diff --git a/CHANGELOG.md b/CHANGELOG.md
index b301ab4d..0d920af6 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,12 @@
 # 印度占星 Skill 更新日志
 
+## 2026-10-23 — 完整数据版最后增加一章独立年度方法参考
+
+- 完整数据版报告末尾新增「独立年度方法参考」。这一章单独用 PyJHora 计算请求年及后两年的 Mudda、Patyayini 和 Tajika，并写明来源。
+- 前面的本机年运、对话卡和校正打分不改用这套数字。PyJHora 只在单独的计算进程里运行。
+- 标准报告没有这一章。
+- Skill 版本不变（BUG-1427、BUG-1428）。
+
 ## 2026-10-23 — 阅读页的 Yogini 子运缺字段时不再整段消失
 
 - 补子运时只在原来的大运行上加子运，不改、不删原来的字段。缺完整年数时仍显示主运，不再整段报错。
diff --git a/deploy/README.md b/deploy/README.md
index e5c4d558..f014055b 100644
--- a/deploy/README.md
+++ b/deploy/README.md
@@ -39,6 +39,8 @@ The volume is not a backup domain. Chart cache is a TTL'd compute cache (default
 
 Heavy rectification scans and high-rigor workflows share a process-wide concurrency gate (`JYOTISH_HEAVY_COMPUTE_CONCURRENCY`, default `2` to match the 2 vCPU host). A saturated request waits up to 8 seconds if fewer than 4 requests are already waiting, then returns HTTP 429 with `Retry-After`. Health checks and other light routes do not take a slot.
 
+The API image installs PyJHora for one chapter only: the full-data report's last section, 「独立年度方法参考」. That chapter runs PyJHora in a child process. The API process does not import it. Chat cards, rectification scoring, and the rest of the report keep the previous no-PyJHora results. AGPL for this chapter is the product decision of 2026-10-21 (issue #93). Do not set `JYOTISH_YEAR_LORD_PYJHORA_REFERENCE`.
+
 ## DNS and Supabase Auth
 
 Final Spaceship resource records (apply only during the approved cutover window):
diff --git a/docs/BUG_HISTORY.md b/docs/BUG_HISTORY.md
index 1f327c3e..302778d2 100644
--- a/docs/BUG_HISTORY.md
+++ b/docs/BUG_HISTORY.md
@@ -17149,6 +17149,7 @@
 - 修复版本：本分支 `codex/special-lagnas-native-20261008`
 - Claude 验收（2026-10-08，Linux）：屏蔽 `jhora` 导入后跑 `cmd_full_reading`，乔布斯、奥巴马、虚构盘 × Lahiri / Raman 共 6 组，8 个点全部 `computed`、`source` 为 `native_special_lagna`，与 `fb8277bf` 调 PyJHora 的结果最大差 0°。Python 全量 62 条失败，与 `2df6ee72` 逐名相同；快速门通过；`tests/test_runtime_jhora_import_scan.py` 在快速门内，人为在 `native_special_lagnas.py` 加一行 `from jhora ...` 后该测试失败（变异检查）。年运资料包屏蔽 `jhora` 前后逐字节相同。
 - 待问占星师：Vighati 上升按每分钟 15° 计（与 PyJHora 一致）；按字面「每个 vighati 一个星座」应为每分钟 75°。
+- 决策推翻（2026-10-22，issue #93，不改本条状态）：产品决定生产镜像可以安装 PyJHora，但只给完整数据版最后一章「独立年度方法参考」用。调用放在单独的子进程里，父进程不加载它。八个特殊上升点仍走本条的本机算法，不改回进程内 PyJHora。AGPL 由产品接受。见 BUG-1427、BUG-1428。
 
 ## BUG-1274 | 中文报告把 Bhava Bala 译成年龄状态，并把 Planets 译成行星s
 
@@ -19350,3 +19351,35 @@
 - 复发自：无
 - 修复版本：分支 `codex/full-data-parity-1-20261021`。
 
