Integrate staged replan context as an isolated benchmark treatment

docs/architecture/rfcs/hierarchical-agent-stride-control-v0.md

@@ -762,17 +762,29 @@ these facts do not change the admitted context/basis identity. Changed source
 requires rereading within the original Turn. Page coordinates persist with the
 decision into subsequent review; node-filtered/planner detail routes remain
 outside this bounded citation path. No separate receipt store is introduced.
-The enabled review also carries the prior decision's existing `todo_delta`
+The replan context also carries the prior decision's existing `todo_delta`
 targets in `decision_followthrough.planned_work`: current canonical status and
 claim, plus references to causally later observations of that exact work item.
 Legacy descriptive Todo deltas remain readable with unresolved linkage; an
 overlong reference is never truncated into a different Todo identity.
 Missing Todo sources and omitted work remain explicit; a missing observation
 does not prove non-execution, and Todo completion does not prove objective gain.
 Related observations survive unrelated recent activity in the bounded window.
-The same bounded followthrough is included in `stride_control` so compact Host
-envelopes expose it without requiring another full-context read. Historical
-decision identities and off-mode output remain unchanged.
+Generic followthrough is available with stride off as well: `result_counts`
+separates advanced, unchanged, blocked and terminal observations, while visible
+rows retain their result class and delivery outcome. The compatibility
+`recorded_progress_count` is an observation count, never successful work.
+Historical decision identities and stride admission/settlement gates remain
+unchanged; off-mode context now includes this feedback without enabling stride.
+
+The [captured replan reader](../../reference/protocols/turn-envelope-v0.md#captured-replan-reads--捕获后的重规划读取)
+separates state review from authoring on one immutable decision snapshot.
+Cumulative acceptance uses lossless content-bound detail reads, and Explore
+receives a separate context increment. Existing uncaptured callers retain their
+full delivery. File/SQLite source tests and populated packet budgets qualify
+transport and writer continuity; installed adoption, real model decisions and
+matched outcome/cost benefit remain unqualified. This introduces no new planning
+receipt, model checkpoint or requirement to change direction.
 
 The next-review input also exposes causally later ordinary writebacks for the
 declared work in `ordinary_work`, with per-Todo `ordinary_evidence_refs`. Its

docs/architecture/rfcs/hierarchical-agent-stride-control-v0.zh-CN.md

@@ -632,14 +632,22 @@ revision、考虑的 finding 与适用/暂缓等理由，不建新图或第二
 拒绝缺少复盘、错误引用以及过期或不可用依据的错误声明，校验提案结构和来源，
 不能证明研究已执行或有收益。
 
-启用策略后，`decision_followthrough.planned_work` 将上次决定中既有 `todo_delta`
+replan context 中的 `decision_followthrough.planned_work` 将上次决定中既有 `todo_delta`
 指向的 Todo 与当前规范状态、认领人和该工作项在决定之后的观测引用一起返回。
 Todo 来源缺失和省略数量明确保留；缺少观测不等于没有执行，Todo 完成也不等于
 目标收益。历史描述性 Todo delta 保持可读并注明关联未解析；过长引用不会截断成
 另一个 Todo 的身份。相关观测不会被大量较新的无关活动挤出有限窗口。
-同一份有界反馈进入
-`stride_control`，紧凑 Host envelope 可直接消费，避免为这些信息再读完整上下文。
-历史决定的引用身份及 off 模式输出不变。
+通用执行回顾在 stride 关闭时也可读取：`result_counts` 区分推进、未变、阻塞和
+终态观测，明细保留结果分类及交付结果。兼容字段 `recorded_progress_count` 表示
+观测数量，不表示成功工作。历史决定引用及 stride 准入/结算门禁不变；off 模式
+新增反馈上下文，但不启用 stride。
+
+[捕获后的重规划读取](../../reference/protocols/turn-envelope-v0.md#captured-replan-reads--捕获后的重规划读取)
+把状态审视和写回格式分开，复用同一不可变决策快照。累计验收通过内容绑定的明细
+读取完整保留，Explore 使用独立的上下文增量预算；未捕获快照的调用仍完整投递。
+File/SQLite 源码测试和真实规模 packet 预算验证传输与写回连续性；已安装采用、真实
+模型决策及可比成果/成本收益仍待资格化。不新增规划回执、模型 checkpoint 或强制
+改变方向的要求。
 
