         # observer basis moved and the Goal owner must re-pin (or not).
         receipts = [item for item in loaded if item["contract_revision"] == pinned]
-        stale = len(loaded) - len(receipts)
+        stale = len(loaded) - len(receipts) + sum(item.get("criterion_current") is False for item in receipts)
     else:
-        receipts, stale = loaded, 0
-    return {
+        receipts, stale = loaded, sum(item.get("criterion_current") is False for item in loaded)
+    result = {
         "policy": policy,
         "receipts": receipts,
         "summary": progress_review_receipt_summary(
@@ -60,6 +94,10 @@ def external_progress_review_context(
             newest_contract_revision=newest_revision,
         ),
     }
+    result["summary"]["read_state"] = "unavailable" if read_error else "observed" if loaded else "missing"
+    if read_error:
+        result["summary"]["read_error"] = read_error
+    return result
 
 
 __all__ = ["external_progress_review_context"]
diff --git a/loopx/capabilities/progress_review/receipt.py b/loopx/capabilities/progress_review/receipt.py
index 336abe35ee..c3706e3c7f 100644
--- a/loopx/capabilities/progress_review/receipt.py
+++ b/loopx/capabilities/progress_review/receipt.py
@@ -267,6 +267,14 @@ def normalize_progress_review_receipt(value: Any) -> dict[str, Any]:
         or float(recorded_at) < 0
     ):
         raise TypeError("receipt.recorded_at must be a non-negative epoch number")
+    scope = None
+    if "evidence_scope" in value:
+        from ...control_plane.effect_runtime import effect_runtime_result, EffectRuntimeRejected
+        try:
+            scope = effect_runtime_result("progress_review.evidence_scope", {"scope": value["evidence_scope"],
+                "goal_id": value.get("goal_id"), "run": value.get("run")})["scope"]
+        except EffectRuntimeRejected as exc:
+            raise ValueError("invalid progress-review evidence scope") from exc
     event_id = _hex64(value.get("event_id"), field="event_id")
     return {
         "schema_version": PROGRESS_REVIEW_RECEIPT_SCHEMA_VERSION,
@@ -291,6 +299,7 @@ def normalize_progress_review_receipt(value: Any) -> dict[str, Any]:
         "usage": usage,
         "recorded_at": float(recorded_at),
         "authority": "none",
+        **({"evidence_scope": scope} if scope is not None else {}),
     }
 
 
@@ -426,6 +435,8 @@ def progress_review_receipt_summary(
                 "drift_signal": dict(receipt["drift_signal"]),
                 "model": receipt["model"],
                 "question_version": receipt["question_version"],
+                **({"evidence_scope": receipt["evidence_scope"]} if "evidence_scope" in receipt else {}),
+                **({"criterion_current": receipt["criterion_current"]} if "criterion_current" in receipt else {}),
             }
     summary: dict[str, Any] = {
         "schema_version": "progress_review_status_v0",
diff --git a/loopx/chat_goal_configuration_api.py b/loopx/chat_goal_configuration_api.py
index bd65e9b87a..0367e6c164 100644
--- a/loopx/chat_goal_configuration_api.py
+++ b/loopx/chat_goal_configuration_api.py
@@ -417,6 +417,14 @@ def _goal_configuration_inspect(self) -> None:
                 result,
                 machine_namespaces=self._goal_configuration_machine_namespaces(),
             )
+            from .agent_registry import load_goal_from_registry
+            from .capabilities.progress_review.context import external_progress_review_context
+            goal = load_goal_from_registry(self.server.registry_path, goal_ids[0].strip())
+            observation = external_progress_review_context(goal or {}, getattr(self.server, "runtime_root", None))
+            if observation is not None:
+                for capability in response.get("capability_catalog", {}).get("capabilities", []):
+                    if capability.get("capability_id") == "progress_review":
+                        capability["observation"] = observation["summary"]
         except (FileNotFoundError, KeyError, TypeError, ValueError) as exc:
             self._send_error(
                 str(exc),
diff --git a/loopx/control_plane/effect_runtime_handlers.ts b/loopx/control_plane/effect_runtime_handlers.ts
index beb2f767ae..da920408e9 100644
--- a/loopx/control_plane/effect_runtime_handlers.ts
+++ b/loopx/control_plane/effect_runtime_handlers.ts
@@ -605,6 +605,8 @@ export function createEffectRuntimeHandlers(
     ["collaboration.delegation.observe_wake", lazyHandler(() => import("./collaboration/delegation.ts"), ({decideDelegationWakeObservation}) => decideDelegationWakeObservation)],
     ["collaboration.delegation.recover_validated_settlement", lazyHandler(() => import("./collaboration/delegation.ts"), ({recoverValidatedDelegationSettlement}) => recoverValidatedDelegationSettlement)],
     ["collaboration.delegation.stop", lazyHandler(() => import("./collaboration/delegation.ts"), ({decideDelegationStop}) => decideDelegationStop)],
+    ["progress_review.evidence_scope", lazyHandler(() => import("./work_items/progress_review_evidence.ts"), ({progressReviewEvidenceScope}) => progressReviewEvidenceScope)],
+    ["progress_review.criterion_basis", lazyHandler(() => import("./work_items/progress_review_evidence.ts"), ({progressReviewCriterionBasis}) => progressReviewCriterionBasis)],
     ["collaboration.delegation.adoption", lazyHandler(() => import("./collaboration/delegation.ts"), ({recordDelegationAdoption}) => recordDelegationAdoption)],
     [
       "collaboration.request.normalize",
diff --git a/loopx/control_plane/work_items/external_progress_review.py b/loopx/control_plane/work_items/external_progress_review.py
index b32fb21806..1328ac0698 100644
--- a/loopx/control_plane/work_items/external_progress_review.py
+++ b/loopx/control_plane/work_items/external_progress_review.py
@@ -169,7 +169,7 @@ def _verdict(
     signal: str,
     pinned: str,
 ) -> Verdict:
-    """Classify one captured transition as drift, on_goal or unevaluated."""
+    """Classify one captured transition as drift, condition_not_met or unevaluated."""
 
     if receipt is None:
         return "unevaluated", "unattributed" if ambiguous else "missing"
@@ -192,7 +192,7 @@ def _verdict(
     if value is True:
         return "drift", None
     if value is False:
-        return "on_goal", None
+        return "condition_not_met", None
     return "unevaluated", "undecided"
 
 
@@ -272,7 +272,7 @@ def external_progress_review_trigger(
       newer `completed` receipt whose selected signal is False: the policy's
       condition no longer holds, so nothing is open;
     - every transition is `drift` (completed, signal True, pinned revision),
-      `on_goal` (completed, signal False) or `unevaluated` for one typed
+      `condition_not_met` (completed, signal False) or `unevaluated` for one typed
       reason in EXTERNAL_PROGRESS_REVIEW_UNEVALUATED_REASONS;
     - a streak forms only from `threshold` consecutive drift transitions with
       no unevaluated transition between them (conservative formation);
@@ -343,7 +343,7 @@ def external_progress_review_trigger(
         verdict, reason = _verdict(
             run, receipt, ambiguous=ambiguous, signal=signal, pinned=pinned
         )
-        if verdict == "on_goal":
+        if verdict == "condition_not_met":
             break
         if verdict == "drift":
             assert receipt is not None
diff --git a/loopx/control_plane/work_items/progress_review_evidence.ts b/loopx/control_plane/work_items/progress_review_evidence.ts
new file mode 100644
index 0000000000..f1ebc827ee
--- /dev/null
+++ b/loopx/control_plane/work_items/progress_review_evidence.ts
@@ -0,0 +1,67 @@
+/** Criterion selection from authorized current facts; observation grants no authority. */
+import type {JsonObject} from "../effect_program.ts";
+import {requireJsonObject} from "../runtime_decode.ts";
+import {EffectRuntimeRequestError} from "../effect_runtime_errors.ts";
+import {canonicalAuthoritySha256} from "../coordination/authority_store_codec.ts";
+export function progressReviewCriterionBasis(params: JsonObject): JsonObject {
+  function requireThat(condition: unknown, message: string): asserts condition {
+    if (!condition) throw new EffectRuntimeRequestError(message);
+  }
+  const requested = params.criterion_ids;
+  if (params.requirements === null) {
+    requireThat(Array.isArray(params.acceptance) && params.acceptance.length > 0 && params.acceptance.length <= 128
+      && params.acceptance.every(row => typeof row === "string" && row.length > 0 && row.length <= 32768),
+    "bounded operator study criteria required");
+    return {acceptance: params.acceptance, binding: {origin: "operator_study",
+      criteria_sha256: canonicalAuthoritySha256(params.acceptance)}};
+  }
+  const basis = requireJsonObject(params.requirements, "current acceptance requirements");
+  requireThat(Array.isArray(requested) && requested.length > 0 && requested.length <= 128
+    && requested.every(row => typeof row === "string" && row.length > 0)
+    && new Set(requested).size === requested.length && Array.isArray(basis.criteria),
+  "explicit current criterion selection required");
+  const criteria = basis.criteria.map(raw => requireJsonObject(raw, "canonical acceptance criterion"))
+    .filter(row => requested.includes(row.id));
+  requireThat(criteria.length === requested.length && new Set(criteria.map(row => row.id)).size === criteria.length,
+    "selected criteria must belong to this current task acceptance scope");
+  requireThat(criteria.every(row => typeof row.description === "string" && row.description.length > 0
+    && row.description.length <= 32768), "bounded canonical criterion descriptions required");
+  requireThat(typeof basis.contract_digest === "string" && Number.isInteger(basis.contract_revision)
+    && typeof basis.todo_id === "string" && typeof basis.todo_semantic_digest === "string",
+  "current canonical acceptance identity required");
+  requireThat(typeof params.goal_id === "string" && params.goal_id.length > 0
+    && typeof params.agent_id === "string" && params.agent_id.length > 0, "canonical goal and agent identity required");
+  return {acceptance: criteria.map(row => row.description), binding: {origin: "goal_acceptance",
+    goal_id: params.goal_id, agent_id: params.agent_id,
+    todo_id: basis.todo_id, contract_digest: basis.contract_digest, contract_revision: basis.contract_revision,
+    todo_semantic_digest: basis.todo_semantic_digest, criterion_ids: criteria.map(row => row.id),
+    criteria_sha256: canonicalAuthoritySha256(criteria)}};
+}
+
+/** Decode path-free observation metadata; this does not verify model answers. */
+export function progressReviewEvidenceScope(params: JsonObject): JsonObject {
+  const scope = requireJsonObject(params.scope, "progress-review evidence scope");
+  const binding = requireJsonObject(scope.criterion_binding, "progress-review criterion binding");
+  const fail = (ok: boolean) => {if (!ok) throw new EffectRuntimeRequestError("invalid bounded progress-review evidence scope");};
+  fail(scope.coverage === "declared_file_net_change" && Array.isArray(scope.files) && scope.files.length <= 32
+    && scope.files.every(ref => typeof ref === "string" && ref.length > 0 && ref.length <= 1024
+      && !ref.startsWith("/") && !ref.includes("\\") && !ref.split("/").includes("..")));
+  fail(binding.origin === "operator_study" || binding.origin === "goal_acceptance");
+  fail(typeof binding.criteria_sha256 === "string" && /^[a-f0-9]{64}$/.test(binding.criteria_sha256));
+  const normalized: JsonObject = {origin: binding.origin, criteria_sha256: binding.criteria_sha256};
+  if (binding.origin === "goal_acceptance") {
+    fail(typeof binding.goal_id === "string" && binding.goal_id.length > 0 && binding.goal_id.length <= 200
+      && typeof binding.agent_id === "string" && binding.agent_id.length > 0 && binding.agent_id.length <= 200
+      && typeof binding.todo_id === "string" && binding.todo_id.length > 0 && binding.todo_id.length <= 200
+      && typeof binding.contract_digest === "string" && /^[a-f0-9]{64}$/.test(binding.contract_digest)
+      && Number.isInteger(binding.contract_revision) && Number(binding.contract_revision) >= 1
+      && typeof binding.todo_semantic_digest === "string" && /^[a-f0-9]{64}$/.test(binding.todo_semantic_digest)
+      && Array.isArray(binding.criterion_ids) && binding.criterion_ids.length > 0 && binding.criterion_ids.length <= 128
+      && binding.criterion_ids.every(id => typeof id === "string" && id.length > 0 && id.length <= 200));
+    const run = requireJsonObject(params.run, "criterion-bound run identity");
+    fail(params.goal_id === binding.goal_id && run.agent_id === binding.agent_id && run.todo_id === binding.todo_id);
+    for (const key of ["goal_id", "agent_id", "todo_id", "contract_digest", "contract_revision", "todo_semantic_digest", "criterion_ids"])
+      normalized[key] = binding[key];
+  }
+  return {scope: {criterion_binding: normalized, coverage: scope.coverage, files: scope.files}};
+}
diff --git a/packages/loopx-jev/src/loopx_jev/drift.py b/packages/loopx-jev/src/loopx_jev/drift.py
index 7127bb6c76..2daac2d7cc 100644
--- a/packages/loopx-jev/src/loopx_jev/drift.py
+++ b/packages/loopx-jev/src/loopx_jev/drift.py
@@ -69,6 +69,8 @@ def contract(path: Path, repo: Path) -> tuple[dict[str, Any], str, Callable[[],
     if not isinstance(basis.get("goal_id"), str) or not basis["goal_id"].strip():
         raise ValueError("missing_goal_identity")
     _, revision = read_json(path, 32768)
+    if basis.get("criterion_binding", {}).get("origin") == "goal_acceptance":
+        revision = digest({"manifest": revision, "criterion_binding": basis["criterion_binding"]})
     return basis, revision, guard
 