+## BUG-1427 | 装上 PyJHora 后，对话交叉验证会在接口进程里加载它
+
+- 状态：resolved
+- 首次发现：2026-10-23
+- 最近更新：2026-10-23
+- 影响面：对话编排里的外部引擎交叉验证。完整数据版最后一章不走这条路径。
+- 用户现象：生产镜像一旦装上 PyJHora，对话在同一进程里探测并加载它。岁差和星历路径会被改掉，同一时刻的月亮经度不再稳定。
+- 触发条件：镜像含 PyJHora，且对话走到交叉验证。
+- 根因：产品路径用「能不能导入 jhora」决定要不要调用。装上之后探测成功，接着在接口进程里导入。
+- 修复：产品报告固定写成今天没装 PyJHora 时的结果，不探测、不导入。命令行诊断工具仍按原样报告是否可用。
+- 验证：`tests/test_pyjhora_annual_method_reference.py` 把探测函数换成会失败的替身，产品报告仍不调用它。`tests/test_unified_consultation_orchestrator.py` 的适配器状态仍是缺少 jhora 模块。`tests/test_pyjhora_adapter_diagnostics.py` 命令行行为不变。
+- 防复发：上面几条。产品路径再调用探测，测试会失败。
+- 相关记录：issue #93。BUG-1273、BUG-1428。
+- 复发自：无
+- 修复版本：分支 `codex/full-data-parity-3-20261022`。
+
+## BUG-1428 | 装上 PyJHora 后，校正三引擎包会在接口进程里加载它
+
+- 状态：resolved
+- 首次发现：2026-10-23
+- 最近更新：2026-10-23
+- 影响面：校正用的三引擎对照包里的 PyJHora 一行。生时校正打分本身不走这条路径。
+- 用户现象：生产镜像一旦装上 PyJHora，这行对照在接口进程里导入它，星盘其余数字可能被带着变。
+- 触发条件：镜像含 PyJHora，且组装三引擎对照包。
+- 根因：这一行直接导入 PyJHora。没装时得到「没有这个模块」；装上之后导入成功，计算发生在父进程。
+- 修复：产品路径固定抛出「没有这个模块」，不执行导入。对照行仍是今天没装时的那一行。
+- 验证：`tests/test_pyjhora_annual_method_reference.py` 检查这段源码没有导入语句，调用时得到模块缺失。封存合同测试 `tests/test_sealed_holdout_contract_freshness.py` 与 `tests/test_rectification_validation_integrity_gate.py` 仍通过。
+- 防复发：上述源码断言。再写回进程内导入，测试会失败。
+- 相关记录：issue #93。BUG-1273、BUG-1427。
+- 复发自：无
+- 修复版本：分支 `codex/full-data-parity-3-20261022`。
+
diff --git a/docs/tasks/PROGRESS-full-data-parity-3-20261022.md b/docs/tasks/PROGRESS-full-data-parity-3-20261022.md
new file mode 100644
index 00000000..87014139
--- /dev/null
+++ b/docs/tasks/PROGRESS-full-data-parity-3-20261022.md
@@ -0,0 +1,87 @@
+# PROGRESS — 完整数据版独立年度方法参考（#93）
+
+基线是已合入 #91 的 `origin/staging` `85e8adddbc9c14e31c687c91ad8666afe3290e64`。#92 没有合入 staging。没有收到「照上游原样」，D6 用默认：当地时间，Patyayini / Tajika 用「年龄 + 1 年」，Mudda 仍用年龄。
+
+## 做了什么
+
+完整数据版报告最后多了一章「独立年度方法参考」。这一章在单独的子进程里用 PyJHora 算请求年及后两年。父进程不加载 PyJHora。对话卡、校正和前面的本机年运仍按没装 PyJHora 时的结果。标准报告没有这一章。
+
+某一年某一张表算不出来时，该表写「未获得结果」，不把内部原因印进报告。
+
+## 换路径开关
+
+| 编号 | 处理 |
+| --- | --- |
+| S1 | 产品路径不启动子进程，仍返回原来的年运结果。研究脚本明确要求时才计算。测试里的旧替身只有 4 个参数，不会把整章打成未判定。 |
+| S2 | 对话交叉验证不探测、不导入。命令行诊断仍可报告是否装了。 |
+| S3 | 校正三引擎对照固定返回「没有这个模块」，不在接口进程里导入。 |
+| S4 | 部署文件不设置 `JYOTISH_YEAR_LORD_PYJHORA_REFERENCE`。Dockerfile 与工作流没改。 |
+| S5 | 没有新增被禁止的那两个模块。 |
+| S6 | 诊断脚本会显示 PyJHora 可用。它们不在产品路径上，没改。 |
+| S7 | 见下面的测试。装上之后这些用例会真跑，不再跳过。 |
+
+## 装与不装
+
+同一进程里，虚构盘 0 连续 5 次：去掉本章后的中文、英文各自哈希不变。父进程没有加载 jhora。英文汉字为 0。
+
+把「找不到 jhora」的拦截放进子进程后，再用同一个进程哈希种子重跑虚构盘 0、1、9：去掉本章后的中文和英文与装上时逐字相同。资料包去掉新键、生成时刻和缓存时刻后，只剩报告引用号。引用号来自结果指纹。结果指纹现在不把这一章算进去，所以装与不装的引用号一致。
+
+不固定哈希种子时，虚构盘 0 的太阳相位力量会差 0.01，虚构盘 1 的第一条王者格句子会换成另一条。这是既有代码用集合遍历，顺序随进程变化，不是这一章引入的。同一进程里的 5 次中文哈希相同。本轮没有改那两处旧代码。
+
+开工前的「改代码之前」对照没有量成。当时的测量停在星历下载，不能当作装与不装的差异清单。上面是改完之后的对照。
+
+## 两种年份口径（虚构盘）
+
+文档口径是本轮采用的默认。上游口径是世界时、并且年份不加 1。只列回归时刻和 Mudda 第一段起点，没有出生资料。小数小时已折成秒。
+
+| 虚构盘 | 岁差回读 | 交点 | 文档口径回归 | 上游口径回归 | 文档口径 Mudda 起点 | 上游口径 Mudda 起点 | 这次子进程秒数 |
+| --- | --- | --- | --- | --- | --- | --- | --- |
+| 0 | RAMAN，一致 | 平交点 | 2026-07-19 08:55:02 | 2025-07-18 23:19:06 | 2026-07-08 02:01:59 | 2026-07-10 19:12:54 | 50.057 |
+| 1 | LAHIRI，一致 | 平交点 | 2026-07-01 06:18:58 | 2025-06-30 21:05:13 | 2026-06-17 18:10:07 | 2026-06-21 00:36:03 | 52.986 |
+| 9 | RAMAN，一致 | 平交点 | 2026-05-08 23:36:33 | 2025-05-09 00:22:13 | 2026-04-29 06:01:58 | 2026-04-23 10:30:48 | 50.567 |
+
+这三次为了对照多算了一遍上游口径，所以比正式出报告的那一次更慢。虚构盘 0 上，正式计算复用同一次年盘，三次复用都与再算一次相同（此前已测）。
+
+## 时长（Windows，虚构盘 0，5 次）
+
+中文先渲染会把中文句子写进资料包，接着的英文会被判成有汉字而整份撤回。计时已改成英文使用渲染前的副本。下表是改完之后的数。第一次作废的英文时长不采用。
+
+| 项目 | 中位 | 最大 |
+| --- | --- | --- |
+| 只出中文 | 54.873 秒 | |
+| 只出英文（含同一次排盘） | 47.344 秒 | |
+| 中英一起 | 57.034 秒 | 59.013 秒 |
+
+排盘约 45.183 秒，中文排版 9.690 秒，英文排版 2.161 秒。中文约 821290 字，英文约 1358880 字。本章中文约 38313 字，英文约 44551 字。
+
+虚构盘 1 中英一起 78.568 秒，虚构盘 9 为 85.168 秒。三张都低于 180 秒。Linux 是否仍低于 180 秒由 Claude 复测。
+
+## 镜像
+
+本地镜像 `jyotisha-api:parity3-20261023`，`docker image inspect` 的 Size 为 932044815 字节。本机没有更早的 `jyotisha-api` 镜像，无法写增量。`docker history` 里安装依赖的那一层约 345MB。容器内可以导入 jhora、swisseph、geocoder、geopy。`pip show pyswisseph` 为 2.10.3.2。`pip check` 无冲突。Dockerfile 没改。
+
+`docker stats` 按让步没测。子进程工作集在这台 Windows 上的采样一直停在约 4.4MB，空进程也是这个数，不能当成峰值。峰值留给 Linux 复测。
+
+## 改过的断言
+
+| 测试 | 原值 | 新值 | 原因 |
+| --- | --- | --- | --- |
+| 阅读页三处泄漏检查 | 全文不得出现 PyJHora | 先去掉「独立年度方法参考」再查 | 这一章必须写来源 |
+| 英文泄漏检查 | 英文全文不得出现 PyJHora | 先去掉 Independent Annual Method Reference 再查 | 同上 |
+
+## 本轮跑过的检查
+
+- 结果指纹不包含这一章：通过。
+- 四参数替身仍能接到年运，不把整章打成未判定：通过。
+- 阅读页、英文版、完整数据章、子进程门禁：在加上结果指纹排除之前已通过（约 291 秒那一轮，外加更早的三张盘真算约 288 秒）。
+- 封存合同与校正完整性：在四参数修复之前的那一轮通过。本轮没重跑。
+- 全量 pytest 和快速门没跑，留给 Claude。
+- 隐私标记、导入扫描、结果指纹，以及 S7 的 Ashtakavarga 与三份大运对照：退出码 0，约 33 秒。
+
+## 偏离
+
+- T0 到 T6 放在同一次提交。章节标题测试要等渲染和依赖同时在，拆开会红。
+- 没装的环境用启动时拦截 `jhora`，没有另建一套虚拟环境。
+- 子进程超时是 120 秒。45 秒会在这台机器上把已经算完的三年掐掉。
+- 没做容器内存曲线。
+- 开工前的基线对照没有形成可用数字。
diff --git a/docs/tasks/README.md b/docs/tasks/README.md
index 358f1049..a64ca0e0 100644
--- a/docs/tasks/README.md
+++ b/docs/tasks/README.md
@@ -526,7 +526,7 @@
 | `TASK-full-data-parity-2-20261021.md` | — | #92 完整数据版补齐（二）：研究型推运附录（Tithi Ashtottari、Yoga / Nakshatra Pravesha、Karana 等，口径换成我方真恒星）、四种星座大运第三层、Kalachakra 第三层（简化模型）、Niryaana Shoola 子运只在有书面出处时做、D81 / D108 / D144 研究型分盘；全部标研究口径，不进对话卡与校正；不设大小上限 | 待执行 | `codex/full-data-parity-2-20261021` |
 | `TASK-goel-jaimini-strength-20261022.md` | `PROGRESS-goel-jaimini-strength-20261022.md` | #94 搬上游公开的 Goel Jaimini 强度与 Brahma / Maheshwara / Rudra（4 个上游原样文件，按 #81 决策 A 先例根目录预算 448→452；来源串改书目引用）；只做计算、不改报告；冻结身份不变 | 已合入 staging | `codex/goel-jaimini-strength-20261022` |
 | `TASK-chart-subject-table-fix-20261022.md` | `PROGRESS-chart-subject-table-fix-20261022.md` | **#86 第 1 次打回**：报告所有表被强制单行（打印跑出纸面）；保存他人复制了档案页代码且已分叉；点编辑框内边距会丢修改 | 已合入 staging | `codex/chart-subject-table-20261021` |
-| `TASK-full-data-parity-3-20261022.md` | — | #93 完整数据版补齐（三）：「独立年度方法参考」正式环境调 PyJHora（子进程隔离；装上后其余输出逐字节不变，7 个换路径开关钉死）；输入默认用本地时间与正确年份（D6 待产品确认） | 待领取（等 #91 合入） | `codex/full-data-parity-3-20261022` |
+| `TASK-full-data-parity-3-20261022.md` | `PROGRESS-full-data-parity-3-20261022.md` | #93 完整数据版补齐（三）：「独立年度方法参考」正式环境调 PyJHora（子进程隔离；装上后其余输出逐字节不变，7 个换路径开关钉死）；输入默认用本地时间与正确年份（D6 默认，未收到「照上游原样」） | 待验收 | `codex/full-data-parity-3-20261022` |
 | `TASK-synastry-staging-fix-fix-20261022.md` | `PROGRESS-synastry-staging-fix-fix-20261022.md` | **#85 第 1 次打回**（T7～T11 主体通过）：新地址没登记进路由清单致快速门红；CHANGELOG 与 T11 矛盾；侧栏点亮测试是空断言 | 已合入 staging | `codex/synastry-staging-fix-20261021` |
 | `TASK-report-english-han-leak-fix-20261022.md` | `PROGRESS-report-english-han-leak-fix-20261022.md` | **#89 第 1 次打回**：改了冻结的校正打分文件 `jaimini.py`；测试的 PyJHora 屏蔽没传到子进程（测的不是生产那份报告，也是慢的原因）；缺英文时留空藏住漏洞 | 已合入 staging | `codex/report-english-han-leak-20261021` |
 | `TASK-full-data-parity-1-fix-20261022.md` | `PROGRESS-full-data-parity-1-fix-20261022.md` | **#91 A 段第 1 次打回**：新六章 69 节里 16 节在三张虚构盘上整节「未判定」，其中 12 节数据在资料包里、是读错了字段；测试抓不到整节未判定；新六章绕过了报告口径替换；中文附录悄悄丢列和瑜伽条目要量出来 | 已合入 staging | `codex/full-data-parity-1-20261021` |
diff --git a/docs/testing/full-data-parity-3-20261022.md b/docs/testing/full-data-parity-3-20261022.md
new file mode 100644
index 00000000..38ba0c00
--- /dev/null
+++ b/docs/testing/full-data-parity-3-20261022.md
@@ -0,0 +1,11 @@
+# 完整数据版「独立年度方法参考」真机验收
+
+用虚构盘或公开名人盘。不要把真实出生资料写进记录。
+
+1. 打开一份完整数据版报告，翻到最后。最后一章标题是「独立年度方法参考」。
+2. 这一章有请求年、后一年、再后一年。每年都能看到来源是 PyJHora，以及 Mudda 两张表、Patyayini、Tajika。
+3. 某一格算不出来时，那一格写「未获得结果」，整份报告仍然能打开。
+4. 切换到英文。同一章标题是 Independent Annual Method Reference。英文里没有汉字。来源行是 Source: PyJHora。
+5. 导出 `.md`。最后一章仍在文件末尾。
+6. 打开一份这次更新之前生成的旧报告。它仍然能从列表打开，内容不因为这章而打不开。
+7. 打开一份标准报告（不是完整数据版）。没有「独立年度方法参考」这一章。
diff --git a/requirements-reference-engines.txt b/requirements-reference-engines.txt
index 0995f855..a24accda 100644
--- a/requirements-reference-engines.txt
+++ b/requirements-reference-engines.txt
@@ -1,7 +1,9 @@
 # Optional isolated reference-engine environment.
 #
-# PyJHora is AGPL-3.0 and must remain an external benchmark/reference engine.
-# Do not merge this file into the commercial production runtime requirements.
+# Product decision 2026-10-21 (issue #93): PyJHora 4.8.7 is installed from
+# requirements.txt for one full-data chapter only, and only inside a child
+# process. AGPL risk for that chapter is accepted. The API process must not
+# import it. This file stays the pinned reference list.
 # PyJHora 4.8.7 omits geocoder from its PyPI dependency metadata even though
 # jhora.utils imports it, so the missing runtime dependency is explicit here.
 PyJHora==4.8.7
diff --git a/requirements.txt b/requirements.txt
index 374dafc1..28b38062 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,6 +3,10 @@
 