 下一轮输入还通过 `ordinary_work` 与各 Todo 的 `ordinary_evidence_refs` 返回已
 声明工作在决定之后的普通回写。详情最多六项，保留总数、省略数和稳定 history

docs/reference/protocols/turn-envelope-v0.md

@@ -77,6 +77,86 @@ CLI from becoming another reader of the legacy guidance field.
 事实仍在当前包内，不需再次执行 guard。不同副本和既有详细视图保留；原始决定与
 capture 不变。动作维度等价不意味着不同渲染可以共用原始 source hash。
 
+### Captured replan reads / 捕获后的重规划读取
+
+For a successful replan with `--decision-output-dir` or `--decision-output-root`,
+the CLI and TurnEnvelope expose `replan_action_packet.review` followed by
+`writeback`. Both are executable `quota replan-view` commands bound to the same
+captured JSON value and source hash. The first read supplies goal context,
+previous decision/results, planning dependencies and evidence references; the
+second supplies the existing authoring shape and settlement commands. No read
+runs another guard, writes a receipt or authorizes work. There is no additional
+read acknowledgement, mandatory model call or prescribed route change.
+
+Captured reads retain the caller's explicit `--cli-bin` route, including later
+acceptance and Explore reads; an ambient executable cannot silently replace it.
+The review stage provides structured facts without a prescribed diagnostic
+method. Writeback omits filled-in Vision/stride examples while retaining field
+limits, legal options, source bindings and settlement rules. Stride exposes its
+required field names and basis options independently of examples. The unchanged
+full capture remains available for optional examples and source inspection.
+This intentionally reduces advisory prompting; it does not remove acceptance,
+authority or stop obligations, nor demonstrate better autonomous decisions.
+
+Cumulative outcome requirements are indexed by exact content hash. Every
+objective, revision and requirement count remains visible; the acceptance read
+returns the complete original outcome. Identical IDs with different revisions
+or evidence remain distinct. Equal context/horizon copies share references;
+differing goal scopes remain separate. The full capture retains all evidence,
+legacy guidance and authoring examples. Explore has an optional, separately
+readable basis; unavailable evidence remains unavailable. These views do not
+assess acceptance, infer execution from counts or make a dependency global.
+
+`quota replan-view --decision-file <file> --source-hash <hash>` defaults to
+`--replan-stage review`; `writeback`, `acceptance --outcome-ref <ref>` and
+`explore` read the corresponding original-source detail. Missing/mutated sources
+fail without guard replay. Existing uncaptured calls and native journal captures
+without the CLI read route retain their prior delivery. The existing generic
+followthrough is now available with stride off; its result-class counts and
+observations remain evidence, not successful-work or goal-gain verdicts.
+
+对使用 `--decision-output-dir` 或 `--decision-output-root` 的成功 replan，CLI 和
+TurnEnvelope 先给出 `review` 状态读取，再给出 `writeback` 格式读取。两步绑定同一
+快照和来源哈希，不重新准入、不新增回执或授权，也不强制换路线。累计验收按完整
+内容哈希建立索引：每个目标、版本及要求数量可见，明细完整读回；相同 ID 的不同
+版本不会覆盖。相同的目标上下文副本共享引用，不同范围保留。Explore 明细单独按需
+读取；不可用保持不可用。完整快照仍保存原指导、示例和证据。没有捕获或没有 CLI
+读取路径的原生 journal 保持原交付方式。通用执行回顾不再依赖 stride 开关；分类
+计数只表示观测，不等于工作成功或目标收益。
+
+捕获后的读取沿用调用方显式选择的 `--cli-bin`，后续验收和 Explore 读取也不会
+悄悄切换到环境中的其他版本。复盘视图提供结构化事实，不预设诊断方法；写回视图
+移除预填的 Vision／stride 答案示例，保留字段限制、合法选项、来源绑定和结算规则。
+stride 的必需字段与 basis 选项独立展示，完整快照仍可用于按需查看示例与源材料。
+这是有意减少建议性提示，不删除验收、权限或停止义务，也不证明自主决策质量提升。
+
+Regression budgets use compact UTF-8 JSON, separately from compute quota and
+provider tokens. The synthetic workload uses a real append index with 1,500
+records, 21 later observations, 24 cumulative outcomes and a populated Explore
+log. Limits protect default input while exact details remain readable:
+
+| View | Regression allocation |
+| --- | --- |
+| Replan entry packet | 2 KiB |
+| Populated history review | 16 KiB |
+| Planning inventory increment | 8 KiB |
+| Cumulative acceptance index increment | 12 KiB |
+| Explore increment in entry / review | 128 B / 768 B |
+| Explore detail / writeback | 8 KiB / 16 KiB |
+
+The whole-envelope differential also prevents a large acceptance copy moving to
+another hot carrier. These are fixture regression allocations, not runtime
+truncation or refusal limits; the existing 8 KiB envelope warning remains.
+File/SQLite CLI tests cover capture → reads → existing writeback, hash rejection
+and unchanged state during reads. They do not establish model choice quality,
+installed adoption, token savings or matched long-horizon benefit.
+
+大小回归使用紧凑 UTF-8 JSON，与算力 quota 和 provider token 分开。通用 review、
+累计验收索引和 Explore 分别约束增量，同时检查整个 envelope，防止把大副本挪到
+另一处热输入。预算不在运行时截断或拒绝合法内容，原 8 KiB envelope 警告保留。
+File/SQLite CLI 验证快照、分步读取、既有写回和哈希拒绝，不据此宣称模型决策、
+已安装采用、token 节省或长程收益改善。
+
 ### Native Turn Host capture
 