 
@@ -170,6 +172,20 @@ def invalidate_baseline(root: Path) -> None:
         atomic_json(root / "state.json", current)
 
 
+def _require_criterion_run(basis, goal_id, run, paths):
+    # The core decoder owns this identity rule for both capture and receipt use.
+    if basis.get("criterion_binding", {}).get("origin") == "goal_acceptance":
+        from loopx.control_plane.effect_runtime import effect_runtime_result, EffectRuntimeRejected
+        try:
+            effect_runtime_result("progress_review.evidence_scope", {
+                "goal_id": goal_id, "run": run,
+                "scope": {"criterion_binding": basis["criterion_binding"],
+                          "coverage": "declared_file_net_change", "files": paths},
+            })
+        except EffectRuntimeRejected as exc:
+            raise ValueError("run_acceptance_scope_mismatch") from exc
+
+
 def enqueue(
     root: Path,
     prepared: dict[str, Any],
@@ -186,6 +202,7 @@ def enqueue(
             "generated_at"
         ):
             raise ValueError("run_goal_or_identity_mismatch")
+        _require_criterion_run(prepared["basis"], current["goal_id"], record, current["paths"])
         # Checkpoint supplements for the same bound Turn are the same transition.
         identity = (
             {
@@ -354,19 +371,27 @@ def drain(
 
             def current() -> bool:
                 try:
+                    _require_criterion_run(job["basis"], initial["goal_id"], job["run"], job["paths"])
                     now = policy(config_path)
+                    # Legacy studies keep their manifest-only guard. New canonical
+                    # scopes must recheck the same complete owner-bound revision.
+                    if job.get("basis", {}).get("criterion_binding", {}).get("origin") == "goal_acceptance":
+                        _, revision, basis_current = contract(Path(initial["basis_path"]), Path(initial["repo"]))
+                        if not basis_current():
+                            return False
+                    else:
+                        _, revision = read_json(Path(initial["basis_path"]), 32768)
                     return (
                         now.mode == "shadow"
                         and now.generation == job["config_generation"]
                         and state(root)["configuration_epoch"]
                         == job["configuration_epoch"]
-                        and read_json(Path(initial["basis_path"]), 32768)[1]
-                        == job["contract_revision"]
+                        and revision == job["contract_revision"]
                         and read_json(job_path, 256 * 1024)[1] == job_digest
                         and read_json(Path(job["source_record"]), 4 * 1024 * 1024)[1]
                         == job["record_digest"]
                     )
-                except (OSError, ValueError, KeyError, TypeError):
+                except (OSError, ValueError, KeyError, TypeError, RuntimeError):
                     return False
 
             options: dict[str, Any] = {}
@@ -472,6 +497,9 @@ def _emit_receipt(
         "timing_ns": timing,
         "usage": result.get("usage") if isinstance(result.get("usage"), dict) else None,
         "recorded_at": time.time(),
+        **({"evidence_scope": {"criterion_binding": job["basis"]["criterion_binding"],
+            "coverage": "declared_file_net_change", "files": job.get("paths", [])}}
+           if (job.get("basis") or {}).get("criterion_binding") else {}),
     }
     try:
         path = write_progress_review_receipt(runtime_root, goal_id, receipt)
diff --git a/packages/loopx-jev/src/loopx_jev/runner.py b/packages/loopx-jev/src/loopx_jev/runner.py
index 8d86ee6c9f..9810552c92 100644
--- a/packages/loopx-jev/src/loopx_jev/runner.py
+++ b/packages/loopx-jev/src/loopx_jev/runner.py
@@ -31,6 +31,7 @@ def read_basis(
         "horizon",
         "evidence",
         "already_known",
+        "acceptance_scope",
     }
     if not isinstance(manifest, dict) or set(manifest) - allowed:
         raise ValueError("unknown basis fields")
@@ -85,16 +86,53 @@ def read_basis(
             }
         )
         checks.append((target, digest))
+    from loopx.control_plane.effect_runtime import effect_runtime_result
+    from loopx.control_plane.goals.acceptance import inspect_goal_acceptance
+
+    scope = manifest.get("acceptance_scope")
+    def criterion_basis():
+        requirements = None
+        requested = None
+        if scope is not None:
+            if not isinstance(scope, dict) or set(scope) != {
+                "registry_ref", "runtime_ref", "agent_id", "todo_id", "criterion_ids"
+            }:
+                raise ValueError("explicit current task acceptance scope required")
+            def local_reference(key):
+                value = scope[key]
+                if not isinstance(value, str) or not value:
+                    raise ValueError("local acceptance reference required")
+                relative = Path(value)
+                target = workspace / relative
+                if relative.is_absolute() or ".." in relative.parts or target.is_symlink() or not target.resolve().is_relative_to(workspace.resolve()):
+                    raise ValueError("acceptance reference escapes the selected workspace")
+                return target
+            current = inspect_goal_acceptance(registry_path=local_reference("registry_ref"),
+                runtime_root=str(local_reference("runtime_ref")), goal_id=manifest.get("goal_id"),
+                agent_id=scope["agent_id"], todo_id=scope["todo_id"])
+            requirements = current.get("completion_requirements")
+            if requirements is None:
+                raise ValueError("current task has no canonical acceptance criteria")
+            requested = scope["criterion_ids"]
+        return effect_runtime_result("progress_review.criterion_basis", {
+            "acceptance": manifest["acceptance"], "requirements": requirements, "criterion_ids": requested,
+            "goal_id": manifest.get("goal_id"), "agent_id": scope.get("agent_id") if scope else None,
+        })
+    selected = criterion_basis()
     basis = {
         **manifest,
         "evidence": observations,
         "basis_origin": "explicit_operator_study_basis_not_completion_authority",
+        "acceptance": selected["acceptance"],
+        "criterion_binding": selected["binding"],
     }
 
     def current() -> bool:
         try:
             if read_json(path, 32768)[1] != manifest_hash:
                 return False
+            if scope is not None and criterion_basis() != selected:
+                return False
             for target, digest in checks:
                 if target.is_symlink() or not target.resolve().is_relative_to(
                     workspace.resolve()
@@ -105,7 +143,7 @@ def current() -> bool:
                 if len(raw) > 32768 or hashlib.sha256(raw).hexdigest() != digest:
                     return False
             return True
-        except (OSError, ValueError):
+        except (OSError, ValueError, RuntimeError):
             return False
 
     return basis, current

From 3a9e308f5949ca6e34aec3f41ade78e11c09f610 Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 10:58:57 +0800
Subject: [PATCH 2/6] docs: qualify criterion-bound review

Signed-off-by: song <liusongstep@gmail.com>
---
 .../agent-failure-containment-and-recovery.md |  3 +-
 ...-failure-containment-and-recovery.zh-CN.md |  2 +-
 ...state-machines-recovery-verification-v0.md | 17 +++++---
 ...machines-recovery-verification-v0.zh-CN.md | 14 +++++--
 .../optional-semantic-assistance-jev-v0.md    | 39 +++++++++++--------
 ...tional-semantic-assistance-jev-v0.zh-CN.md | 30 ++++++++------
 loopx/capabilities/progress_review/README.md  | 29 ++++++++++++--
 .../progress_review/README.zh-CN.md           | 22 +++++++++--
 packages/loopx-jev/DRIFT_SHADOW.md            | 36 ++++++++++++++++-
 packages/loopx-jev/DRIFT_SHADOW.zh-CN.md      | 27 +++++++++++++
 10 files changed, 170 insertions(+), 49 deletions(-)

diff --git a/docs/architecture/agent-failure-containment-and-recovery.md b/docs/architecture/agent-failure-containment-and-recovery.md
index 307e792287..fbd7ae43c5 100644
--- a/docs/architecture/agent-failure-containment-and-recovery.md
+++ b/docs/architecture/agent-failure-containment-and-recovery.md
@@ -197,8 +197,7 @@ contracts; passing them does not establish live model correction, arbitrary
 remote exactly-once effects or production recovery time. Diagnostics deployment,
 overhead and long-running non-interference need separate evidence.
 
-The RFCs specify three remaining integration gaps in these partially implemented
-paths:
+The RFCs track three integration paths with separate implementation checkpoints:
 
 1. [Current-use checks for explicit delegation ancestry](rfcs/shared-goal-alignment-and-governed-amendment-v0.md#38-invalid-evidence-and-affected-consumers):
    refuse new dependent use when its declared basis is unavailable, preserving
diff --git a/docs/architecture/agent-failure-containment-and-recovery.zh-CN.md b/docs/architecture/agent-failure-containment-and-recovery.zh-CN.md
index 26c4017824..fabc6a84c2 100644
--- a/docs/architecture/agent-failure-containment-and-recovery.zh-CN.md
+++ b/docs/architecture/agent-failure-containment-and-recovery.zh-CN.md
@@ -88,7 +88,7 @@ Note 能保留尝试过的路径与下一步，但 summary 仍由发送者提供
 
 这些测试设计包含失败、陈旧来源、输入改变和 effect 中断等负例。Fixture provider 与一次性本地状态验证各自合同；通过不等于真实模型纠错、任意远端 exactly-once 或生产恢复时间合格。诊断部署、开销与长时间 non-interference 需要独立证据。
 
-RFC 明确了这些部分实现路径中仍需补齐的三项整合缺口：
+RFC 将以下三项整合路径分别记录为独立 checkpoint：
 
 1. [显式委派来源链的当前使用检查](rfcs/shared-goal-alignment-and-governed-amendment-v0.zh-CN.md#38-失效证据与受影响消费者)：声明依据不可用时拒绝新的依赖使用，保留历史完成，限制遍历与验证成本。
 2. [原任务修复与复验](rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md#基于源码的实现顺序)：把独立失败依据传入可操作恢复路径，按当前 criterion 复验，再对账原效果。
diff --git a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
index 2b68db8414..1da8ad873e 100644
--- a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
+++ b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
@@ -304,15 +304,13 @@ lease/GoalRef fence, followed by M3's actual delivery callers.
 
 ### Source-grounded implementation sequence
 
-These are unimplemented composition deltas within existing M2/M3 and roadmap
-R2/R3/R4, not new milestones. Source owners were rechecked at
-`44931b6d22a50b949d43354e6ea498fb6b68d231`.
+The table tracks independent composition boundaries within M2/M3 and roadmap R2/R3/R4. Candidate checkpoints do not establish installed behavior or full M2 acceptance. The source baseline is `233cc76fd22760947d73e1501032b8b77e28148b`.
 
-| Bounded outcome | Existing entry and owner | Current gap and decisive exit |
+| Bounded outcome | Existing entry and owner | Current checkpoint and decisive exit |
 | --- | --- | --- |
 | Unavailable declared ancestry prevents new dependent use | `Delegations._read_current/start`, `delegation_results.require_dependencies/adoption_evidence`, `delegation.ts` | Alignment §3.8: source → A → B with A input invalidated must refuse current dependent use even if A output is unchanged. Qualify real read/start/adopt/settlement and packaged evidence readback; keep historical completion. |
 | Independent check failure reaches an actionable original-task recovery journey | `executor._task_validation_stage`, `ValidatedTurnReceipt`, canonical `turn_loop_controller_contract_v0.json`, `turn_journal.ts` | Preserve qualified failure scope and repair/replan detail; complete bounded repair or verifier-only retry, current validation and original-effect settlement. Host assertions cannot become trusted validation. |
-| Optional semantic review explains its evidence and coverage | Existing progress-review receipt/loader and canonical Goal acceptance inspection | Bind a selected criterion and evidence coverage in shadow readback; show missing/stale basis and separate judgment dimensions. Model quality and intervention qualify separately. |
+| Shadow review explains criteria and coverage | progress-review receipt/context, `progress_review_evidence.ts`, canonical acceptance inspect | Select current canonical task criteria; changed task bindings withdraw old judgments. Manual studies and legacy receipts retain distinct strength. Show separate dimensions, declared-file net change, missing evidence and unreadable storage. Default off and existing assist trigger rules retain their semantics. |
 
 For the second outcome, `_task_validation_stage` already saves the independent
 result and blocks settlement. `ValidatedTurnReceipt` omits that validation's
@@ -345,6 +343,15 @@ complete the journey. Bound repeated verification and source-chain traversal
 using measured workloads. Roll back code through its owner while retaining
 receipts, completed effects and unresolved recovery obligations.
 