 # 核心：Swiss Ephemeris 天文计算库（必需）
 pyswisseph
+# 仅供完整数据版「独立年度方法参考」子进程使用，AGPL，产品 10-21 决定。
+PyJHora==4.8.7
+geocoder==1.38.1
+geopy==2.4.1
 timezonefinder>=6.5
 
 # VedAstro 官方 Python SDK（官方快照与交叉验证）
diff --git a/scripts/annual_pyjhora_replay.py b/scripts/annual_pyjhora_replay.py
index 5f02a453..a82a3ed9 100755
--- a/scripts/annual_pyjhora_replay.py
+++ b/scripts/annual_pyjhora_replay.py
@@ -39,6 +39,13 @@ def _jhora_importable() -> bool:
     return _jhora_spec_cache
 
 
+def _call_isolated_runner(fn, profile, birth_julian_day, place, age, *, research: bool):
+    """Call a 4-arg test stand-in, or the real runner that also takes research."""
+    if "research" in signature(fn).parameters:
+        return fn(profile, birth_julian_day, place, age, research=research)
+    return fn(profile, birth_julian_day, place, age)
+
+
 def _no_jhora_tajaka_replay(settings: dict[str, Any]) -> dict[str, Any]:
     """Same dict the child returns when ``import jhora`` raises ModuleNotFoundError."""
     copied = dict(settings)
@@ -92,6 +99,7 @@ def probe_pyjhora_year_lord(
     import_module: Callable[[str], Any] = _import_module,
     distribution_version: Callable[[str], str] = _distribution_version,
     isolated_replay: Callable[[dict[str, Any], float, Any, int | float], dict[str, Any]] | None = None,
+    research: bool = False,
 ) -> dict[str, Any]:
     """Probe PyJHora for an annual Tajaka/Panchadhikari Year Lord API.
 
@@ -123,8 +131,13 @@ def probe_pyjhora_year_lord(
         "year_lord_replay": replay,
     }
 
-    replay_runner = isolated_replay or _run_isolated_tajaka_year_lord
-    isolated = replay_runner(profile, birth_julian_day, place, age)
+    if isolated_replay is None:
+        isolated = _call_isolated_runner(
+            _run_isolated_tajaka_year_lord,
+            profile, birth_julian_day, place, age, research=research,
+        )
+    else:
+        isolated = isolated_replay(profile, birth_julian_day, place, age)
     if isolated.get("reason") != "pyjhora_tajaka_api_missing":
         replay.update(isolated)
         replay["normalized_candidates"] = _normalize_tajaka_candidates(replay.get("raw_candidates"))
@@ -159,6 +172,8 @@ def _run_isolated_tajaka_year_lord(
     birth_julian_day: float,
     place: Any,
     age: int | float,
+    *,
+    research: bool = False,
 ) -> dict[str, Any]:
     settings = _replay_settings(profile)
     request = {
@@ -168,7 +183,9 @@ def _run_isolated_tajaka_year_lord(
         "settings": settings,
     }
     request_hash = _request_hash(request)
-    if not _jhora_importable():
+    # Product calls leave research false and return the no-install golden.
+    # They do not call find_spec. Only an explicit research replay may spawn.
+    if not research or not _jhora_importable():
         replay = _no_jhora_tajaka_replay(settings)
         replay["request_hash"] = request_hash
         return replay
@@ -298,6 +315,8 @@ def _run_isolated_annual_replay(
     birth_julian_day: float,
     place: Any,
     age: int | float,
+    *,
+    research: bool = False,
 ) -> dict[str, Any]:
     """Call optional PyJHora annual surfaces with a child-process Place."""
 
@@ -309,7 +328,8 @@ def _run_isolated_annual_replay(
         "settings": settings,
     }
     request_hash = _request_hash(request)
-    if not _jhora_importable():
+    # Same pin as the Tajaka year-lord child: product traffic stays on the golden.
+    if not research or not _jhora_importable():
         replay = _no_jhora_annual_replay(settings)
         replay["request_hash"] = request_hash
         return replay
@@ -662,16 +682,25 @@ def replay_pyjhora_annual(
     import_module: Callable[[str], Any] = _import_module,
     year_lord_probe: Callable[..., dict[str, Any]] | None = None,
     isolated_annual_replay: Callable[[dict[str, Any], float, Any, int | float], dict[str, Any]] | None = None,
+    research: bool = False,
 ) -> dict[str, Any]:
     """Replay annual PyJHora methods for the same profile, if available."""
 
-    year_lord_observer = year_lord_probe or probe_pyjhora_year_lord
-    year_lord_result = year_lord_observer(
-        profile,
-        birth_julian_day=birth_julian_day,
-        place=place,
-        age=age,
-    )
+    if year_lord_probe is None:
+        year_lord_result = probe_pyjhora_year_lord(
+            profile,
+            birth_julian_day=birth_julian_day,
+            place=place,
+            age=age,
+            research=research,
+        )
+    else:
+        year_lord_result = year_lord_probe(
+            profile,
+            birth_julian_day=birth_julian_day,
+            place=place,
+            age=age,
+        )
     year_lord_replay = year_lord_result.get("year_lord_replay") if isinstance(year_lord_result, dict) else None
     blocked = _base(profile)
     blocked["year_lord_replay"] = year_lord_replay if isinstance(year_lord_replay, dict) else {
@@ -681,8 +710,13 @@ def replay_pyjhora_annual(
     blocked["evidence_scope"] = "pyjhora_behavior_only"
     blocked["parity_status"] = "not_multiengine_parity"
     if isolated_annual_replay is not None or (import_module is _import_module and year_lord_probe is None):
-        observer = isolated_annual_replay or _run_isolated_annual_replay
-        observed = observer(profile, birth_julian_day, place, age)
+        if isolated_annual_replay is None:
+            observed = _call_isolated_runner(
+                _run_isolated_annual_replay,
+                profile, birth_julian_day, place, age, research=research,
+            )
+        else:
+            observed = isolated_annual_replay(profile, birth_julian_day, place, age)
         if not isinstance(observed, dict):
             observed = _annual_replay_blocked(
                 "pyjhora_isolated_annual_replay_invalid_payload",
diff --git a/scripts/diagnose_pyjhora_adapter.py b/scripts/diagnose_pyjhora_adapter.py
index fba2a704..a8038c52 100644
--- a/scripts/diagnose_pyjhora_adapter.py
+++ b/scripts/diagnose_pyjhora_adapter.py
@@ -17,6 +17,38 @@ ROOT = Path(__file__).resolve().parents[1]
 ADAPTER_RELATIVE = "benchmarks/jyotish/scripts/run_pyjhora_compare.py"
 
 
+def build_product_report() -> dict:
+    """Product path: the same dict as a machine without PyJHora.
+
+    Does not call find_spec and does not import jhora. The CLI still uses
+    ``build_report``, which may probe.
+    """
+    adapter = ROOT / ADAPTER_RELATIVE
+    module_name = os.environ.get("PYJHORA_MODULE_NAME", "jhora").strip() or "jhora"
+    adapter_exists = adapter.exists()
+    status = "missing_adapter" if not adapter_exists else "missing_dependency"
+    return {
+        "scope": "pyjhora_adapter_diagnostics",
+        "status": status,
+        "adapter_command": f"python3 {ADAPTER_RELATIVE}",
+        "adapter_exists": adapter_exists,
+        "dependency_module": module_name,
+        "dependency_available": False,
+        "missing_dependency": module_name,
+        "dependency_import_error": None,
+        "install_hint": {
+            "package": "PyJHora",
+            "commands": [
+                "pip install -r requirements.txt -r requirements-reference-engines.txt"
+            ],
+            "note": "Install in an isolated optional benchmark environment, not as a hard runtime dependency.",
+        },
+        "license_boundary": "AGPL external benchmark only; do not vendor or make it a runtime dependency.",
+        "ephemeris_data_note": "Recent PyJHora releases may require separate Swiss Ephemeris data download/configuration before full chart comparison can run.",
+        "boundary": "This verifies required PyJHora modules import successfully, but does not run chart comparison.",
+    }
+
+
 def build_report() -> dict:
     adapter = ROOT / ADAPTER_RELATIVE
     module_name = os.environ.get("PYJHORA_MODULE_NAME", "jhora").strip() or "jhora"
diff --git a/scripts/jyotish_engine.py b/scripts/jyotish_engine.py
index 4ba95d48..14345307 100644
--- a/scripts/jyotish_engine.py
+++ b/scripts/jyotish_engine.py
@@ -2332,6 +2332,29 @@ def _attach_annual_tajika_pack(packet: dict, args) -> dict:
     return packet
 