 Executing `turn run-once --execute` now saves the original full decision before

loopx/cli_commands/quota.py

@@ -1,11 +1,12 @@
 from __future__ import annotations
 
 import argparse
+import json
 import time
 from collections.abc import Callable, Mapping
 from pathlib import Path
 
-from .quota_capture import bind_capture_selection_transport, capture_decision, handoff_decision, prepare_decision_capture
+from .quota_capture import bind_capture_selection_transport, capture_decision, handoff_decision, handle_replan_view, prepare_quota_capture
 from ..usage_goal import observe_quota_result
 
 from ..capabilities.explore.composition_frontier import (
@@ -411,6 +412,9 @@ def _project_quota_cli_payload(
         if args.quota_command in {"status", "plan"}:
             return compact_quota_plan_cli_payload(payload, detail_sections=detail_sections)
         if args.quota_command == "should-run":
+            if captured_decision_path and "vision" not in detail_sections:
+                from ..control_plane.quota.turn_envelope import _turn_envelope_result
+                payload = _turn_envelope_result("replan_cli_view", payload=payload)
             return compact_quota_should_run_cli_payload(
                 payload,
                 include_todo_summary_detail="agent-todos" in detail_sections,
@@ -464,6 +468,11 @@ def _emit_quota_result(
             started_at=usage_quota_started,
         )
     bind_capture_selection_transport(payload, args)
+    if capture_directory is not None and payload.get("replan_action_packet"):
+        from ..command_invocation import loopx_cli_argv
+
+        payload = {**payload, "replan_read_path": str(capture_directory / "decision.json"),
+            "replan_read_argv": loopx_cli_argv()}
     try:
         capture_decision(capture_directory, payload)
         if payload.get("ok") is True:
@@ -504,6 +513,8 @@ def handle_quota_command(
     append_cli_rollout_event: RolloutEventAppender,
     capture_directory: Path | None = None,
 ) -> int:
+    if args.quota_command == "replan-view":
+        return handle_replan_view(args, print_payload)
     usage_quota_started = time.time_ns() // 1_000_000
     heartbeat_turn_id: str | None = None
     heartbeat_receipt_existing: dict[str, object] | None = None
@@ -518,13 +529,7 @@ def handle_quota_command(
     context: QuotaCommandContext | None = None
     goal_ref: dict[str, str] | None = None
     try:
-        if capture_directory is None and (
-            getattr(args, "decision_output_dir", None) is not None
-            or getattr(args, "decision_output_root", None) is not None
-            or getattr(args, "decision_handoff_socket", None) is not None
-        ):
-            validate_quota_command_context_request(args)
-            capture_directory = prepare_decision_capture(args)
+        capture_directory = prepare_quota_capture(args, capture_directory)
         goal_ref, turn_start_hook_dispatch, context = (
             _prepare_quota_command_execution(
                 args,

loopx/cli_commands/quota_capture.py

@@ -8,12 +8,28 @@
 import shlex
 import socket
 import tempfile
+from collections.abc import Callable
 from pathlib import Path
 