+An optional `acceptance_scope` refers only to registry/runtime paths inside the
+selected workspace. Criteria come from the current owner; private commands and
+paths stay out of model questions. Hashes, versions and declared coverage are
+observations, not proof of model correctness, a durable business checkpoint or
+whole-task completion. Existing team evidence and capability settings carry App
+readback, and CLI/MCP share the owner; no Lark-specific protocol is introduced.
+Live correction, remote exactly-once, takeover, observer quality and intervention
+retain their existing RFC acceptance owners.
+
 ## 12. Open decisions
 
 The first M2 implementation must choose its smallest sufficient exploration
diff --git a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
index 95bf06b563..972be4f5b7 100644
--- a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
+++ b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
@@ -249,14 +249,13 @@ PostgreSQL authority、successor 调度和 App/Lark 送达尚未覆盖。因此
 
 ### 基于源码的实现顺序
 
-以下是既有 M2/M3 与路线图 R2/R3/R4 中尚未实现的组合增量，不新增里程碑。
-Source owner 在 `44931b6d22a50b949d43354e6ea498fb6b68d231` 复核。
+以下表格记录既有 M2/M3 与路线图 R2/R3/R4 的独立组合边界；候选 checkpoint 不表示已安装能力或完整 M2 验收。源码基线为 `233cc76fd22760947d73e1501032b8b77e28148b`。
 
-| 有界结果 | 既有入口与 owner | 当前缺口与决定性出口 |
+| 有界结果 | 既有入口与 owner | 当前 checkpoint 与决定性出口 |
 | --- | --- | --- |
 | 声明来源链不可用时，阻止新的依赖使用 | `Delegations._read_current/start`、`delegation_results.require_dependencies/adoption_evidence`、`delegation.ts` | 遵循对齐 §3.8：source → A → B 中 A input 失效，即使 A output 不变也须拒绝当前依赖使用；验证真实 read/start/adopt/settlement 和打包证据读回，保留历史完成。 |
 | 独立检查失败进入原任务可操作的恢复旅程 | `executor._task_validation_stage`、`ValidatedTurnReceipt`、canonical `turn_loop_controller_contract_v0.json`、`turn_journal.ts` | 传递已资格化失败范围与 repair/replan 细节；完成有界修复或只运行 verifier 的重试、当前验证与原效果结算。Host 声明不能冒充可信验证。 |
-| 可选语义审查说明证据和覆盖范围 | 既有 progress-review receipt/loader 与 canonical Goal acceptance inspect | 在 shadow 读回绑定选中 criterion 和证据覆盖；显示缺失/陈旧依据及独立判断维度。模型质量和干预另行资格化。 |
+| shadow 说明条款及观察范围 | progress-review receipt/context、`progress_review_evidence.ts`、canonical acceptance inspect | 可选择任务当前规范 criterion；任务关联变化后撤回旧判断。手工研究和旧回执保持不同证据强度；显示独立维度、净文件变化覆盖、缺失及存储未知。默认 off 和 assist 既有触发规则保持不变。 |
 
 第二项中，`_task_validation_stage` 已保存独立结果并阻断结算；`ValidatedTurnReceipt`
 未携带该验证的 `recovery_kind`，canonical controller 有意将 legacy
@@ -280,6 +279,13 @@ result，只重跑 validation，不再次调用 Host。**实际修复工作**需
 复制命令按钮或后端 receipt 不能独自完成旅程。用实际测量限制重复验证和来源链遍历
 成本；经既有 owner 回退代码，同时保留 receipt、已提交效果和未解决恢复义务。
 
+规范审查 basis 的 `acceptance_scope` 只引用选定 workspace 内的 registry/runtime，
+条款内容从当前 owner 读取，原命令及私有路径不进入模型问题。criterion hash、
+版本和覆盖是来源观察，不证明模型答案正确、完整业务 checkpoint 或全部任务完成。
+前端沿既有团队证据和能力设置入口读回；受影响 CLI/MCP 共用 owner，未新增
+Lark 专有协议。真实模型纠错、远端 exactly-once、接管、observer 质量与受控干预
+仍由原 RFC 验收所有者负责。
+
 ## 12. 未决事项
 
 首个 M2 实现由测试与领域维护者选择足够小的探索方法及边界。优先复用 fixture
diff --git a/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md b/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md
index 3d4af87063..2f721a9d93 100644
--- a/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md
+++ b/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.md
@@ -238,24 +238,29 @@ delay, false intervention against legitimate work and missed failures. Sample
 cases that never generated an alert as well as alerted cases; alert-only review
 cannot estimate missed failures.
 
-Source recheck at `44931b6d22a50b949d43354e6ea498fb6b68d231`:
-the optional `packages/loopx-jev` capture reads a declared file set and its net
-change; it does not observe every intermediate assertion. Core
-[`external_progress_review_trigger`](../../../loopx/control_plane/work_items/external_progress_review.py)
-consumes attributable receipts at a pinned assessment revision. In `assist`, a
-qualified streak can open the existing replan obligation; a newer completed
-non-triggering verdict can end that streak. This is the current signal policy,
-not a business-correctness certificate. The
-[`progress-review` contract](../../../loopx/capabilities/progress_review/README.md)
-and its owner retain off/shadow/assist compatibility. Any change to the combined
-signal or its clearing policy needs explicit disclosure and counterexamples;
-this design does not silently change it.
-
-The downstream contract is [evidence invalidation in shared work](shared-goal-alignment-and-governed-amendment-v0.md#38-invalid-evidence-and-affected-consumers)
+The local candidate now binds an explicitly selected canonical task criterion
+through [`progress_review_evidence.ts`](../../../loopx/control_plane/work_items/progress_review_evidence.ts).
+The optional observer's `acceptance_scope` reads the existing acceptance owner;
+private registry/runtime references stay outside the model packet. A manual
+basis remains an `operator_study`. Receipts declare selected-file net-change
+coverage. Capture and core receipt decoding require the canonical Goal/Agent/Todo
+scope to match the actual run; mismatches never reach model dispatch or drift
+consumption. Core readback also rechecks bindings against current criteria. A changed or unavailable basis withdraws that receipt's signal without
+rewriting its history. Legacy receipt handling and default-off isolation remain.
+
+The packaged capability editor separately displays relationship and increment,
+criterion origin, coverage, stale/missing/unavailable states and an explicit
+refresh. `off_goal` plus `new_evidence` does not become an on-Goal verdict; a
+non-triggering combined signal retains the existing streak behavior. File/SQLite
+canonical criterion edits and the production HTTP/frontend journey have local
+validation; no new live model-quality experiment or automatic intervention was
+performed. See the [capability contract](../../../loopx/capabilities/progress_review/README.md)
+and [observer guide](../../../packages/loopx-jev/DRIFT_SHADOW.md).
+
+These observations connect to [shared evidence invalidation](shared-goal-alignment-and-governed-amendment-v0.md#38-invalid-evidence-and-affected-consumers)
 and [recovery qualification](composable-state-machines-recovery-verification-v0.md#detection-containment-and-recovery-evidence).
-First connect existing verifier results and dependency readback to those owners.
-Additional model inference is conditional on the comparison in section 6; it is
-not a prerequisite for fixing a demonstrated deterministic gap.
+Model reasoning remains subject to the comparisons in section 6 and is not a
+prerequisite for repairing an evidenced deterministic gap.
 
 ### 5.3 Provider facts and deferred mechanisms
 
diff --git a/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.zh-CN.md b/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.zh-CN.md
index 286e7354d2..1c14b45a14 100644
--- a/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.zh-CN.md
+++ b/docs/architecture/rfcs/optional-semantic-assistance-jev-v0.zh-CN.md
@@ -224,19 +224,25 @@ D1 分别研究目标关系与证据增量。保留必要前置、合理阻塞
 行为。可选评估不可用时，必要的确定性检查仍生效。除检测延迟外，还要测量对合法
 工作的误干预与漏报。告警和从未告警的记录都要抽查；只审查告警记录无法估计漏报。
 
-源码在 `44931b6d22a50b949d43354e6ea498fb6b68d231` 复核：可选
-`packages/loopx-jev` capture 读取声明文件集合及净变化，不观察每条中间断言。
-核心 [`external_progress_review_trigger`](../../../loopx/control_plane/work_items/external_progress_review.py)
-消费身份可归属、评估版本已固定的回执。`assist` 可把合格连续信号转为既有 replan
-义务；较新的已完成、未触发 verdict 可以结束该连续信号。这是当前信号政策，不是
-业务正确证明。[progress-review 合同](../../../loopx/capabilities/progress_review/README.zh-CN.md)
-继续拥有 off/shadow/assist 兼容性。修改组合信号或解除政策须明确披露并提供反例，
-本设计不隐式改变当前行为。
-
-后续连接[共享工作中的证据失效](shared-goal-alignment-and-governed-amendment-v0.zh-CN.md#38-失效证据与受影响消费者)
+本地候选实现通过
+[`progress_review_evidence.ts`](../../../loopx/control_plane/work_items/progress_review_evidence.ts)
+绑定显式选中的规范任务 criterion。可选观察器的 `acceptance_scope` 读取既有验收
+owner；私有 registry/runtime 引用不进入模型请求。手写 basis 仍为 `operator_study`。
+回执声明选中文件的净变化覆盖；采集与核心回执解码共用 Goal/Agent/Todo 身份约束，
+不匹配时拒绝模型派发及漂移消费。核心读回也把新 canonical binding 与当前 criterion
+复核；依据变化或不可用时撤回该回执信号，不改历史。旧回执兼容性与默认关闭隔离
+保持原合同。
+
+打包能力编辑器分别展示目标关系与证据增量、criterion 来源、覆盖范围及过期／缺失／
+不可用状态，提供显式刷新。`off_goal` 加 `new_evidence` 不被改写为符合目标；组合
+信号未触发仍遵循原连续段政策。File/SQLite 下规范 criterion 编辑及真实 HTTP／界面
+流程已做本地验证；未开展新的真实模型质量实验或自动干预。见
+[能力合同](../../../loopx/capabilities/progress_review/README.zh-CN.md)与
+[观察器指南](../../../packages/loopx-jev/DRIFT_SHADOW.zh-CN.md)。
+
+这些观察衔接[共享工作中的证据失效](shared-goal-alignment-and-governed-amendment-v0.zh-CN.md#38-失效证据与受影响消费者)
 与[恢复资格](composable-state-machines-recovery-verification-v0.zh-CN.md#检测控制与恢复证据)。
-优先把现有 verifier 结果与依赖读回接到这些 owner；增加模型推理仍取决于第 6 节
-比较，不能成为修复已证实确定性缺口的前置条件。
+增加模型推理仍取决于第 6 节比较，不能成为修复已证实确定性缺口的前置条件。
 
 ### 5.3 提供方事实与延后机制
 
diff --git a/loopx/capabilities/progress_review/README.md b/loopx/capabilities/progress_review/README.md
index 0f2d6f5650..6d25063e3f 100644
--- a/loopx/capabilities/progress_review/README.md
+++ b/loopx/capabilities/progress_review/README.md
@@ -44,7 +44,7 @@ loopx configure-goal --goal-id <goal-id> --clear-progress-review-configuration -
 | `mode` | `off`, `shadow`, `assist` | `off` loads nothing; `shadow` records and displays; `assist` may raise the obligation |
 | `signal` | `noul`, `choice` | Which receipt judgment pair counts as drift |
 | `drift_threshold` | 2–20 | Consecutive completed drift receipts before an obligation |
-| `contract_revision` | sha256 or empty | The observer basis revision receipts must be bound to; printed by `loopx-jev drift init`. Required for `assist`. The pin is manual: receipts bound to other revisions are never counted, changing the basis does not retire earlier receipts by itself, and status reports `rebind_hint: newer_receipts_under_unpinned_revision` when the newest receipt is bound elsewhere |
+| `contract_revision` | sha256 or empty | The observer basis revision receipts must be bound to; printed by `loopx-jev drift init`. Required for `assist`. The pin is manual: receipts bound to other revisions are never counted, changing the basis does not retire legacy or operator-study receipts by itself, and status reports `rebind_hint: newer_receipts_under_unpinned_revision` when the newest receipt is bound elsewhere |
 
 The policy lives at `control_plane.progress_review` in the goal registry and is
 visible in `loopx configure-goal --goal-id <goal-id>` under `feature_summary`
@@ -65,7 +65,10 @@ Each receipt carries only typed fields:
 - `judgments.noul`: probabilities for `behavior_change`, `serves_acceptance`,
   `evidence_increment`, or null;
 - `drift_signal.noul` and `drift_signal.choice`: `true`, `false` or null;
-- `timing_ns`, `usage`, `label_probability_threshold`, `recorded_at`.
+- `timing_ns`, `usage`, `label_probability_threshold`, `recorded_at`;
+- optional `evidence_scope`: criterion origin/version/hash and declared relative
+  files, with coverage `declared_file_net_change`. New canonical bindings are
+  rechecked against current task acceptance before their signals are consumed.
 