 
+def _attach_pyjhora_annual_method_reference(packet: dict, args):
+    """One call into the subprocess-backed annual method reference."""
+    try:
+        from scripts.report_render.annual_method_reference import attach_pyjhora_annual_method_reference
+    except ModuleNotFoundError:  # pragma: no cover - direct script execution
+        from report_render.annual_method_reference import attach_pyjhora_annual_method_reference
+    return attach_pyjhora_annual_method_reference(packet, args)
+
+
+def _bind_result_without_annual_method_reference(packet: dict, profile: dict) -> dict:
+    """Keep the new chapter out of result_hash. The chart result itself is unchanged."""
+    worksheets = packet.get("worksheets") if isinstance(packet, dict) else None
+    timing = worksheets.get("timing_and_predictive_systems") if isinstance(worksheets, dict) else None
+    detached = None
+    if isinstance(timing, dict) and "pyjhora_annual_method_reference" in timing:
+        detached = timing.pop("pyjhora_annual_method_reference")
+    try:
+        return bind_result_to_profile(packet, profile)
+    finally:
+        if isinstance(timing, dict) and detached is not None:
+            timing["pyjhora_annual_method_reference"] = detached
+
+
 def _uncertainty_minutes_from_args(args):
     explicit = getattr(args, 'uncertainty_minutes', None)
     if explicit not in (None, ''):
@@ -18083,6 +18106,7 @@ def build_professional_report_reference_packet(
     packet = build_pl9_style_export_packet(full_reading, include_raw=True)
     packet = attach_calculation_profile(packet, args)
     packet = _attach_annual_tajika_pack(packet, args)
+    packet = _attach_pyjhora_annual_method_reference(packet, args)
     packet = _attach_report_governance_contracts(packet, args)
     packet = _attach_base_charts_pack(packet)
     packet = _attach_visual_chart_pack(packet, args)
@@ -18100,7 +18124,7 @@ def build_professional_report_reference_packet(
         and isinstance(profile.get('input_hash'), str)
         and len(profile['input_hash']) == 64
     ):
-        delivered = bind_result_to_profile(delivered, profile)
+        delivered = _bind_result_without_annual_method_reference(delivered, profile)
     if isinstance(delivered, dict) and delivered.get('selected_report_scope') == 'full':
         evaluate_full_report = _try_attr_import('full_report_quality_gate', 'evaluate_full_report')
         if evaluate_full_report is None:
diff --git a/scripts/pl9_full_data_export.py b/scripts/pl9_full_data_export.py
index 143c2c6f..48f1092f 100644
--- a/scripts/pl9_full_data_export.py
+++ b/scripts/pl9_full_data_export.py
@@ -2417,7 +2417,8 @@ def render_full_data_markdown(packet: dict) -> str:
             _pl9_english_report_title(packet),
         )
         rendered = _append_parity_chapters(rendered, packet, "en")
-        return _apply_local_report_conventions(rendered, language="en")
+        rendered = _apply_local_report_conventions(rendered, language="en")
+        return _append_annual_method_reference(rendered, packet, "en")
     zh_structured_appendix = re.sub(
         r"^#\s+.*?(?:\n{2,}|\n)",
         "",
@@ -2439,7 +2440,20 @@ def render_full_data_markdown(packet: dict) -> str:
         )
     rendered = _sanitize_pl9_ai_density_markdown(zh_markdown, _pl9_public_report_title(packet))
     rendered = _append_parity_chapters(rendered, packet, "zh")
-    return _apply_local_report_conventions(rendered, language="zh")
+    rendered = _apply_local_report_conventions(rendered, language="zh")
+    return _append_annual_method_reference(rendered, packet, "zh")
+
+
+def _append_annual_method_reference(markdown: str, packet: dict, language: str) -> str:
+    """After the sanitizers. Those sanitizers delete any line that names PyJHora."""
+    try:
+        from scripts.report_render.annual_method_reference import render_annual_method_reference
+    except ModuleNotFoundError:  # pragma: no cover
+        from report_render.annual_method_reference import render_annual_method_reference
+    extra = render_annual_method_reference(packet, language).strip()
+    if not extra:
+        return markdown
+    return markdown.rstrip() + "\n\n" + extra + "\n"
 
 
 def read_full_data_domain(packet: Mapping, domain: str) -> dict:
@@ -2452,7 +2466,7 @@ def read_full_data_domain(packet: Mapping, domain: str) -> dict:
         raise KeyError(domain)
     if not isinstance(packet, Mapping):
         raise TypeError("packet must be a mapping")
-    return copy.deepcopy({
+    copied = copy.deepcopy({
         "domain": domain,
         "label": FULL_DATA_DOMAINS[domain],
         "report_version": packet.get("report_version"),
@@ -2461,3 +2475,8 @@ def read_full_data_domain(packet: Mapping, domain: str) -> dict:
         "calculation_profile": packet.get("calculation_profile"),
         "worksheets": packet.get("worksheets"),
     })
+    worksheets = copied.get("worksheets")
+    timing = worksheets.get("timing_and_predictive_systems") if isinstance(worksheets, dict) else None
+    if isinstance(timing, dict):
+        timing.pop("pyjhora_annual_method_reference", None)
+    return copied
diff --git a/scripts/rectification_three_engine_packet.py b/scripts/rectification_three_engine_packet.py
index da266d31..4f508782 100644
--- a/scripts/rectification_three_engine_packet.py
+++ b/scripts/rectification_three_engine_packet.py
@@ -1,7 +1,6 @@
 """Build a privacy-safe, request-level three-engine rectification parity packet."""
 from __future__ import annotations
 
-import importlib
 import os
 import sys
 from datetime import datetime
@@ -45,12 +44,12 @@ def _local_d1(case: dict[str, Any]) -> dict[str, str]:
 
 
 def _pyjhora_d1(case: dict[str, Any]) -> dict[str, str]:
-    """PyJHora D1 signs. The jhora import stays in the reference module."""
-    try:
-        reference = importlib.import_module("pyjhora_rectification_d1_reference")
-    except ModuleNotFoundError:
-        reference = importlib.import_module("scripts.pyjhora_rectification_d1_reference")
-    return reference.pyjhora_d1(case)
+    """Product path keeps today's missing-install row and does not load PyJHora.
+
+    ``case`` stays in the signature so the packet caller is unchanged.
+    """
+    del case
+    raise ModuleNotFoundError("No module named 'jhora'")
 