+from ..control_plane.effect_runtime import EffectRuntimeRejected
 from ..control_plane.quota.error_codes import QuotaCommandValidationError, QuotaDecisionTransportError
 from ..control_plane.turn_driver.decision_capture import validate_handoff_socket
 
 
+def prepare_quota_capture(args: argparse.Namespace, directory: Path | None) -> Path | None:
+    """Validate transport options before allocating a new guard capture."""
+    if directory is None and (
+        getattr(args, "decision_output_dir", None) is not None
+        or getattr(args, "decision_output_root", None) is not None
+        or getattr(args, "decision_handoff_socket", None) is not None
+    ):
+        from .quota_context import validate_quota_command_context_request
+
+        validate_quota_command_context_request(args)
+        return prepare_decision_capture(args)
+    return directory
+
+
 def prepare_decision_capture(args: argparse.Namespace) -> Path | None:
     requested = getattr(args, "decision_output_dir", None)
     root = getattr(args, "decision_output_root", None)
@@ -123,3 +139,29 @@ def handoff_decision(directory: Path | None, socket_path: str | None) -> None:
             "Decision handoff failed after guard evaluation; inspect the original "
             f"Turn receipt and {path}. Do not blindly rerun the guard."
         ) from exc
+
+
+def read_replan_view(args: argparse.Namespace) -> dict:
+    """Read a captured observation only; no registry, provider, guard or receipts."""
+    from ..control_plane.quota.turn_envelope import _turn_envelope_result
+
+    if not args.decision_file or not args.source_hash:
+        raise ValueError("replan-view requires --decision-file and --source-hash")
+    path = Path(args.decision_file).expanduser()
+    with path.open(encoding="utf-8") as stream:
+        payload = json.load(stream)
+    return _turn_envelope_result("replan_view", payload=payload, source_hash=args.source_hash,
+                                 stage=args.replan_stage, outcome_ref=args.outcome_ref)
+
+
+def handle_replan_view(
+    args: argparse.Namespace,
+    print_payload: Callable[[dict[str, object], str, Callable[[dict[str, object]], str]], None],
+) -> int:
+    """Render the captured read without entering live quota dispatch or recovery."""
+    try:
+        view = read_replan_view(args)
+    except (OSError, ValueError, EffectRuntimeRejected) as exc:
+        view = {"ok": False, "reason": str(exc)}
+    print_payload(view, args.format, lambda value: json.dumps(value, ensure_ascii=False, indent=2))
+    return 0 if view["ok"] else 1

loopx/cli_commands/quota_registration.py

@@ -27,6 +27,7 @@ def register_quota_command(
             "status",
             "plan",
             "should-run",
+            "replan-view",
             "monitor-poll",
             "scheduler-ack",
             "scheduler-ack-current",
@@ -37,6 +38,11 @@ def register_quota_command(
         default="status",
         help="Use status for all groups, plan for next-turn groups, should-run for one goal, monitor-poll for no-spend quiet poll evidence, scheduler-ack for successful Codex App RRULE state, scheduler-fail-current to suppress a repeated failed host update pair, spend-slot for accounting, or void-slot for a non-destructive accounting correction.",
     )
+    quota_parser.add_argument("--decision-file", help="Original captured decision for read-only replan-view.")
+    quota_parser.add_argument("--outcome-ref", help="Exact captured acceptance outcome reference for replan-view.")
+    quota_parser.add_argument("--source-hash", help="Original source_decision_hash for replan-view.")
+    quota_parser.add_argument("--replan-stage", choices=["review", "writeback", "acceptance", "explore"], default="review",
+                              help="Read decision context before opening its writeback format.")
     quota_parser.add_argument(
         "--goal-id",
         help="Goal id to check. Optional for status/plan (omit for all goals); required for should-run, scheduler ACK/failure, spend, and void. A scoped status/plan ranks only the selected goal, not the global next turn.",