 `sequence` is the writing observer's local counter and restarts at zero when
 `drift init` creates a new observer state. The core never compares sequences
@@ -105,7 +108,7 @@ This tightens the original stage-0 `assist` compatibility: incompletely bound
 historical receipts remain visible in shadow/status but cannot form or clear an
 obligation. Existing off behavior, model requests, manual revision pins and typed
 replan outcomes are unchanged. Every captured transition is one of three things: **drift**
-(completed, selected signal `true`, pinned revision), **on-goal** (completed,
+(completed, selected signal `true`, pinned revision), **condition not met** (completed,
 signal `false`) or **unevaluated** for one typed reason (`pending`, `failed`,
 `abstained`, `stale`, `undecided`, `missing`, `unattributed`,
 `identity_conflict`, `other_revision`, `not_evaluated`). Formation is
@@ -113,12 +116,30 @@ conservative: an obligation needs `drift_threshold` consecutive drift
 transitions with no unevaluated transition between them. Persistence is not:
 once formed, newer unevaluated transitions neither extend nor dissolve the
 obligation, and their count is reported as `unevaluated_transitions` on the
-trigger. Only an acknowledged replan or a newer completed on-goal verdict ends
+trigger. Only an acknowledged replan or a newer completed non-triggering verdict ends
 it; the Goal owner can also set the mode back to `shadow` or `off`. Receipts
 bound to a revision other than the pinned one are unevaluated history and
 never counted. The scan covers the run history the Goal keeps
 (`latest_runs`), so an obligation can only be as old as that window.
 
+## Criterion and observation readback
+
+For canonical task criteria, initialize the optional observer with an explicit
+[`acceptance_scope`](../../../packages/loopx-jev/DRIFT_SHADOW.md#bind-canonical-task-criteria).
+The observer reads the existing acceptance owner instead of trusting the
+manifest's acceptance text. Manual bases remain labeled `operator_study`.
+A changed or unavailable canonical criterion binding makes that receipt stale
+and withdraws its signal; stored history and the policy's manual revision pin
+remain intact. Legacy receipts retain their existing behavior.
+
+In **Goal settings → capability center → progress-review sentinel**, enabled
+Goals show relationship and evidence increment separately, criterion origin,
+selected-file coverage and missing/stale/unavailable states. Use **重新读取审查证据**
+to refresh; the old judgment is withdrawn during the read. `off_goal` together
+with `new_evidence` can be a non-triggering signal without being on Goal.
+Receipt-store failure is unavailable, never cached success. Model scores do not
+validate a task or authorize stop, rollback or completion. `off` reads nothing.
+
 ## Discharge
 
 An `assist` obligation is discharged only the way every autonomous replan
diff --git a/loopx/capabilities/progress_review/README.zh-CN.md b/loopx/capabilities/progress_review/README.zh-CN.md
index 784384acbd..d340ffae42 100644
--- a/loopx/capabilities/progress_review/README.zh-CN.md
+++ b/loopx/capabilities/progress_review/README.zh-CN.md
@@ -35,7 +35,7 @@ loopx configure-goal --goal-id <goal-id> --clear-progress-review-configuration -
 | `mode` | `off`、`shadow`、`assist` | `off` 不加载任何内容；`shadow` 记录并展示；`assist` 可以触发义务 |
 | `signal` | `noul`、`choice` | 哪一组判断算作漂移 |
 | `drift_threshold` | 2–20 | 触发义务前需要的连续已完成漂移回执数 |
-| `contract_revision` | sha256 或空 | 回执必须绑定的观察器 basis 修订，由 `loopx-jev drift init` 打印；`assist` 必需。pin 是手动的：绑定到其他修订的回执永不计数，basis 变化本身不会让早先回执退休，最新回执绑定到别处时 status 报告 `rebind_hint: newer_receipts_under_unpinned_revision` |
+| `contract_revision` | sha256 或空 | 回执必须绑定的观察器 basis 修订，由 `loopx-jev drift init` 打印；`assist` 必需。pin 是手动的：绑定到其他修订的回执永不计数，basis 变化本身不会让旧格式或 operator-study 回执退休，最新回执绑定到别处时 status 报告 `rebind_hint: newer_receipts_under_unpinned_revision` |
 
 策略保存在 Goal 注册表的 `control_plane.progress_review`，可在 `loopx configure-goal --goal-id <goal-id>` 输出的 `feature_summary` 和 Dashboard 能力编辑器中看到。格式错误的配置块会安全地退回 `off`。
 
@@ -48,7 +48,9 @@ loopx configure-goal --goal-id <goal-id> --clear-progress-review-configuration -
 - `judgments.choice`：`relation` 与 `increment` 标签或 null；
 - `judgments.noul`：`behavior_change`、`serves_acceptance`、`evidence_increment` 的概率或 null；
 - `drift_signal.noul`、`drift_signal.choice`：`true`、`false` 或 null；
-- `timing_ns`、`usage`、`label_probability_threshold`、`recorded_at`。
+- `timing_ns`、`usage`、`label_probability_threshold`、`recorded_at`；
+- 可选 `evidence_scope`：criterion 来源／版本／摘要、声明的相对文件及
+  `declared_file_net_change` 覆盖。新 canonical binding 的信号被消费前会与当前任务验收复核。
 
 `sequence` 是写回执的观察器的本地计数，`drift init` 创建新的观察器状态时会从 0 重新计数。核心从不跨观察器比较 sequence：加载、加载上限、最新修订判断和同一转换两条回执的合并，都按 run 的 `generated_at`、再 `recorded_at`、再 `sequence` 排序（`progress_review_receipt_order_key`）。
 
@@ -63,7 +65,21 @@ loopx configure-goal --goal-id <goal-id> --clear-progress-review-configuration -
 
 这收紧了原 stage-0 `assist` 的兼容行为：身份不完整的历史回执仍可在 shadow/status 查看，但不能形成或解除义务。off 行为、模型请求、手动 revision pin 与类型化 replan 出口保持原契约。
 
-每个被捕获的转换只有三种：**漂移**（completed、所选信号为 `true`、pin 修订）、**on-goal**（completed、信号为 `false`）或**未评估**，后者带一个类型化原因（`pending`、`failed`、`abstained`、`stale`、`undecided`、`missing`、`unattributed`、`identity_conflict`、`other_revision`、`not_evaluated`）。形成是保守的：义务需要 `drift_threshold` 个连续漂移转换，中间不能有未评估转换。存续则不然：一旦形成，更新的未评估转换既不延长也不解除义务，其数量作为 `unevaluated_transitions` 记在 trigger 上。只有已确认的重规划或更新的 completed on-goal 判定能结束它；Goal 负责人也可以把模式调回 `shadow` 或 `off`。绑定到非 pin 修订的回执是未评估的历史，永不计数。扫描范围是 Goal 保留的 run 历史（`latest_runs`），义务最多只能与该窗口同样老。
+每个被捕获的转换只有三种：**漂移**（completed、所选信号为 `true`、pin 修订）、**干预条件不成立**（completed、信号为 `false`）或**未评估**，后者带一个类型化原因（`pending`、`failed`、`abstained`、`stale`、`undecided`、`missing`、`unattributed`、`identity_conflict`、`other_revision`、`not_evaluated`）。形成是保守的：义务需要 `drift_threshold` 个连续漂移转换，中间不能有未评估转换。存续则不然：一旦形成，更新的未评估转换既不延长也不解除义务，其数量作为 `unevaluated_transitions` 记在 trigger 上。只有已确认的重规划或更新的 completed 未触发判定能结束它；Goal 负责人也可以把模式调回 `shadow` 或 `off`。绑定到非 pin 修订的回执是未评估的历史，永不计数。扫描范围是 Goal 保留的 run 历史（`latest_runs`），义务最多只能与该窗口同样老。
+
+## Criterion 与观察读回
+
+规范任务 criterion 通过观察器的显式
+[`acceptance_scope`](../../../packages/loopx-jev/DRIFT_SHADOW.zh-CN.md#绑定规范任务-criterion)
+绑定：读取既有验收 owner，不信任 manifest 自述的 acceptance。手写 basis 仍标为
+`operator_study`。规范 criterion binding 变化或不可用时，该回执投影为过期并撤回信号；
+保留存储历史与政策的手动 revision pin，旧格式回执沿用原行为。
+
+在 **Goal 设置 → 能力中心 → 进展评估哨兵** 中，已启用的 Goal 分别展示目标关系与
+证据增量、criterion 来源、选中文件覆盖及缺失／过期／不可用状态。点
+**重新读取审查证据** 刷新，读取期间撤回旧判断。`off_goal` 与 `new_evidence` 可以
+同时成立且组合信号未触发，不能据此称为符合目标。回执存储读取失败保持不可用，
+不沿用缓存成功。模型分数不构成任务验收，也不授予停止、回滚或完成权限；`off` 不读取。
 
 ## 解除
 
diff --git a/packages/loopx-jev/DRIFT_SHADOW.md b/packages/loopx-jev/DRIFT_SHADOW.md
index c100c0c859..0fdcd039e0 100644
--- a/packages/loopx-jev/DRIFT_SHADOW.md
+++ b/packages/loopx-jev/DRIFT_SHADOW.md
@@ -47,7 +47,7 @@ collector does not discover relevant files or scan the entire repository.
 
 | Material | How it enters the assessment |
 | --- | --- |
-| Goal and acceptance criteria | Operator-provided basis JSON |
+| Goal and acceptance criteria | Operator-provided basis; with explicit `acceptance_scope`, current canonical task criteria |
 | Files named by `--path` | Before/after contents and net changes; a not-yet-created file is allowed |
 | Optional `evidence` references in the basis | Explicitly named regular files, such as a test or probe report |
 | Other source, dependencies or conversation history | Not automatically read; its absence limits the judgment |
@@ -64,6 +64,40 @@ the test that exercises it. A judgment from that packet cannot certify the full
 behavior. Include relevant tests, results and dependencies deliberately; if the
 necessary material does not fit, do not present the partial packet as complete.
 