 
 def _ensure_jyotishganit_data_dir() -> str:
diff --git a/scripts/report_render/annual_method_reference.py b/scripts/report_render/annual_method_reference.py
new file mode 100644
index 00000000..bb1ca2c0
--- /dev/null
+++ b/scripts/report_render/annual_method_reference.py
@@ -0,0 +1,542 @@
+"""Parent side of the full-data annual method reference.
+
+This module does not import PyJHora. The worker runs in a child process.
+The worker module name appears only as the ``-m`` argument below.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import json
+import logging
+import subprocess
+import sys
+from datetime import datetime, timedelta
+from pathlib import Path
+from typing import Any, Callable
+
+log = logging.getLogger("jyotisha.annual_method_reference")
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+WORKER_MODULE = "scripts.report_render.pyjhora_annual_reference_worker"
+# Windows measured about 50s for three years. 45s killed a finished call.
+# Linux is expected to be faster. 120s still leaves the report under the 180s wait.
+DEFAULT_TIMEOUT_SECONDS = 120
+TABLES = ("mudda_birth", "mudda_annual", "patyayini", "tajika")
+CHAPTER_ZH = "独立年度方法参考"
+CHAPTER_EN = "Independent Annual Method Reference"
+DASHA_LORDS = {
+    0: "Sun",
+    1: "Moon",
+    2: "Mars",
+    3: "Mercury",
+    4: "Jupiter",
+    5: "Venus",
+    6: "Saturn",
+    7: "Rahu",
+    8: "Ketu",
+}
+YOGA_NAMES = {
+    "ithasala": "Ithasala",
+    "eesarpha": "Esharpha",
+    "manahoo": "Manahoo",
+    "kamboola": "Kamboola",
+    "gairi_kamboola": "Gairi-Kamboola",
+    "khallasara": "Khallsara",
+    "radda": "Radda",
+    "nakta": "Nakta",
+    "yamaya": "Yamaya",
+}
+YOGA_ORDER = (
+    "ithasala",
+    "eesarpha",
+    "manahoo",
+    "kamboola",
+    "gairi_kamboola",
+    "khallasara",
+    "radda",
+    "nakta",
+    "yamaya",
+)
+ZH_PLANETS = {
+    "Sun": "太阳",
+    "Moon": "月亮",
+    "Mars": "火星",
+    "Mercury": "水星",
+    "Jupiter": "木星",
+    "Venus": "金星",
+    "Saturn": "土星",
+    "Rahu": "罗睺",
+    "Ketu": "计都",
+    "Lagna": "上升",
+}
+
+_CACHE: dict[str, dict[str, Any]] = {}
+_RUNNER: Callable[[dict[str, Any], int], dict[str, Any]] | None = None
+
+
+def set_reference_runner(runner: Callable[[dict[str, Any], int], dict[str, Any]] | None) -> None:
+    """Tests and the 200-chart sweep install a stand-in. ``None`` restores the child process."""
+    global _RUNNER
+    _RUNNER = runner
+
+
+def reset_reference_cache() -> None:
+    _CACHE.clear()
+
+
+def strip_annual_method_chapter(markdown: str) -> str:
+    """Drop this chapter so older full-text checks still see the previous report."""
+    lines = (markdown or "").splitlines(keepends=True)
+    start = None
+    for index, line in enumerate(lines):
+        title = line.strip()
+        if title in {f"## {CHAPTER_ZH}", f"## {CHAPTER_EN}"}:
+            start = index
+            break
+    if start is None:
+        return markdown or ""
+    body = "".join(lines[:start]).rstrip()
+    return (body + "\n") if body else ""
+
+
+def _request_hash(payload: dict[str, Any]) -> str:
+    encoded = json.dumps(payload, ensure_ascii=False, sort_keys=True, separators=(",", ":"))
+    return hashlib.sha256(encoded.encode("utf-8")).hexdigest()
+
+
+def _pick(args: Any, packet: dict[str, Any], name: str) -> Any:
+    value = getattr(args, name, None)
+    if value not in (None, ""):
+        return value
+    birth = packet.get("birth_info") if isinstance(packet.get("birth_info"), dict) else {}
+    if birth.get(name) not in (None, ""):
+        return birth.get(name)
+    profile = packet.get("calculation_profile") if isinstance(packet.get("calculation_profile"), dict) else {}
+    if profile.get(name) not in (None, ""):
+        return profile.get(name)
+    return None
+
+
+def _payload_from(args: Any, packet: dict[str, Any]) -> dict[str, Any] | None:
+    try:
+        year = int(_pick(args, packet, "year"))
+        target_year = int(_pick(args, packet, "target_year") or year)
+        birth = {
+            "name": "request",
+            "year": year,
+            "month": int(_pick(args, packet, "month")),
+            "day": int(_pick(args, packet, "day")),
+            "hour": int(_pick(args, packet, "hour") or 0),
+            "minute": int(_pick(args, packet, "minute") or 0),
+            "second": int(_pick(args, packet, "second") or 0),
+            "lat": float(_pick(args, packet, "lat")),
+            "lon": float(_pick(args, packet, "lon")),
+            "tz": float(_pick(args, packet, "tz")),
+        }
+    except (TypeError, ValueError):
+        return None
+    ayanamsa = _pick(args, packet, "ayanamsa") or "lahiri"
+    node_mode = _pick(args, packet, "node_mode") or "mean"
+    return {
+        "birth": birth,
+        "ayanamsa": str(ayanamsa),
+        "node_mode": str(node_mode),
+        "target_year": target_year,
+    }
+
+
+def _invoke_worker(payload: dict[str, Any], timeout: int) -> dict[str, Any]:
+    if _RUNNER is not None:
+        return _RUNNER(payload, timeout)
+    completed = subprocess.run(
+        [sys.executable, "-m", WORKER_MODULE],
+        input=json.dumps(payload, ensure_ascii=False),
+        capture_output=True,
+        text=True,
+        timeout=timeout,
+        check=False,
+        cwd=str(REPO_ROOT),
+    )
+    if completed.returncode != 0:
+        log.warning(
+            "annual method reference worker exit %s: %s",
+            completed.returncode,
+            (completed.stderr or "")[-500:],
+        )
+        return {"ok": False, "reason": f"exit_{completed.returncode}"}
+    text = (completed.stdout or "").strip()
+    try:
+        parsed = json.loads(text)
+    except json.JSONDecodeError:
+        parsed = None
+        for line in reversed(text.splitlines()):
+            if line.strip().startswith("{"):
+                try:
+                    parsed = json.loads(line)
+                    break
+                except json.JSONDecodeError:
+                    continue
+        if parsed is None:
+            log.warning("annual method reference worker returned invalid json")
+            return {"ok": False, "reason": "invalid_json"}
+    if not isinstance(parsed, dict):
+        log.warning("annual method reference worker returned %s", type(parsed).__name__)
+        return {"ok": False, "reason": "invalid_payload"}
+    if completed.stderr and not parsed.get("ok"):
+        log.warning("annual method reference worker stderr: %s", completed.stderr[-500:])
+    return parsed
+
+
+def _year_shell(year: int, *, ayanamsa: str | None = None, node_header: str = "default_true") -> dict[str, Any]:
+    return {
+        "requested_year": year,
+        "ayanamsa": ayanamsa,
+        "ayanamsa_ok": False,
+        "node_header": node_header,
+        "patyayini_anchor": None,
+        "tajika_anchor": None,
+        "tables": {name: {"ok": False, "rows": []} for name in TABLES},
+    }
+
+
+def _unavailable(target_year: int) -> dict[str, Any]:
+    return {"years": [_year_shell(target_year + offset) for offset in range(3)]}
+
+
+def _lord_label(value: Any) -> str:
+    if isinstance(value, str) and value.strip().upper() == "L":
+        return "Lagna"
+    try:
+        return str(DASHA_LORDS.get(int(value), value))
+    except (TypeError, ValueError):
+        return str(value)
+
+
+def _display_clock(start: Any) -> str | None:
+    if not isinstance(start, (list, tuple)) or len(start) != 4:
+        return None
+    try:
+        year, month, day = (int(part) for part in start[:3])
+        hour_decimal = float(start[3])
+        stamp = datetime(year, month, day) + timedelta(hours=hour_decimal)
+        return stamp.strftime("%Y-%m-%d %H:%M:%S")
+    except (TypeError, ValueError, OverflowError):
+        return None
+
+
+def _duration_text(value: Any) -> str:
+    try:
+        number = float(value)
+    except (TypeError, ValueError):
+        return ""
+    text = f"{number:.4f}".rstrip("0").rstrip(".")
+    return text
+
+
+def _period_rows(periods: Any, *, with_end: bool) -> list[dict[str, Any]]:
+    rows: list[dict[str, Any]] = []
+    if not isinstance(periods, list):
+        return rows
+    for order, period in enumerate(periods, start=1):
+        if not isinstance(period, (list, tuple)) or len(period) < 3:
+            continue
+        lords, start, duration = period[0], period[1], period[2]
+        if not isinstance(lords, (list, tuple)) or len(lords) < 2:
+            continue
+        display = _display_clock(start)
+        if display is None:
+            continue
+        rows.append({
+            "order": order,
+            "main_lord": _lord_label(lords[0]),
+            "sub_lord": _lord_label(lords[1]),
+            "start": display,
+            "end": None,
+            "duration": _duration_text(duration),
+        })
+    if with_end:
+        for index, row in enumerate(rows):
+            following = rows[index + 1] if index + 1 < len(rows) else None
+            row["end"] = following["start"] if following else None
+    return rows
+
+
+def _anchor_text(anchor: Any) -> str | None:
+    if not isinstance(anchor, dict):
+        return None
+    date = anchor.get("date")
+    time = anchor.get("time")
+    if not isinstance(date, list) or len(date) != 3 or not isinstance(time, str):
+        return None
+    try:
+        return f"{int(date[0]):04d}-{int(date[1]):02d}-{int(date[2]):02d} {time}"
+    except (TypeError, ValueError):
+        return None
+
+
+def _planet_name(value: Any) -> str:
+    return _lord_label(value)
+
+
+def _relationship_groups(raw: Any) -> list[list[str]]:
+    if raw in (None, []):
+        return []
+    if isinstance(raw, (list, tuple)):
+        if raw and all(not isinstance(item, (list, tuple, dict)) for item in raw):
+            return [[_planet_name(item) for item in raw]]
+        groups = []
+        for item in raw:
+            if isinstance(item, (list, tuple)):
+                groups.append([_planet_name(part) for part in item])
+            else:
+                groups.append([_planet_name(item)])
+        return groups
+    return [[_planet_name(raw)]]
+
+
+def _tajika_rows(raw: Any) -> list[dict[str, Any]]:
+    yogas = raw.get("yogas") if isinstance(raw, dict) and isinstance(raw.get("yogas"), dict) else {}
+    failures = raw.get("failures") if isinstance(raw, dict) and isinstance(raw.get("failures"), dict) else {}
+    rows = []
+    for name in YOGA_ORDER:
+        if name in failures:
+            state = "unavailable"
+            groups: list[list[str]] = []
+        elif name not in yogas:
+            state = "unavailable"
+            groups = []
+        elif yogas[name] is None:
+            state = "unknown"
+            groups = []
+        elif yogas[name] == []:
+            state = "empty"
+            groups = []
+        else:
+            state = "returned"
+            groups = _relationship_groups(yogas[name])
+        rows.append({
+            "yoga": YOGA_NAMES.get(name, name.title()),
+            "state": state,
+            "relationships": groups,
+        })
+    return rows
+
+
+def _normalize_year(raw: dict[str, Any], *, ayanamsa: str | None, ayanamsa_ok: bool, node_header: str) -> dict[str, Any]:
+    year = int(raw.get("requested_year"))
+    shell = _year_shell(year, ayanamsa=ayanamsa, node_header=node_header)
+    shell["ayanamsa_ok"] = ayanamsa_ok
+    if not ayanamsa_ok:
+        return shell
+    anchor = _anchor_text(raw.get("anchor"))
+    shell["patyayini_anchor"] = anchor
+    shell["tajika_anchor"] = anchor
+    mudda_birth = _period_rows(raw.get("mudda_birth"), with_end=False)
+    mudda_annual = _period_rows(raw.get("mudda_annual"), with_end=False)
+    patyayini = _period_rows(raw.get("patyayini"), with_end=True)
+    tajika = _tajika_rows(raw.get("tajika")) if isinstance(raw.get("tajika"), dict) else []
+    shell["tables"] = {
+        "mudda_birth": {"ok": bool(mudda_birth) and "mudda_birth_error" not in raw, "rows": mudda_birth},
+        "mudda_annual": {"ok": bool(mudda_annual) and "annual_return_error" not in raw and "annual_chart_error" not in raw, "rows": mudda_annual},
+        "patyayini": {"ok": bool(patyayini) and "patyayini_error" not in raw, "rows": patyayini},
+        "tajika": {"ok": bool(tajika) and "annual_chart_error" not in raw, "rows": tajika},
+    }
+    return shell
+
+
+def normalize_worker_payload(payload: dict[str, Any], target_year: int) -> dict[str, Any]:
+    """Turn the child JSON into reader rows. Failure text stays out of the packet."""
+    if not isinstance(payload, dict) or not payload.get("ok") or not payload.get("ayanamsa_ok"):
+        reference = _unavailable(target_year)
+        if isinstance(payload, dict):
+            reference["ayanamsa_requested"] = payload.get("ayanamsa_requested")
+            reference["node_header"] = payload.get("node_header") or "default_true"
+            for year in reference["years"]:
+                year["node_header"] = reference["node_header"]
+                year["ayanamsa"] = payload.get("ayanamsa_readback") or payload.get("ayanamsa_requested")
+        return reference
+    years = []
+    for raw in payload.get("years") or []:
+        if isinstance(raw, dict):
+            years.append(_normalize_year(
+                raw,
+                ayanamsa=payload.get("ayanamsa_readback"),
+                ayanamsa_ok=True,
+                node_header=str(payload.get("node_header") or "default_true"),
+            ))
+    if len(years) != 3:
+        return _unavailable(target_year)
+    return {"years": years}
+
+
+def build_pyjhora_annual_method_reference(args: Any, packet: dict[str, Any], *, timeout: int = DEFAULT_TIMEOUT_SECONDS) -> dict[str, Any]:
+    payload = _payload_from(args, packet if isinstance(packet, dict) else {})
+    if payload is None:
+        return _unavailable(0)
+    key = _request_hash(payload)
+    if _RUNNER is None and key in _CACHE:
+        return _CACHE[key]
+    try:
+        raw = _invoke_worker(payload, timeout)
+    except subprocess.TimeoutExpired:
+        log.warning("annual method reference worker timed out after %s seconds", timeout)
+        raw = {"ok": False, "reason": "timeout"}
+    except Exception as exc:
+        log.warning("annual method reference worker could not start: %s", exc.__class__.__name__)
+        raw = {"ok": False, "reason": exc.__class__.__name__}
+    reference = normalize_worker_payload(raw, int(payload["target_year"]))
+    if _RUNNER is None and raw.get("ok"):
+        _CACHE[key] = reference
+    return reference
+
+
+def attach_pyjhora_annual_method_reference(packet: dict[str, Any], args: Any) -> dict[str, Any]:
+    """Store the chapter under the new packet key. Existing keys stay untouched."""
+    if not isinstance(packet, dict):
+        return packet
+    worksheets = packet.get("worksheets")
+    timing = worksheets.get("timing_and_predictive_systems") if isinstance(worksheets, dict) else None
+    if not isinstance(timing, dict):
+        return packet
+    timing["pyjhora_annual_method_reference"] = build_pyjhora_annual_method_reference(args, packet)
+    return packet
+
+
+def _copy(packet: dict[str, Any]) -> dict[str, Any]:
+    worksheets = packet.get("worksheets") if isinstance(packet.get("worksheets"), dict) else {}
+    timing = worksheets.get("timing_and_predictive_systems") if isinstance(worksheets.get("timing_and_predictive_systems"), dict) else {}
+    reference = timing.get("pyjhora_annual_method_reference")
+    return reference if isinstance(reference, dict) else {}
+
+
+def _planet(name: str, language: str) -> str:
+    if language == "zh":
+        return ZH_PLANETS.get(name, name)
+    return name
+
+
+def _relationships(groups: list[list[str]], language: str) -> str:
+    rendered = []
+    for group in groups:
+        rendered.append("–".join(_planet(name, language) for name in group))
+    return "; ".join(item for item in rendered if item)
+
+
+def _state_label(state: str, language: str) -> str:
+    labels = {
+        "returned": ("Combination returned; conditions still need checking", "已返回组合，需核对条件"),
+        "empty": ("No combination detected", "检测为空"),
+        "unknown": ("Unknown", "未知"),
+        "unavailable": ("Result unavailable", "未获得结果"),
+    }
+    pair = labels.get(state, labels["unavailable"])
+    return pair[1] if language == "zh" else pair[0]
+
+
+def _table(headers: list[str], rows: list[list[str]]) -> str:
+    head = "| " + " | ".join(headers) + " |"
+    rule = "| " + " | ".join("---" for _ in headers) + " |"
+    body = ["| " + " | ".join(row) + " |" for row in rows]
+    return "\n".join([head, rule, *body])
+
+
+def _one_year(year: dict[str, Any], language: str) -> str:
+    requested = year.get("requested_year")
+    if language == "zh":
+        lines = [f"### 年度方法补充数据（请求年份 {requested}）", "", "来源：PyJHora。本节数字由 PyJHora 计算，与前面的原生周期分开读。"]
+        node = {"mean": "平交点", "true": "真交点"}.get(str(year.get("node_header")), "PyJHora 默认（真交点）")
+        ayanamsa = year.get("ayanamsa") or "未能按请求设定"
+        lines.append(f"岁差：{ayanamsa}。交点：{node}。")
+        lines.append(f"Patyayini 回归锚点：{year.get('patyayini_anchor') or '未取得'}")
+        lines.append(f"Tajika 回归锚点：{year.get('tajika_anchor') or '未取得'}")
+        headings = {
+            "mudda_birth": "#### Mudda（出生时月亮起算）",
+            "mudda_annual": "#### Mudda（年盘月亮起算）",
+            "patyayini": "#### Patyayini",
+            "tajika": "#### Tajika 瑜伽",
+        }
+        notes = {
+            "mudda_birth": "时长单位是天。这一表没有结束时间。",
+            "mudda_annual": "时长单位是天。这一表没有结束时间。",
+            "patyayini": "时长单位是年。结束时间是下一段的开始。最后一段没有结束时间。",
+            "tajika": "空的组合写检测为空。没有返回结果时写未获得结果。",
+        }
+        period_headers = ["序号", "大运主星", "小运主星", "开始", "时长（天）"]
+        patyayini_headers = ["序号", "大运主星", "小运主星", "开始", "结束", "时长（年）"]
+        yoga_headers = ["瑜伽", "结果", "行星关系"]
+        missing = "未获得结果"
+    else:
+        lines = [
+            f"### Supplementary Annual Method Data (requested year {requested})",
+            "",
+            "Source: PyJHora. These figures are calculated by PyJHora and are read separately from the native cycles above.",
+        ]
+        node = {"mean": "mean", "true": "true"}.get(str(year.get("node_header")), "PyJHora default (true node)")
+        ayanamsa = year.get("ayanamsa") or "could not be set to the requested mode"
+        lines.append(f"Ayanamsa: {ayanamsa}. Nodes: {node}.")
+        lines.append(f"Patyayini return anchor: {year.get('patyayini_anchor') or 'Not obtained'}")
+        lines.append(f"Tajika return anchor: {year.get('tajika_anchor') or 'Not obtained'}")
+        headings = {
+            "mudda_birth": "#### Mudda (Moon at birth)",
+            "mudda_annual": "#### Mudda (Moon in annual chart)",
+            "patyayini": "#### Patyayini",
+            "tajika": "#### Tajika Yoga",
+        }
+        notes = {
+            "mudda_birth": "Duration is in days. This table has no end time.",
+            "mudda_annual": "Duration is in days. This table has no end time.",
+            "patyayini": "Duration is in years. The end is the next row's start. The last row has no end time.",
+            "tajika": "An empty detection is written as no combination. A failed call is written as result unavailable.",
+        }
+        period_headers = ["No.", "MD lord", "AD lord", "Start", "Duration (days)"]
+        patyayini_headers = ["No.", "MD lord", "AD lord", "Start", "End", "Duration (years)"]
+        yoga_headers = ["Yoga", "Result", "Planet relationships"]
+        missing = "Result unavailable"
+    tables = year.get("tables") if isinstance(year.get("tables"), dict) else {}
+    for key in TABLES:
+        lines.extend(["", headings[key], "", notes[key], ""])
+        table = tables.get(key) if isinstance(tables.get(key), dict) else {}
+        rows = table.get("rows") if table.get("ok") and isinstance(table.get("rows"), list) else []
+        if not rows:
+            lines.append(missing)
+            continue
+        if key == "tajika":
+            rendered = [[
+                str(row.get("yoga") or ""),
+                _state_label(str(row.get("state") or ""), language),
+                _relationships(row.get("relationships") or [], language),
+            ] for row in rows if isinstance(row, dict)]
+            lines.append(_table(yoga_headers, rendered))
+            continue
+        headers = patyayini_headers if key == "patyayini" else period_headers
+        rendered = []
+        for row in rows:
+            if not isinstance(row, dict):
+                continue
+            end = row.get("end") or ("—" if language == "zh" else "-")
+            base = [
+                str(row.get("order") or ""),
+                _planet(str(row.get("main_lord") or ""), language),
+                _planet(str(row.get("sub_lord") or ""), language),
+                str(row.get("start") or ""),
+            ]
+            if key == "patyayini":
+                base.append(str(end))
+            base.append(str(row.get("duration") or ""))
+            rendered.append(base)
+        lines.append(_table(headers, rendered))
+    return "\n".join(lines)
+
+
+def render_annual_method_reference(packet: dict[str, Any], language: str) -> str:
+    reference = _copy(packet if isinstance(packet, dict) else {})
+    years = [year for year in reference.get("years") or [] if isinstance(year, dict)]
+    if not years:
+        return ""
+    if language == "zh":
+        intro = "## 独立年度方法参考\n\n本节与前面的原生周期分开使用。数字不替换、不合并前面的年运。"
+    else:
+        intro = "## Independent Annual Method Reference\n\nUse this section separately from the native cycles above. These figures do not replace the annual cycles earlier in the report."
+    body = "\n\n".join(_one_year(year, "en" if language == "en" else "zh") for year in years)
+    return intro + "\n\n" + body + "\n"
diff --git a/scripts/report_render/pyjhora_annual_reference_worker.py b/scripts/report_render/pyjhora_annual_reference_worker.py
new file mode 100644
index 00000000..fa194724
--- /dev/null
+++ b/scripts/report_render/pyjhora_annual_reference_worker.py
@@ -0,0 +1,288 @@
+"""PyJHora annual-method worker. Run only as ``python -m``.
+
+Product modules must not import this file. Importing it loads PyJHora and
+changes the process ayanamsa and ephemeris path.
+"""
+
+from __future__ import annotations
+
+import contextlib
+import io
+import json
+import logging
+import sys
+from typing import Any
+
+log = logging.getLogger("jyotisha.pyjhora_annual_reference_worker")
+
+AYANAMSA_MAP = {
+    "lahiri": "LAHIRI",
+    "raman": "RAMAN",
+    "kp": "KP",
+    "krishnamurti": "KP",