+## Bind canonical task criteria
+
+The basis may include this private, local scope in addition to its existing
+`goal_id`, `objective` and nonempty `acceptance` fields:
+
+```json
+"acceptance_scope": {
+  "registry_ref": ".loopx/registry.json",
+  "runtime_ref": ".loopx/runtime",
+  "agent_id": "reviewer",
+  "todo_id": "todo_review",
+  "criterion_ids": ["criterion_review"]
+}
+```
+
+Use real owner paths and criterion IDs for that task. Both references must stay
+inside the selected workspace; escapes and symlinks are rejected. The examples
+are local operator configuration, not files to publish. The observer reads the
+current acceptance owner and replaces the model's acceptance list with the
+selected canonical descriptions. Private registry/runtime references are not
+sent to the model or persisted in public receipt scope. Without this block the
+basis is explicitly an operator study, not canonical acceptance.
+
+The binding pins Goal/Agent/Todo identity, contract revision/digest, task semantics
+and selected criteria. Capture rejects another task or agent before model dispatch;
+core receipt decoding applies the same identity rule. Select a new matching basis
+when changing tasks.
+A change invalidates in-flight assessment; a fresh observation needs a new basis
+revision/baseline through the existing initialization path. Core readback also
+withdraws a newly scoped receipt when its canonical binding is no longer current.
+The Goal's `assist` revision pin stays manual. Declared file net changes do not
+cover every intermediate decision or omitted dependency, and binding does not
+certify model correctness or grant mutation authority.
+
 ## Run it
 
 Use Python 3.11+ and the Node runtime required by the LoopX checkout. This
diff --git a/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md b/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
index cef0cfeb0a..02a51391d1 100644
--- a/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
+++ b/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
@@ -27,6 +27,33 @@
 
 例如只选择函数所在文件，而漏掉被调用的 helper 和对应测试，评估就不能证明完整行为。应主动纳入相关测试、结果和依赖；必要材料装不下时，不能把部分材料包装成完整证据。
 
+## 绑定规范任务 criterion
+
+在既有 `goal_id`、`objective` 和非空 `acceptance` 字段之外，basis 可以包含这个
+私有本机范围：
+
+```json
+"acceptance_scope": {
+  "registry_ref": ".loopx/registry.json",
+  "runtime_ref": ".loopx/runtime",
+  "agent_id": "reviewer",
+  "todo_id": "todo_review",
+  "criterion_ids": ["criterion_review"]
+}
+```
+
+替换为该任务真实 owner 路径与 criterion ID。两个引用必须位于选中 workspace 内，
+拒绝越界与符号链接；这些是本地操作者配置，不应发布。观察器读取当前验收 owner，
+用选中的规范描述替换模型 acceptance 列表。私有 registry/runtime 引用不发送给模型，
+也不进入公开回执范围。没有该块时，basis 明确属于 operator study，不是规范验收。
+
+binding 固定 Goal/Agent/Todo 身份、contract 修订／摘要、任务语义及所选 criterion。
+采集在模型派发前拒绝错配任务或 Agent，核心回执解码使用同一规则；切换任务须选用
+匹配的新依据。变更使在途评估失效；
+新的观察沿既有初始化入口重建 basis revision／baseline。核心读回也会在新格式回执的
+规范 binding 不再当前时撤回信号。Goal 的 `assist` revision pin 仍是手动的。
+声明文件净变化不覆盖每条中间判断或遗漏依赖；绑定不证明模型正确，也不授予修改权限。
+
 ## 操作方法
 
 使用 Python 3.11+ 和 LoopX 检出要求的 Node 运行时。采集器目前面向 Linux/macOS 的 POSIX 文件处理；Windows 采集未验证，缺少所需文件原语时记录观察不可用。此可选发行包不进入 LoopX 默认 wheel；从源码根目录在新环境中安装，再执行不需要 key 的集成测试：

From 6001cadde7f2eb36fb1b2beb227e5741ea03aa16 Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 10:58:58 +0800
Subject: [PATCH 3/6] tests: qualify criterion-bound review

Signed-off-by: song <liusongstep@gmail.com>
---
 .../smoke/progress-review-evidence-smoke.mjs  |  69 ++++++++++++
 tests/capabilities/test_progress_review.py    |  21 ++++
 .../test_external_progress_review.py          |   2 +-
 .../progress_review_evidence.test.ts          |  28 +++++
 .../progress_review_evidence_fixture.py       |  93 ++++++++++++++++
 .../test_progress_review_criterion_binding.py | 104 ++++++++++++++++++
 6 files changed, 316 insertions(+), 1 deletion(-)
 create mode 100644 apps/presentation/dashboard/smoke/progress-review-evidence-smoke.mjs
 create mode 100644 tests/control_plane_ts/progress_review_evidence.test.ts
 create mode 100644 tests/presentation/progress_review_evidence_fixture.py
 create mode 100644 tests/test_progress_review_criterion_binding.py

diff --git a/apps/presentation/dashboard/smoke/progress-review-evidence-smoke.mjs b/apps/presentation/dashboard/smoke/progress-review-evidence-smoke.mjs
new file mode 100644
index 0000000000..4b409bba31
--- /dev/null
+++ b/apps/presentation/dashboard/smoke/progress-review-evidence-smoke.mjs
@@ -0,0 +1,69 @@
+// Packaged reader + production Chat HTTP, delegation workers and independent checks.
+import assert from "node:assert/strict";
+import {spawn} from "node:child_process";
+import {once} from "node:events";
+import {createInterface} from "node:readline";
+import {mkdir} from "node:fs/promises";
+import {resolve} from "node:path";
+import {resolveTestPython} from "../../../../scripts/test-python.mjs";
+import {launchBrowser, loadPlaywright, waitForHttp} from "../../../../examples/dashboard-browser-smoke-support.mjs";
+process.env.LOOPX_PERSONAL_WORKSPACE_PACKAGED = "1";
+const {repoRoot, outputDir, port, startServer} = await import("../../../../examples/personal-workspace-browser/fixture.mjs");
+const {openWorkspacePage} = await import("../../../../examples/personal-workspace-browser/scenario-context.mjs");
+const fixture = spawn(resolveTestPython({repoRoot}), ["-u", "tests/presentation/progress_review_evidence_fixture.py"], {cwd: repoRoot, stdio: ["pipe", "pipe", "pipe"]});
+const exited = once(fixture, "exit");
+const lines = createInterface({input: fixture.stdout});
+const iterator = lines[Symbol.asyncIterator]();
+let stderr = "";
+fixture.stderr.on("data", chunk => {stderr += chunk;});
+async function next() {const line = await iterator.next(); assert.equal(line.done, false, stderr); return JSON.parse(line.value);}
+async function command(value) {fixture.stdin.write(`${value}\n`); return next();}
+let server, browser, workspace;
+try {
+  const {origin, goal_id} = await next();
+  server = await startServer();
+  const url = `http://127.0.0.1:${port}/chat/?statusUrl=/status.json`;
+  await waitForHttp(url);
+  browser = await launchBrowser(loadPlaywright().chromium);
+  workspace = await openWorkspacePage({newPage: options => browser.newPage({locale: "zh-CN", ...options})}, url);
+  const {page, api} = workspace;
+  await page.locator(".personal-goal-link", {hasText: "Product Release"}).click();
+  await mkdir(outputDir, {recursive: true});
+  await page.setViewportSize({width: 390, height: 844});
+  await page.route("**/api/chat/goal-configuration?*", async route => {
+    const response = await page.request.get(`${origin}/api/chat/goal-configuration?goal_id=${encodeURIComponent(goal_id)}`);
+    return route.fulfill({response});
+  });
+  await page.getByRole("button", {name: "Goal 设置", exact: true}).click();
+  await page.getByRole("button", {name: "能力中心", exact: true}).click();
+  await page.getByRole("button", {name: /进展评估哨兵/}).click();
+  const review = page.getByRole("region", {name: "审查证据读回"});
+  await review.getByText("绑定任务的规范验收条款", {exact: true}).waitFor();
+  await review.getByText("与目标的关系: off_goal", {exact: true}).waitFor();
+  await review.getByText("新证据增量: new_evidence", {exact: true}).waitFor();
+  await review.getByText("仅观察声明文件在两个检查点之间的净变化", {exact: true}).waitFor();
+  const coverage = review.getByText("仅观察声明文件在两个检查点之间的净变化", {exact: true});
+  await coverage.scrollIntoViewIfNeeded();
+  await review.getByText("观察范围与条款版本", {exact: true}).click();
+  await review.locator("code").scrollIntoViewIfNeeded();
+  await page.screenshot({path: resolve(outputDir, "progress-review-mixed-dimensions-mobile.png")});
+  assert.equal(await page.evaluate(() => document.documentElement.scrollWidth > window.innerWidth), false);
+  await page.setViewportSize({width: 1280, height: 960});
+  await review.scrollIntoViewIfNeeded();
+  await page.screenshot({path: resolve(outputDir, "progress-review-mixed-dimensions-desktop.png")});
+  await command("review-task-edit");
+  await review.getByRole("button", {name: "重新读取审查证据", exact: true}).click();
+  await review.getByText("条款或任务关联已变化，旧判断不可作为当前依据。", {exact: true}).waitFor();
+  await review.getByText("最新审查: stale", {exact: true}).waitFor();
+  await command("review-missing");
+  await review.getByRole("button", {name: "重新读取审查证据", exact: true}).click();
+  await review.getByText("尚无审查回执。", {exact: true}).waitFor();
+  assert.equal(await review.getByText("与目标的关系: off_goal", {exact: true}).count(), 0);
+  console.log("PASS: packaged scoped shadow readback, separate dimensions, changed-task withdrawal and missing evidence via production HTTP");
+} catch (error) {
+  if (workspace) console.error({body: (await workspace.page.locator("body").innerText()).slice(-3000), errors: workspace.errors});
+  throw error;
+} finally {
+  await workspace?.close(); await browser?.close(); server?.kill("SIGTERM");
+  fixture.stdin.end(); const [code] = await exited; lines.close(); assert.equal(code, 0, stderr);
+}
diff --git a/tests/capabilities/test_progress_review.py b/tests/capabilities/test_progress_review.py
index 9468e986cc..3b12a86f96 100644
--- a/tests/capabilities/test_progress_review.py
+++ b/tests/capabilities/test_progress_review.py
@@ -362,3 +362,24 @@ def test_cli_flags_reach_configure_goal(tmp_path: Path) -> None:
         "drift_threshold": 5,
         "contract_revision": None,
     }
+
+
+def test_readback_separates_dimensions_and_keeps_missing_and_io_unknown(tmp_path, monkeypatch):
+    from loopx.capabilities.progress_review.context import external_progress_review_context
+    from loopx.capabilities.progress_review import receipt as receipt_module
+    goal = {"id": GOAL_ID, "control_plane": {"progress_review": {"mode": "shadow"}}}
+    assert external_progress_review_context({}, tmp_path) is None
+    assert external_progress_review_context(goal, tmp_path)["summary"]["read_state"] == "missing"
+    mixed = receipt(judgments={"choice": {"relation": "off_goal", "increment": "new_evidence"}, "noul": {
+        "behavior_change": 0.9, "serves_acceptance": 0.1, "evidence_increment": 0.9}},
+        drift_signal={"noul": False, "choice": False})
+    write_progress_review_receipt(tmp_path, GOAL_ID, mixed)
+    latest = external_progress_review_context(goal, tmp_path)["summary"]["latest"]
+    assert latest["judgments"]["choice"] == {"relation": "off_goal", "increment": "new_evidence"}
+    def unreadable(*args, **kwargs):
+        raise OSError("disposable storage unavailable")
+    monkeypatch.setattr(receipt_module, "load_progress_review_receipts", unreadable)
+    summary = external_progress_review_context(goal, tmp_path)["summary"]
+    assert summary["read_state"] == "unavailable"
+    assert summary["read_error"] == "receipt_store_unavailable"
+    assert summary["latest"] is None
diff --git a/tests/control_plane/test_external_progress_review.py b/tests/control_plane/test_external_progress_review.py
index 2c1bd14fed..bfffd12c87 100644
--- a/tests/control_plane/test_external_progress_review.py
+++ b/tests/control_plane/test_external_progress_review.py
@@ -369,7 +369,7 @@ def test_newer_drift_after_formation_extends_and_rebinds_the_baseline() -> None:
     assert result["oldest_counted_generated_at"] == "2026-09-21T00:00:01Z"
 
 
-def test_a_newer_completed_on_goal_verdict_ends_the_open_condition() -> None:
+def test_a_newer_completed_nontriggering_signal_ends_the_open_condition() -> None:
     """A positive verdict on a newer transition is evidence; a missing one is not."""
 
     runs = [run(3, turn="t3"), run(2, turn="t2"), run(1, turn="t1")]
diff --git a/tests/control_plane_ts/progress_review_evidence.test.ts b/tests/control_plane_ts/progress_review_evidence.test.ts
new file mode 100644
index 0000000000..02fb5e1b98
--- /dev/null
+++ b/tests/control_plane_ts/progress_review_evidence.test.ts
@@ -0,0 +1,28 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import {progressReviewCriterionBasis, progressReviewEvidenceScope} from "../../loopx/control_plane/work_items/progress_review_evidence.ts";
+const identity = {goal_id: "goal", agent_id: "agent"};
+const run = {agent_id: "agent", todo_id: "task"};
+const requirements = {todo_id: "task", contract_revision: 1, contract_digest: "a".repeat(64),
+  todo_semantic_digest: "b".repeat(64), criteria: [{id: "check", description: "Prove the declared postcondition"}]};
+test("canonical criteria come from the task's current scope, not operator assertions", () => {
+  const result = progressReviewCriterionBasis({...identity, requirements, criterion_ids: ["check"], acceptance: ["Claim success"]});
+  assert.deepEqual(result.acceptance, ["Prove the declared postcondition"]);
+  assert.equal((result.binding as any).origin, "goal_acceptance");
+  assert.throws(() => progressReviewCriterionBasis({...identity, requirements, criterion_ids: ["another"]}));
+  assert.throws(() => progressReviewCriterionBasis({...identity, requirements, criterion_ids: ["check", "check"]}));
+  const manual = progressReviewCriterionBasis({requirements: null, acceptance: ["Study this behavior"]});
+  assert.equal((manual.binding as any).origin, "operator_study");
+  const scope = {criterion_binding: result.binding, coverage: "declared_file_net_change", files: ["code.ts"]};
+  assert.deepEqual(progressReviewEvidenceScope({...identity, run, scope}).scope, scope);
+  assert.throws(() => progressReviewEvidenceScope({...identity, run, scope: {...scope, files: ["../outside"]}}));
+  assert.throws(() => progressReviewEvidenceScope({...identity, run, scope: {...scope, coverage: "all_steps"}}));
+});
+
+test("canonical criteria cannot be used for a different goal, agent or task", () => {
+  const basis = progressReviewCriterionBasis({...identity, requirements, criterion_ids: ["check"]});
+  const scope = {criterion_binding: basis.binding, coverage: "declared_file_net_change", files: []};
+  for (const changed of [{goal_id: "other", run}, {goal_id: "goal", run: {...run, todo_id: "other"}},
+    {goal_id: "goal", run: {...run, agent_id: "other"}}, {goal_id: "goal", run: {}}])
+    assert.throws(() => progressReviewEvidenceScope({...changed, scope}));
+});
diff --git a/tests/presentation/progress_review_evidence_fixture.py b/tests/presentation/progress_review_evidence_fixture.py
new file mode 100644
index 0000000000..779b075d9c
--- /dev/null
+++ b/tests/presentation/progress_review_evidence_fixture.py
@@ -0,0 +1,93 @@
+"""Disposable production delegation + Chat HTTP for packaged recovery validation."""
+from pathlib import Path
+import json
+import sys
+import tempfile
+import threading
+from types import SimpleNamespace
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[2]))
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+import pytest
+from test_local_delegation import service
+from loopx.chat_runtime import ChatRuntimeController
+from loopx.chat_server import ChatHTTPServer, ChatRequestHandler
+from loopx.chat_store import ChatSessionStore
+
+
+def main():
+    cleanups = []
+    patch = pytest.MonkeyPatch()
+    with tempfile.TemporaryDirectory(prefix="loopx-result-use-") as directory:
+        try:
+            root, runner = service.__wrapped__(Path(directory),
+                SimpleNamespace(param="file", addfinalizer=cleanups.append), patch)
+            config = root / "project/.loopx/config/delegations.json"
+            config.parent.mkdir(parents=True, exist_ok=True)
+            catalog = json.loads(runner.config.read_text())
+            config.write_text(json.dumps(catalog))
+            runner.config = config
+            registry = json.loads(runner.registry.read_text())
+            goal = next(row for row in registry["goals"] if row["id"] == runner.goal_id)
+            goal.setdefault("control_plane", {})["progress_review"] = {"mode": "shadow"}
+            runner.registry.write_text(json.dumps(registry))
+            from tests.capabilities.test_progress_review import receipt
+            from loopx.capabilities.progress_review.receipt import write_progress_review_receipt
+            from loopx.control_plane.effect_runtime import effect_runtime_result
+            from loopx.control_plane.goals.acceptance import inspect_goal_acceptance
+            requirements = inspect_goal_acceptance(registry_path=runner.registry, runtime_root=str(runner.root),
+                goal_id=runner.goal_id, agent_id="analyst", todo_id="todo_analyst-initial")["completion_requirements"]
+            binding = effect_runtime_result("progress_review.criterion_basis", {
+                "requirements": requirements, "acceptance": [], "criterion_ids": ["analyst-initial"],
+                "goal_id": runner.goal_id, "agent_id": "analyst",
+            })["binding"]
+            review = receipt(goal_id=runner.goal_id,
+                judgments={"choice": {"relation": "off_goal", "increment": "new_evidence"}, "noul": {
+                    "behavior_change": 0.9, "serves_acceptance": 0.1, "evidence_increment": 0.9}},
+                drift_signal={"noul": False, "choice": False},
+                evidence_scope={"criterion_binding": binding, "coverage": "declared_file_net_change", "files": ["code.ts"]})
+            review["run"].update(agent_id="analyst", todo_id="todo_analyst-initial")
+            review_path = write_progress_review_receipt(runner.root, runner.goal_id, review)
+            store = ChatSessionStore(runner.root)
+            controller = ChatRuntimeController(store=store, codex_bin="missing-codex", registry_path=runner.registry)
+            store.create_session(goal_id=runner.goal_id, agent_id="codex", channel_id=f"goal.{runner.goal_id}",
+                                 upstream_thread_id="fixture", session_id="recovery", adapter_kind="codex_app_server", upstream_mode="chat")
+            store.update_session("recovery", native_goal={"status": "active"}, loopx_mode={"settings": {
+                "agent_id": "lead", "execution_config": ".loopx/config/delegations.json"}})
+            server = ChatHTTPServer(("127.0.0.1", 0), ChatRequestHandler)
+            server.verbose = False
+            server.registry_path = runner.registry
+            server.runtime_root = runner.root
+            server.chat_store = store
+            server.runtime_controller = controller
+            thread = threading.Thread(target=server.serve_forever, daemon=True)
+            thread.start()
+            print(json.dumps({"origin": f"http://127.0.0.1:{server.server_port}", "goal_id": runner.goal_id}), flush=True)
+            try:
+                for command in sys.stdin:
+                    action = command.strip()
+                    if action == "review-missing":
+                        review_path.unlink()
+                    elif action == "review-task-edit":
+                        changed = runner._cli(runner.binding("analysis"), "todo", "update", "--goal-id", runner.goal_id,
+                            "--agent-id", "analyst", "--todo-id", "todo_analyst-initial", "--text", "Changed declared scope",
+                            "--update-operation-id", "review-task-scope-edit")
+                        assert changed["ok"] is True, changed
+                    elif action == "review-restore":
+                        write_progress_review_receipt(runner.root, runner.goal_id, review)
+                    else:
+                        raise ValueError("unknown fixture command")
+                    print(json.dumps({"ok": True}), flush=True)
+            finally:
+                server.shutdown()
+                server.server_close()
+                thread.join(timeout=5)
+                controller.close()
+        finally:
+            for cleanup in reversed(cleanups):
+                cleanup()
+            patch.undo()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/tests/test_progress_review_criterion_binding.py b/tests/test_progress_review_criterion_binding.py
new file mode 100644
index 0000000000..e8df697891
--- /dev/null
+++ b/tests/test_progress_review_criterion_binding.py
@@ -0,0 +1,104 @@
+"""Version-bound requester decisions through real Turn acceptance and durable IO."""
+import json
+import subprocess
+import sys
+
+import pytest
+
+from test_local_delegation import service as delegation_service
+
+service = delegation_service
+
+
+def test_shadow_basis_uses_current_canonical_criteria_and_withdraws_after_task_edit(service):
+    from pathlib import Path
+    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "packages/loopx-jev/src"))
+    from loopx_jev.runner import read_basis
+    root, runner = service
+    manifest = root / "basis.json"
+    manifest.write_text(json.dumps({"goal_id": runner.goal_id, "objective": "Review declared task change",
+        "acceptance": ["Operator claim must not replace the canonical criterion"], "evidence": [],
+        "acceptance_scope": {"registry_ref": "registry.json", "runtime_ref": "runtime", "agent_id": "analyst",
+            "todo_id": "todo_analyst-initial", "criterion_ids": ["analyst-initial"]}}))
+    basis, current = read_basis(manifest, root)
+    assert basis["criterion_binding"]["origin"] == "goal_acceptance"
+    assert basis["criterion_binding"]["criterion_ids"] == ["analyst-initial"]
+    assert basis["acceptance"] == ["Independent checks for analyst initial"]
+    assert current()
+    from tests.capabilities.test_progress_review import receipt
+    from loopx.capabilities.progress_review.receipt import write_progress_review_receipt
+    from loopx.capabilities.progress_review.context import external_progress_review_context
+    goal = {"id": runner.goal_id, "control_plane": {"progress_review": {"mode": "shadow"}}}
+    canonical_receipt = receipt(goal_id=runner.goal_id,
+        evidence_scope={"criterion_binding": basis["criterion_binding"], "coverage": "declared_file_net_change", "files": ["output.json"]})
+    canonical_receipt["run"].update(agent_id="analyst", todo_id="todo_analyst-initial")
+    for mismatch in ({"todo_id": "todo_reviewer-corrected"}, {"agent_id": "reviewer"}):
+        with pytest.raises(ValueError, match="evidence scope"):
+            write_progress_review_receipt(runner.root, runner.goal_id,
+                {**canonical_receipt, "run": {**canonical_receipt["run"], **mismatch}})
+    persisted = write_progress_review_receipt(runner.root, runner.goal_id, canonical_receipt)
+    persisted.write_text(json.dumps({**canonical_receipt, "run": {**canonical_receipt["run"], "todo_id": "todo_reviewer-corrected"}}))
+    rejected_context = external_progress_review_context(goal, runner.root)
+    assert rejected_context["receipts"] == []
+    write_progress_review_receipt(runner.root, runner.goal_id, canonical_receipt)
+    assert external_progress_review_context(goal, runner.root)["summary"]["latest"]["criterion_current"] is True
+    # Exercise the real consumer too: canonical revisions include the owner
+    # binding, so comparing only the unchanged manifest digest would reject all jobs.
+    from loopx_jev import drift
+    from loopx_jev.store import atomic_json
+    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "packages/loopx-jev/tests"))
+    from drift_fixtures import response
+    def git(*args):
+        subprocess.run(["git", "-C", str(root), *args], check=True, capture_output=True)
+    git("init", "-q")
+    git("config", "user.email", "fixture@example.invalid")
+    git("config", "user.name", "Fixture")
+    probe = root / "probe.txt"
+    probe.write_text("before\n")
+    git("add", "probe.txt")
+    git("commit", "-qm", "fixture baseline")
+    config = root / "shadow-config.json"
+    atomic_json(config, {"schema_version": "loopx_jev_drift_config_v0", "mode": "shadow",
+        "scenarios": ["progress_review"], "model": "fixture-v1", "allow_egress": True})
+    observer = root / "observer"
+    drift.initialize(observer, root, manifest, config, ["probe.txt"], runtime_root=runner.root)
+    def queue(sequence):
+        probe.write_text(f"changed {sequence}\n")
+        record = root / f"run-{sequence}.json"
+        atomic_json(record, {"goal_id": runner.goal_id, "generated_at": f"2026-01-01T00:00:0{sequence}Z",
+            "agent_id": "analyst", "todo_id": "todo_analyst-initial", "turn_instance_id": f"shadow-{sequence}"})
+        prepared = drift.prepare(observer, config)
+        original = json.loads(record.read_text())
+        for mismatch in ({"todo_id": "todo_reviewer-corrected"}, {"agent_id": "reviewer"}):
+            atomic_json(record, {**original, **mismatch})
+            with pytest.raises(ValueError, match="run_acceptance_scope_mismatch"):
+                drift.enqueue(observer, prepared, record)
+        atomic_json(record, original)
+        return drift.enqueue(observer, prepared, record)["event_id"]
+    calls = []
+    def send(request, *_):
+        calls.append(request)
+        assert "acceptance_scope" not in request["state"]["goal_basis"]
+        assert request["state"]["goal_basis"]["acceptance"] == ["Independent checks for analyst initial"]
+        return {"response": response(request, ["off_goal", "new_evidence"])}
+    event = queue(1)
+    drift.drain(observer, config, transport=send, credential=lambda: "fixture")
+    evaluated = next(row for row in drift.status(observer)["events"] if row["event_id"] == event)
+    assert evaluated["status"] == "completed", evaluated
+    assert len(calls) == 1
+    queue(2)
+    def edit_during_request(request, *args):
+        result = runner._cli(runner.binding("analysis"), "todo", "update", "--goal-id", runner.goal_id,
+            "--agent-id", "analyst", "--todo-id", "todo_analyst-initial", "--text", "Changed declared task scope",
+            "--update-operation-id", "shadow-scope-edit")
+        assert result["ok"] is True, result
+        return send(request, *args)
+    drift.drain(observer, config, transport=edit_during_request, credential=lambda: "fixture")
+    assert len(calls) == 2
+    assert all(row["status"] == "stale" for row in drift.status(observer)["events"] if row["event_id"] != event)
+    assert not current(), "A stale owner binding cannot be promoted by an unchanged study manifest"
+    latest = external_progress_review_context(goal, runner.root)["summary"]["latest"]
+    assert latest["status"] == "stale" and latest["criterion_current"] is False
+    assert latest["drift_signal"] == {"noul": None, "choice": None}
+
+

From ddd1ed764c0d8b95e502ad3f787a345cfffdbc8a Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 13:00:01 +0800
Subject: [PATCH 4/6] fix(progress-review): preserve exact Goal identity and
 legacy provider compatibility