+    "fagan_bradley": "FAGAN",
+    "true_citra": "TRUE_CITRA",
+    "true_pushya": "TRUE_PUSHYA",
+}
+YOGA_CALLABLES = (
+    ("ithasala", "get_ithasala_yoga_planet_pairs"),
+    ("eesarpha", "get_eesarpha_yoga_planet_pairs"),
+    ("manahoo", "get_manahoo_yoga_planet_pairs"),
+    ("kamboola", "get_kamboola_yoga_planet_pairs"),
+    ("gairi_kamboola", "get_gairi_kamboola_yoga_planet_pairs"),
+    ("khallasara", "get_khallasara_yoga_planet_pairs"),
+    ("radda", "get_radda_yoga_planet_pairs"),
+    ("nakta", "get_nakta_yoga_planet_triples"),
+    ("yamaya", "get_yamaya_yoga_planet_triples"),
+)
+
+
+def _jsonify(value: Any) -> Any:
+    if isinstance(value, dict):
+        return {str(key): _jsonify(item) for key, item in value.items()}
+    if isinstance(value, (list, tuple)):
+        return [_jsonify(item) for item in value]
+    if isinstance(value, (str, int, float, bool)) or value is None:
+        return value
+    return str(value)
+
+
+def _anchor(chart_payload: Any) -> dict[str, Any] | None:
+    if not isinstance(chart_payload, (list, tuple)) or len(chart_payload) < 2:
+        return None
+    marker = chart_payload[1]
+    if not isinstance(marker, (list, tuple)) or len(marker) < 2:
+        return None
+    date_value = marker[0]
+    if not isinstance(date_value, (list, tuple)) or len(date_value) != 3:
+        return None
+    return {"date": [int(part) for part in date_value], "time": str(marker[1])}
+
+
+def _jd_from_anchor(utils: Any, anchor: dict[str, Any]) -> float:
+    hour, minute, second = (int(part) for part in str(anchor["time"]).split(":"))
+    return float(utils.julian_day_number(tuple(anchor["date"]), (hour, minute, second)))
+
+
+def _verify_node(const: Any, drik: Any, swe: Any, local_jd: float, timezone: float, requested_true: bool) -> tuple[bool, float | None]:
+    if bool(const._use_true_nodes_for_rahu_ketu) != requested_true:
+        return False, None
+    jd_utc = float(local_jd) - float(timezone) / 24.0
+    observed = float(drik.sidereal_longitude(jd_utc, const._RAHU)) % 360.0
+    flags = swe.FLG_SWIEPH | swe.FLG_SIDEREAL
+    swe_lon = float(swe.calc_ut(jd_utc, int(const._RAHU), flags)[0][0]) % 360.0
+    delta = abs((observed - swe_lon + 180.0) % 360.0 - 180.0)
+    return delta < 1.0, round(observed, 6)
+
+
+def _one_mode(
+    *,
+    mudda: Any,
+    patyayini: Any,
+    tajaka: Any,
+    yoga_module: Any,
+    utils: Any,
+    place: Any,
+    birth_jd: float,
+    age: int,
+    annual_years: int,
+) -> dict[str, Any]:
+    year: dict[str, Any] = {"mudda_birth": None, "mudda_annual": None, "patyayini": None, "tajika": None}
+    try:
+        with contextlib.redirect_stdout(sys.stderr):
+            periods = mudda.mudda_dhasa_bhukthi(birth_jd, place, age)
+        year["mudda_birth"] = _jsonify(periods)
+    except Exception as exc:
+        log.warning("mudda birth-moon failed: %s", exc.__class__.__name__)
+        year["mudda_birth_error"] = exc.__class__.__name__
+
+    chart_payload = None
+    try:
+        with contextlib.redirect_stdout(sys.stderr):
+            chart_payload = tajaka.annual_chart(birth_jd, place, years=annual_years)
+    except Exception as exc:
+        log.warning("annual_chart failed: %s", exc.__class__.__name__)
+        year["annual_chart_error"] = exc.__class__.__name__
+
+    anchor = _anchor(chart_payload)
+    positions = chart_payload[0] if isinstance(chart_payload, (list, tuple)) and chart_payload else None
+    year["anchor"] = anchor
+    if anchor is not None:
+        try:
+            annual_jd = _jd_from_anchor(utils, anchor)
+            with contextlib.redirect_stdout(sys.stderr):
+                annual_periods = mudda.mudda_dhasa_bhukthi(annual_jd, place, 0)
+            year["mudda_annual"] = _jsonify(annual_periods)
+            with contextlib.redirect_stdout(sys.stderr):
+                patyayini_periods = patyayini.get_dhasa_bhukthi(annual_jd, place)
+            year["patyayini"] = _jsonify(patyayini_periods)
+            year["anchor_mode"] = "annual_chart"
+        except Exception as exc:
+            log.warning("annual-return tables failed: %s", exc.__class__.__name__)
+            year["annual_return_error"] = exc.__class__.__name__
+    if year.get("patyayini") is None:
+        try:
+            fallback_jd = float(birth_jd) + float(age) * 365.256364
+            with contextlib.redirect_stdout(sys.stderr):
+                year["patyayini"] = _jsonify(patyayini.get_dhasa_bhukthi(fallback_jd, place))
+            year["anchor_mode"] = "tropical_year_fallback"
+        except Exception as exc:
+            log.warning("patyayini fallback failed: %s", exc.__class__.__name__)
+            year["patyayini_error"] = exc.__class__.__name__
+
+    yogas: dict[str, Any] = {}
+    failures: dict[str, str] = {}
+    if isinstance(positions, (list, tuple)):
+        for yoga_name, callable_name in YOGA_CALLABLES:
+            candidate = getattr(yoga_module, callable_name, None)
+            if not callable(candidate):
+                failures[yoga_name] = "missing"
+                continue
+            try:
+                with contextlib.redirect_stdout(sys.stderr):
+                    yogas[yoga_name] = _jsonify(candidate(positions))
+            except Exception as exc:
+                log.warning("tajika %s failed: %s", yoga_name, exc.__class__.__name__)
+                failures[yoga_name] = exc.__class__.__name__
+        year["tajika"] = {"yogas": yogas, "failures": failures}
+    return year
+
+
+def compute(request: dict[str, Any]) -> dict[str, Any]:
+    import swisseph as swe
+    import jhora.const as const
+
+    birth = request.get("birth") if isinstance(request.get("birth"), dict) else {}
+    timezone = float(birth.get("tz") or 0)
+    clock = float(birth.get("hour") or 0) + float(birth.get("minute") or 0) / 60.0 + float(birth.get("second") or 0) / 3600.0
+    local_jd = float(swe.julday(int(birth["year"]), int(birth["month"]), int(birth["day"]), clock))
+    requested_name = str(request.get("ayanamsa") or "").strip().lower()
+    mode = AYANAMSA_MAP.get(requested_name)
+    node_mode = str(request.get("node_mode") or "mean").strip().lower()
+    requested_true = node_mode == "true"
+    target_year = int(request["target_year"])
+    birth_year = int(birth["year"])
+    result: dict[str, Any] = {
+        "ok": True,
+        "ayanamsa_requested": mode,
+        "ayanamsa_readback": None,
+        "ayanamsa_ok": False,
+        "node_applied": False,
+        "node_header": "default_true",
+        "years": [],
+    }
+    if mode is None:
+        log.warning("unsupported ayanamsa for the annual method reference")
+        return result
+
+    const.set_node_mode(requested_true)
+    from jhora.panchanga import drik
+    from jhora.horoscope.dhasa.annual import mudda, patyayini
+    from jhora.horoscope.transit import tajaka, tajaka_yoga
+    from jhora import utils
+
+    node_ok, node_longitude = _verify_node(const, drik, swe, local_jd, timezone, requested_true)
+    result["node_longitude"] = node_longitude
+    if node_ok:
+        result["node_applied"] = True
+        result["node_header"] = "true" if requested_true else "mean"
+    else:
+        log.warning("node mode readback failed; leaving PyJHora's default true node")
+        const.set_node_mode(True)
+        result["node_header"] = "default_true"
+
+    with contextlib.redirect_stdout(sys.stderr):
+        drik.set_ayanamsa_mode(mode)
+    readback = str(const._DEFAULT_AYANAMSA_MODE).upper()
+    result["ayanamsa_readback"] = readback
+    result["ayanamsa_ok"] = readback == mode
+    if not result["ayanamsa_ok"]:
+        log.warning("ayanamsa readback %s did not match %s", readback, mode)
+        return result
+
+    place = drik.Place(str(birth.get("name") or "request"), float(birth["lat"]), float(birth["lon"]), timezone, 0.0)
+    prove_reuse = bool(request.get("prove_reuse"))
+    include_upstream = bool(request.get("include_upstream_diff"))
+    ut_jd = local_jd - timezone / 24.0
+    for offset in range(3):
+        year = target_year + offset
+        age = year - birth_year
+        documented = _one_mode(
+            mudda=mudda,
+            patyayini=patyayini,
+            tajaka=tajaka,
+            yoga_module=tajaka_yoga,
+            utils=utils,
+            place=place,
+            birth_jd=local_jd,
+            age=age,
+            annual_years=age + 1,
+        )
+        documented["requested_year"] = year
+        documented["age"] = age
+        if prove_reuse:
+            try:
+                with contextlib.redirect_stdout(sys.stderr):
+                    again = tajaka.annual_chart(local_jd, place, years=age + 1)
+                documented["reuse_matches"] = _anchor(again) == documented.get("anchor")
+            except Exception as exc:
+                log.warning("reuse check failed: %s", exc.__class__.__name__)
+                documented["reuse_matches"] = False
+        if include_upstream and offset == 0:
+            upstream = _one_mode(
+                mudda=mudda,
+                patyayini=patyayini,
+                tajaka=tajaka,
+                yoga_module=tajaka_yoga,
+                utils=utils,
+                place=place,
+                birth_jd=ut_jd,
+                age=age,
+                annual_years=age,
+            )
+            documented["upstream_diff"] = {
+                "documented_anchor": documented.get("anchor"),
+                "upstream_anchor": upstream.get("anchor"),
+                "documented_mudda_start": _first_start(documented.get("mudda_birth")),
+                "upstream_mudda_start": _first_start(upstream.get("mudda_birth")),
+            }
+        result["years"].append(documented)
+    return result
+
+
+def _first_start(periods: Any) -> Any:
+    if not isinstance(periods, list) or not periods:
+        return None
+    first = periods[0]
+    if isinstance(first, list) and len(first) >= 2:
+        return first[1]
+    return None
+
+
+def main() -> int:
+    logging.basicConfig(stream=sys.stderr, level=logging.WARNING)
+    try:
+        request = json.loads(sys.stdin.read() or "{}")
+    except json.JSONDecodeError as exc:
+        log.warning("worker stdin was not json: %s", exc.__class__.__name__)
+        sys.stdout.write(json.dumps({"ok": False, "reason": "invalid_json"}))
+        return 0
+    if not isinstance(request, dict):
+        sys.stdout.write(json.dumps({"ok": False, "reason": "invalid_request"}))
+        return 0
+    buffer = io.StringIO()
+    try:
+        with contextlib.redirect_stdout(buffer):
+            payload = compute(request)
+    except Exception as exc:
+        log.warning("worker failed: %s", exc.__class__.__name__)
+        payload = {"ok": False, "reason": exc.__class__.__name__}
+    leaked = buffer.getvalue()
+    if leaked:
+        print(leaked, file=sys.stderr)
+    sys.stdout.write(json.dumps(payload, ensure_ascii=False))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/scripts/unified_consultation_orchestrator.py b/scripts/unified_consultation_orchestrator.py
index 27b945b1..cb5bfa0e 100644
--- a/scripts/unified_consultation_orchestrator.py
+++ b/scripts/unified_consultation_orchestrator.py
@@ -41,9 +41,9 @@ def _formal_divisions() -> tuple[int, ...]:
 FORMAL_DIVISIONS = _formal_divisions()
 