Signed-off-by: song <liusongstep@gmail.com>
---
 loopx/capabilities/progress_review/context.py |  4 ++++
 packages/loopx-jev/src/loopx_jev/runner.py    | 22 +++++++++++++------
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/loopx/capabilities/progress_review/context.py b/loopx/capabilities/progress_review/context.py
index 8f8eed1530..66ba9f909c 100644
--- a/loopx/capabilities/progress_review/context.py
+++ b/loopx/capabilities/progress_review/context.py
@@ -34,6 +34,7 @@ def external_progress_review_context(
     if policy["mode"] == "off" or runtime_root is None or not goal_id:
         return None
     from ...control_plane.effect_runtime import effect_runtime_result, EffectRuntimeRemoteError
+    from ...control_plane.goals.goal_ref_validation import exact_goal_ref
     read_error = None
     try:
         loaded, rejected = load_progress_review_receipts(Path(runtime_root), goal_id)
@@ -50,8 +51,11 @@ def external_progress_review_context(
             todo_id = binding["todo_id"]
             if todo_id not in current_by_todo:
                 try:
+                    goal_instance_id = goal.get("goal_instance_id")
                     observation = effect_runtime_result("goal.acceptance.inspect", {
                         "goal_id": goal_id, "runtime_root": str(runtime_root), "todo_id": todo_id,
+                        **({"goal_ref": exact_goal_ref(goal_id, goal_instance_id)}
+                           if goal_instance_id is not None else {}),
                     })
                     current_by_todo[todo_id] = observation.get("completion_requirements")
                 except (OSError, ValueError, EffectRuntimeRemoteError):
diff --git a/packages/loopx-jev/src/loopx_jev/runner.py b/packages/loopx-jev/src/loopx_jev/runner.py
index 9810552c92..c8458c6309 100644
--- a/packages/loopx-jev/src/loopx_jev/runner.py
+++ b/packages/loopx-jev/src/loopx_jev/runner.py
@@ -86,14 +86,14 @@ def read_basis(
             }
         )
         checks.append((target, digest))
-    from loopx.control_plane.effect_runtime import effect_runtime_result
-    from loopx.control_plane.goals.acceptance import inspect_goal_acceptance
+    from loopx.control_plane.effect_runtime import effect_runtime_result, EffectRuntimeRejected
 
     scope = manifest.get("acceptance_scope")
     def criterion_basis():
         requirements = None
         requested = None
         if scope is not None:
+            from loopx.control_plane.goals.acceptance import inspect_goal_acceptance
             if not isinstance(scope, dict) or set(scope) != {
                 "registry_ref", "runtime_ref", "agent_id", "todo_id", "criterion_ids"
             }:
@@ -114,17 +114,25 @@ def local_reference(key):
             if requirements is None:
                 raise ValueError("current task has no canonical acceptance criteria")
             requested = scope["criterion_ids"]
-        return effect_runtime_result("progress_review.criterion_basis", {
-            "acceptance": manifest["acceptance"], "requirements": requirements, "criterion_ids": requested,
-            "goal_id": manifest.get("goal_id"), "agent_id": scope.get("agent_id") if scope else None,
-        })
+        try:
+            return effect_runtime_result("progress_review.criterion_basis", {
+                "acceptance": manifest["acceptance"], "requirements": requirements, "criterion_ids": requested,
+                "goal_id": manifest.get("goal_id"), "agent_id": scope.get("agent_id") if scope else None,
+            })
+        except EffectRuntimeRejected as exc:
+            # Older independently installed cores still support manual studies.
+            # Keep their legacy shape; never fabricate a canonical binding or
+            # bypass a current core's validation/IO failure.
+            if scope is None and exc.diagnostic_code == "unsupported_method":
+                return {"acceptance": manifest["acceptance"]}
+            raise
     selected = criterion_basis()
     basis = {
         **manifest,
         "evidence": observations,
         "basis_origin": "explicit_operator_study_basis_not_completion_authority",
         "acceptance": selected["acceptance"],
-        "criterion_binding": selected["binding"],
+        **({"criterion_binding": selected["binding"]} if "binding" in selected else {}),
     }
 
     def current() -> bool:

From 508d6a582a9a849af8ae4881f8a10d2fd843ce09 Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 13:00:01 +0800
Subject: [PATCH 5/6] docs(progress-review): clarify instance and independent
 install boundaries

Signed-off-by: song <liusongstep@gmail.com>
---
 ...osable-state-machines-recovery-verification-v0.md |  2 +-
 ...-state-machines-recovery-verification-v0.zh-CN.md |  2 +-
 packages/loopx-jev/DRIFT_SHADOW.md                   | 12 ++++++++++++
 packages/loopx-jev/DRIFT_SHADOW.zh-CN.md             |  8 ++++++++
 4 files changed, 22 insertions(+), 2 deletions(-)

diff --git a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
index 1da8ad873e..7d92d0726c 100644
--- a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
+++ b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.md
@@ -310,7 +310,7 @@ The table tracks independent composition boundaries within M2/M3 and roadmap R2/
 | --- | --- | --- |
 | Unavailable declared ancestry prevents new dependent use | `Delegations._read_current/start`, `delegation_results.require_dependencies/adoption_evidence`, `delegation.ts` | Alignment §3.8: source → A → B with A input invalidated must refuse current dependent use even if A output is unchanged. Qualify real read/start/adopt/settlement and packaged evidence readback; keep historical completion. |
 | Independent check failure reaches an actionable original-task recovery journey | `executor._task_validation_stage`, `ValidatedTurnReceipt`, canonical `turn_loop_controller_contract_v0.json`, `turn_journal.ts` | Preserve qualified failure scope and repair/replan detail; complete bounded repair or verifier-only retry, current validation and original-effect settlement. Host assertions cannot become trusted validation. |
-| Shadow review explains criteria and coverage | progress-review receipt/context, `progress_review_evidence.ts`, canonical acceptance inspect | Select current canonical task criteria; changed task bindings withdraw old judgments. Manual studies and legacy receipts retain distinct strength. Show separate dimensions, declared-file net change, missing evidence and unreadable storage. Default off and existing assist trigger rules retain their semantics. |
+| Shadow review explains criteria and coverage | progress-review receipt/context, `progress_review_evidence.ts`, canonical acceptance inspect | Select current canonical task criteria; exact GoalRef preserves instance identity, while changed tasks or recreated instances withdraw old judgments. Independent older cores retain only the legacy manual-study shape; canonical scope never downgrades. Show separate dimensions, declared-file net change, missing evidence and unreadable storage. Default off and existing assist trigger rules retain their semantics. |
 
 For the second outcome, `_task_validation_stage` already saves the independent
 result and blocks settlement. `ValidatedTurnReceipt` omits that validation's
diff --git a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
index 972be4f5b7..e24de651af 100644
--- a/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
+++ b/docs/architecture/rfcs/composable-state-machines-recovery-verification-v0.zh-CN.md
@@ -255,7 +255,7 @@ PostgreSQL authority、successor 调度和 App/Lark 送达尚未覆盖。因此
 | --- | --- | --- |
 | 声明来源链不可用时，阻止新的依赖使用 | `Delegations._read_current/start`、`delegation_results.require_dependencies/adoption_evidence`、`delegation.ts` | 遵循对齐 §3.8：source → A → B 中 A input 失效，即使 A output 不变也须拒绝当前依赖使用；验证真实 read/start/adopt/settlement 和打包证据读回，保留历史完成。 |
 | 独立检查失败进入原任务可操作的恢复旅程 | `executor._task_validation_stage`、`ValidatedTurnReceipt`、canonical `turn_loop_controller_contract_v0.json`、`turn_journal.ts` | 传递已资格化失败范围与 repair/replan 细节；完成有界修复或只运行 verifier 的重试、当前验证与原效果结算。Host 声明不能冒充可信验证。 |
-| shadow 说明条款及观察范围 | progress-review receipt/context、`progress_review_evidence.ts`、canonical acceptance inspect | 可选择任务当前规范 criterion；任务关联变化后撤回旧判断。手工研究和旧回执保持不同证据强度；显示独立维度、净文件变化覆盖、缺失及存储未知。默认 off 和 assist 既有触发规则保持不变。 |
+| shadow 说明条款及观察范围 | progress-review receipt/context、`progress_review_evidence.ts`、canonical acceptance inspect | 可选择任务当前规范 criterion；精确 GoalRef 保持实例身份，任务变化或实例重建撤回旧判断。独立旧核心只保留手工研究的旧格式；规范 scope 不降级。显示独立维度、净文件变化覆盖、缺失及存储未知。默认 off 和 assist 既有触发规则保持不变。 |
 
 第二项中，`_task_validation_stage` 已保存独立结果并阻断结算；`ValidatedTurnReceipt`
 未携带该验证的 `recovery_kind`，canonical controller 有意将 legacy
diff --git a/packages/loopx-jev/DRIFT_SHADOW.md b/packages/loopx-jev/DRIFT_SHADOW.md
index 0fdcd039e0..d754179081 100644
--- a/packages/loopx-jev/DRIFT_SHADOW.md
+++ b/packages/loopx-jev/DRIFT_SHADOW.md
@@ -87,6 +87,15 @@ selected canonical descriptions. Private registry/runtime references are not
 sent to the model or persisted in public receipt scope. Without this block the
 basis is explicitly an operator study, not canonical acceptance.
 
+Independent installations negotiate this method boundary. A core without
+`progress_review.criterion_basis` still runs manual studies using the legacy
+basis/receipt shape, without criterion scope metadata. This fallback applies
+only to the typed `unsupported_method` rejection and only without
+`acceptance_scope`; other failures remain errors. Canonical scope requires a
+core implementing the criterion and evidence-scope methods. If either is
+unavailable, upgrade the core before using that scope; it never becomes a
+manual study automatically.
+
 The binding pins Goal/Agent/Todo identity, contract revision/digest, task semantics
 and selected criteria. Capture rejects another task or agent before model dispatch;
 core receipt decoding applies the same identity rule. Select a new matching basis
@@ -94,6 +103,9 @@ when changing tasks.
 A change invalidates in-flight assessment; a fresh observation needs a new basis
 revision/baseline through the existing initialization path. Core readback also
 withdraws a newly scoped receipt when its canonical binding is no longer current.
+For an instantiated Goal, readback uses its exact GoalRef. Recreating the Goal
+retires the old acceptance, so an unchanged task ID cannot keep its old
+observation current. The retained historical receipt is not rewritten.
 The Goal's `assist` revision pin stays manual. Declared file net changes do not
 cover every intermediate decision or omitted dependency, and binding does not
 certify model correctness or grant mutation authority.
diff --git a/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md b/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
index 02a51391d1..b261e91683 100644
--- a/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
+++ b/packages/loopx-jev/DRIFT_SHADOW.zh-CN.md
@@ -47,11 +47,19 @@
 用选中的规范描述替换模型 acceptance 列表。私有 registry/runtime 引用不发送给模型，
 也不进入公开回执范围。没有该块时，basis 明确属于 operator study，不是规范验收。
 
+独立安装按方法能力区分兼容边界。核心缺少 `progress_review.criterion_basis` 时，
+手工研究仍使用原 basis／回执格式，不生成 criterion scope 元数据。只在没有
+`acceptance_scope` 且收到类型化 `unsupported_method` 拒绝时保留这条路径；其他失败
+继续报错。规范 scope 要求核心实现 criterion 和 evidence-scope 方法，缺少任一方法
+须先升级核心；不会自动改成手工研究。
+
 binding 固定 Goal/Agent/Todo 身份、contract 修订／摘要、任务语义及所选 criterion。
 采集在模型派发前拒绝错配任务或 Agent，核心回执解码使用同一规则；切换任务须选用
 匹配的新依据。变更使在途评估失效；
 新的观察沿既有初始化入口重建 basis revision／baseline。核心读回也会在新格式回执的
 规范 binding 不再当前时撤回信号。Goal 的 `assist` revision pin 仍是手动的。
+已实例化 Goal 的读回使用精确 GoalRef；重建会退休旧验收，同一任务 ID 不能使旧观察
+继续有效。保留的历史回执不被改写。
 声明文件净变化不覆盖每条中间判断或遗漏依赖；绑定不证明模型正确，也不授予修改权限。
 
 ## 操作方法

From 8e47dc13ab321fa4eb88bd3903d9945cbdc1d7b1 Mon Sep 17 00:00:00 2001
From: song <liusongstep@gmail.com>
Date: Sat, 10 Oct 2026 13:00:01 +0800
Subject: [PATCH 6/6] test(progress-review): qualify exact lifecycle and typed
 compatibility failures

Signed-off-by: song <liusongstep@gmail.com>
---
 packages/loopx-jev/tests/test_protocol.py     | 31 ++++++
 .../progress_review_evidence_fixture.py       |  6 +-
 .../test_progress_review_criterion_binding.py | 94 +++++++++++++++++--
 3 files changed, 124 insertions(+), 7 deletions(-)

diff --git a/packages/loopx-jev/tests/test_protocol.py b/packages/loopx-jev/tests/test_protocol.py
index d709d9d639..a5176f315b 100644
--- a/packages/loopx-jev/tests/test_protocol.py
+++ b/packages/loopx-jev/tests/test_protocol.py
@@ -13,6 +13,37 @@
 from loopx_jev.transport import send, TransportFailure
 
 
+@pytest.mark.parametrize("code", ["unsupported_method", "invalid_request", "io_permission_denied"])
+def test_manual_basis_preserves_only_the_typed_legacy_method_boundary(tmp_path, monkeypatch, code):
+    from loopx.control_plane.effect_runtime import EffectRuntimeRemoteError, EffectRuntimeRejected, EffectRuntimePermanentIOError
+    from loopx_jev.runner import read_basis
+
+    path = tmp_path / "basis.json"
+    manifest = {"goal_id": "study", "objective": "Study a declared file",
+                "acceptance": ["Observe its net change"], "evidence": [{"ref": "code.py"}]}
+    path.write_text(json.dumps(manifest))
+    evidence = tmp_path / "code.py"
+    evidence.write_text("before")
+    calls = []
+    def unsupported(method, params):
+        calls.append(method)
+        if code == "io_permission_denied":
+            raise EffectRuntimePermanentIOError("Fixture receipt store denied access", diagnostic_code=code)
+        raise EffectRuntimeRejected("Fixture core rejected the method", diagnostic_code=code)
+    monkeypatch.setattr("loopx.control_plane.effect_runtime.effect_runtime_result", unsupported)
+    if code != "unsupported_method":
+        with pytest.raises(EffectRuntimeRemoteError):
+            read_basis(path, tmp_path)
+    else:
+        basis, current = read_basis(path, tmp_path)
+        assert basis["acceptance"] == manifest["acceptance"]
+        assert "criterion_binding" not in basis
+        assert current()
+        evidence.write_text("after")
+        assert not current()
+    assert calls == ["progress_review.criterion_basis"]
+
+
 @pytest.mark.parametrize("raw", ['{"x":1,"x":2}', '{"x":NaN}', '{"x":Infinity}'])
 def test_ambiguous_json_rejected(raw):
     with pytest.raises(ValueError):
diff --git a/tests/presentation/progress_review_evidence_fixture.py b/tests/presentation/progress_review_evidence_fixture.py
index 779b075d9c..672dcd27f5 100644
--- a/tests/presentation/progress_review_evidence_fixture.py
+++ b/tests/presentation/progress_review_evidence_fixture.py
@@ -29,12 +29,16 @@ def main():
             runner.config = config
             registry = json.loads(runner.registry.read_text())
             goal = next(row for row in registry["goals"] if row["id"] == runner.goal_id)
+            goal["goal_instance_id"] = "ginst_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
             goal.setdefault("control_plane", {})["progress_review"] = {"mode": "shadow"}
             runner.registry.write_text(json.dumps(registry))
             from tests.capabilities.test_progress_review import receipt
             from loopx.capabilities.progress_review.receipt import write_progress_review_receipt
             from loopx.control_plane.effect_runtime import effect_runtime_result
-            from loopx.control_plane.goals.acceptance import inspect_goal_acceptance
+            from loopx.control_plane.goals.acceptance import inspect_goal_acceptance, transition_goal_acceptance_lifecycle
+            transition_goal_acceptance_lifecycle(runtime_root=runner.root, goal_id=runner.goal_id,
+                operation_id="bind-presented-review-instance", transition={"kind": "bind_existing", "goal_ref": {
+                    "goal_id": runner.goal_id, "goal_instance_id": goal["goal_instance_id"]}})
             requirements = inspect_goal_acceptance(registry_path=runner.registry, runtime_root=str(runner.root),
                 goal_id=runner.goal_id, agent_id="analyst", todo_id="todo_analyst-initial")["completion_requirements"]
             binding = effect_runtime_result("progress_review.criterion_basis", {
diff --git a/tests/test_progress_review_criterion_binding.py b/tests/test_progress_review_criterion_binding.py
index e8df697891..4cecff7c2a 100644
--- a/tests/test_progress_review_criterion_binding.py
+++ b/tests/test_progress_review_criterion_binding.py
@@ -1,7 +1,7 @@
 """Version-bound requester decisions through real Turn acceptance and durable IO."""
 import json
+from pathlib import Path
 import subprocess
-import sys
 
 import pytest
 
@@ -10,11 +10,30 @@
 service = delegation_service
 
 
-def test_shadow_basis_uses_current_canonical_criteria_and_withdraws_after_task_edit(service):
-    from pathlib import Path
-    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "packages/loopx-jev/src"))
+@pytest.fixture(autouse=True)
+def provider_imports(monkeypatch):
+    package = Path(__file__).resolve().parents[1] / "packages/loopx-jev"
+    monkeypatch.syspath_prepend(str(package / "src"))
+    monkeypatch.syspath_prepend(str(package / "tests"))
+
+
+def bind_exact_goal(runner):
+    from loopx.control_plane.goals.acceptance import transition_goal_acceptance_lifecycle
+    registry = json.loads(runner.registry.read_text())
+    goal = next(row for row in registry["goals"] if row["id"] == runner.goal_id)
+    goal["goal_instance_id"] = "ginst_aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa"
+    runner.registry.write_text(json.dumps(registry))
+    transition_goal_acceptance_lifecycle(runtime_root=runner.root, goal_id=runner.goal_id,
+        operation_id="bind-review-instance", transition={"kind": "bind_existing",
+            "goal_ref": {"goal_id": runner.goal_id, "goal_instance_id": goal["goal_instance_id"]}})
+    return goal
+
+
+@pytest.mark.parametrize("exact_instance", [False, True])
+def test_shadow_basis_uses_current_canonical_criteria_and_withdraws_after_task_edit(service, exact_instance):
     from loopx_jev.runner import read_basis
     root, runner = service
+    registered_goal = bind_exact_goal(runner) if exact_instance else {"id": runner.goal_id}
     manifest = root / "basis.json"
     manifest.write_text(json.dumps({"goal_id": runner.goal_id, "objective": "Review declared task change",
         "acceptance": ["Operator claim must not replace the canonical criterion"], "evidence": [],
@@ -28,7 +47,7 @@ def test_shadow_basis_uses_current_canonical_criteria_and_withdraws_after_task_e
     from tests.capabilities.test_progress_review import receipt
     from loopx.capabilities.progress_review.receipt import write_progress_review_receipt
     from loopx.capabilities.progress_review.context import external_progress_review_context
-    goal = {"id": runner.goal_id, "control_plane": {"progress_review": {"mode": "shadow"}}}
+    goal = {**registered_goal, "control_plane": {"progress_review": {"mode": "shadow"}}}
     canonical_receipt = receipt(goal_id=runner.goal_id,
         evidence_scope={"criterion_binding": basis["criterion_binding"], "coverage": "declared_file_net_change", "files": ["output.json"]})
     canonical_receipt["run"].update(agent_id="analyst", todo_id="todo_analyst-initial")
@@ -42,11 +61,13 @@ def test_shadow_basis_uses_current_canonical_criteria_and_withdraws_after_task_e
     assert rejected_context["receipts"] == []
     write_progress_review_receipt(runner.root, runner.goal_id, canonical_receipt)
     assert external_progress_review_context(goal, runner.root)["summary"]["latest"]["criterion_current"] is True
+    if exact_instance:
+        wrong_instance = {**goal, "goal_instance_id": "ginst_bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb"}
+        assert external_progress_review_context(wrong_instance, runner.root)["summary"]["latest"]["criterion_current"] is False
     # Exercise the real consumer too: canonical revisions include the owner
     # binding, so comparing only the unchanged manifest digest would reject all jobs.
     from loopx_jev import drift
     from loopx_jev.store import atomic_json
-    sys.path.insert(0, str(Path(__file__).resolve().parents[1] / "packages/loopx-jev/tests"))
     from drift_fixtures import response
     def git(*args):
         subprocess.run(["git", "-C", str(root), *args], check=True, capture_output=True)
@@ -102,3 +123,64 @@ def edit_during_request(request, *args):
     assert latest["drift_signal"] == {"noul": None, "choice": None}
 
 
+def test_scoped_basis_never_downgrades_an_unsupported_core_to_manual(service, monkeypatch):
+    from loopx.control_plane import effect_runtime
+    from loopx_jev.runner import read_basis
+    root, runner = service
+    manifest = root / "scoped-basis.json"
+    manifest.write_text(json.dumps({"goal_id": runner.goal_id, "objective": "Review the declared task",
+        "acceptance": ["Manual wording cannot replace canonical criteria"],
+        "acceptance_scope": {"registry_ref": "registry.json", "runtime_ref": "runtime", "agent_id": "analyst",
+            "todo_id": "todo_analyst-initial", "criterion_ids": ["analyst-initial"]}}))
+    original = effect_runtime.effect_runtime_result
+    def old_core(method, params, **kwargs):
+        if method == "progress_review.criterion_basis":
+            raise effect_runtime.EffectRuntimeRejected("Method unavailable", diagnostic_code="unsupported_method")
+        return original(method, params, **kwargs)
+    monkeypatch.setattr(effect_runtime, "effect_runtime_result", old_core)
+    with pytest.raises(effect_runtime.EffectRuntimeRejected):
+        read_basis(manifest, root)
+
+
+def test_recreated_goal_with_same_task_never_reuses_the_retired_observation(service):
+    from loopx.control_plane.projects.registry_codec import source_session_registry_transaction
+    from loopx.control_plane.goals.source_session_recreation import RecreateGoalRequest, recreate_goal_instance
+    from loopx_jev.runner import read_basis
+    from tests.capabilities.test_progress_review import receipt
+    from loopx.capabilities.progress_review.receipt import write_progress_review_receipt
+    from loopx.capabilities.progress_review.context import external_progress_review_context
+
+    root, runner = service
+    goal = bind_exact_goal(runner)
+    goal["control_plane"] = {"progress_review": {"mode": "shadow"}}
+    source_registry = root / "source-registry.json"
+    payload = {"schema_version": "0.2", "registry_role": "project-local", "profile_id": "source_session_v1",
+        "common_runtime_root": str(runner.root), "projects": [], "goals": [goal],
+        "session_bindings": [], "session_receipts": [], "lifetime_receipts": [], "retired_goal_instances": []}
+    with source_session_registry_transaction(source_registry, operation="review-source-fixture", create=lambda: payload) as tx:
+        tx.commit(tx.payload_copy())
+    manifest = root / "source-basis.json"
+    manifest.write_text(json.dumps({"goal_id": runner.goal_id, "objective": "Review exact instance work",
+        "acceptance": ["Canonical basis required"], "acceptance_scope": {
+            "registry_ref": "registry.json", "runtime_ref": "runtime", "agent_id": "analyst",
+            "todo_id": "todo_analyst-initial", "criterion_ids": ["analyst-initial"]}}))
+    basis, current = read_basis(manifest, root)
+    observed = receipt(goal_id=runner.goal_id, evidence_scope={"criterion_binding": basis["criterion_binding"],
+        "coverage": "declared_file_net_change", "files": ["output.json"]})
+    observed["run"].update(agent_id="analyst", todo_id="todo_analyst-initial")
+    path = write_progress_review_receipt(runner.root, runner.goal_id, observed)
+    historical = path.read_bytes()
+    assert current()
+    assert external_progress_review_context(goal, runner.root)["summary"]["latest"]["criterion_current"] is True
+    recreated = recreate_goal_instance(RecreateGoalRequest(registry_path=source_registry,
+        goal_id=runner.goal_id, goal_instance_id=goal["goal_instance_id"], operation_id="recreate-review-instance"))
+    assert recreated["ok"] is True, recreated
+    assert recreated["goal_ref"]["goal_instance_id"] != goal["goal_instance_id"]
+    active = {**goal, "goal_instance_id": recreated["goal_ref"]["goal_instance_id"]}
+    assert not current()
+    for instance in (goal, active):
+        latest = external_progress_review_context(instance, runner.root)["summary"]["latest"]
+        assert latest["criterion_current"] is False and latest["status"] == "stale"
+        assert latest["drift_signal"] == {"noul": None, "choice": None}
+    assert path.read_bytes() == historical
+