 try:
-    from diagnose_pyjhora_adapter import build_report as build_pyjhora_adapter_report
+    from diagnose_pyjhora_adapter import build_product_report as build_pyjhora_adapter_report
 except Exception:  # pragma: no cover - import path varies in tests/CLI
-    from scripts.diagnose_pyjhora_adapter import build_report as build_pyjhora_adapter_report
+    from scripts.diagnose_pyjhora_adapter import build_product_report as build_pyjhora_adapter_report
 
 try:
     from diagnose_jyotishganit_adapter import build_report as build_jyotishganit_adapter_report
diff --git a/tests/report_english_fictional.py b/tests/report_english_fictional.py
index 78ef53a7..d743bf15 100644
--- a/tests/report_english_fictional.py
+++ b/tests/report_english_fictional.py
@@ -235,6 +235,16 @@ def _guard_snapshot() -> dict:
     }
 
 
+def _install_annual_reference_stub() -> None:
+    """The 200-chart sweep must not start one PyJHora child per chart."""
+    from scripts.report_render.annual_method_reference import set_reference_runner
+
+    def _unavailable(_payload, _timeout):
+        return {"ok": False, "reason": "sweep_stub"}
+
+    set_reference_runner(_unavailable)
+
+
 def _prepare_process() -> None:
     root = Path(__file__).resolve().parents[1]
     for entry in (str(root), str(root / "scripts")):
@@ -242,6 +252,7 @@ def _prepare_process() -> None:
             sys.path.insert(0, entry)
     block_jhora()
     _install_replay_stubs()
+    _install_annual_reference_stub()
 
 
 def init_english_worker() -> None:
diff --git a/tests/test_api_chart_subprocess_gate.py b/tests/test_api_chart_subprocess_gate.py
index 1e1ae8cb..d5001512 100644
--- a/tests/test_api_chart_subprocess_gate.py
+++ b/tests/test_api_chart_subprocess_gate.py
@@ -73,6 +73,9 @@ def test_no_jhora_annual_skips_subprocess_and_matches_golden(monkeypatch) -> Non
 
 
 def test_jhora_available_still_uses_subprocess(monkeypatch) -> None:
+    # 原值：_jhora_importable 为真时两个隔离函数各起一次子进程（calls == 2）
+    # 新值：产品调用不传 research 时，即使可导入也不起子进程、返回 golden；传 research=True 时仍起子进程（calls == 2）
+    # 原因：#93 红线 1：装包后旧回放不得进年运章
     reset_jhora_import_cache()
     monkeypatch.setattr("scripts.annual_pyjhora_replay._jhora_importable", lambda: True)
     calls = {"n": 0}
@@ -85,12 +88,46 @@ def test_jhora_available_still_uses_subprocess(monkeypatch) -> None:
         calls["n"] += 1
         return _Completed()
 
-    monkeypatch.setattr("scripts.annual_pyjhora_replay.subprocess.run", _fake_run)
+    def _forbid_subprocess(*_args, **_kwargs):
+        raise AssertionError("product replay must not spawn a child")
+
+    monkeypatch.setattr("scripts.annual_pyjhora_replay.subprocess.run", _forbid_subprocess)
     tajaka = _run_isolated_tajaka_year_lord(_PROFILE, 2447893.0, _Place(), 30)
     annual = _run_isolated_annual_replay(_PROFILE, 2447893.0, _Place(), 30)
+    assert calls["n"] == 0
+    tajaka_golden = json.loads((FIXTURES / "tajaka-no-jhora.json").read_text(encoding="utf-8"))
+    annual_golden = json.loads((FIXTURES / "annual-no-jhora.json").read_text(encoding="utf-8"))
+    assert _without_hash(tajaka) == tajaka_golden
+    assert _without_hash(annual) == annual_golden
+
+    monkeypatch.setattr("scripts.annual_pyjhora_replay.subprocess.run", _fake_run)
+    _run_isolated_tajaka_year_lord(_PROFILE, 2447893.0, _Place(), 30, research=True)
+    _run_isolated_annual_replay(_PROFILE, 2447893.0, _Place(), 30, research=True)
     assert calls["n"] == 2
-    assert tajaka["request_hash"]
-    assert annual["request_hash"]
+
+
+def test_product_replay_calls_a_four_arg_stand_in(monkeypatch) -> None:
+    """The 200-chart sweep replaces the runners with four-argument functions."""
+    from scripts.annual_pyjhora_replay import (
+        _no_jhora_annual_replay,
+        _no_jhora_tajaka_replay,
+        replay_pyjhora_annual,
+    )
+
+    calls = []
+
+    def blocked(profile, birth_julian_day, place, age):
+        calls.append(age)
+        settings = {"ayanamsa": "LAHIRI", "node_mode": "mean"}
+        result = _no_jhora_tajaka_replay(settings) if len(calls) == 1 else _no_jhora_annual_replay(settings)
+        result["request_hash"] = "stand-in"
+        return result
+
+    monkeypatch.setattr("scripts.annual_pyjhora_replay._run_isolated_tajaka_year_lord", blocked)
+    monkeypatch.setattr("scripts.annual_pyjhora_replay._run_isolated_annual_replay", blocked)
+    result = replay_pyjhora_annual(_PROFILE, birth_julian_day=2447893.0, place=_Place(), age=30)
+    assert calls == [30, 30]
+    assert result["status"] == "blocked"
 
 
 def test_pyjhora_interpreter_lookup_is_cached(monkeypatch) -> None:
diff --git a/tests/test_full_data_parity_chapters.py b/tests/test_full_data_parity_chapters.py
index cf5a0e4b..4c12a581 100644
--- a/tests/test_full_data_parity_chapters.py
+++ b/tests/test_full_data_parity_chapters.py
@@ -171,8 +171,8 @@ def test_three_fictional_charts_render_the_same_chapters(fictional_reports: dict
         assert chinese_prose_gates(chinese) == {"english_signs": 0, "han_function_glue": 0}
         assert HAN.findall(english) == []
         assert "[translation missing:" not in english
-        zh = _chapter_text(chinese, "zh")
-        en = _chapter_text(english, "en")
+        zh = _chapter_text(chinese, "zh").split("\n## 独立年度方法参考\n", 1)[0]
+        en = _chapter_text(english, "en").split("\n## Independent Annual Method Reference\n", 1)[0]
         zh_at = 0
         en_at = 0
         for title_en, title_zh in SECTIONS:
@@ -335,20 +335,29 @@ def test_six_chapters_are_reader_chapters_and_sections_sit_below_them(monkeypatc
     full_zh = render_full_data_markdown(copy.deepcopy(packet))
     full_en = _english_edition(copy.deepcopy(packet), True).get("markdown_en") or ""
     monkeypatch.setattr(export, "_append_parity_chapters", lambda markdown, _packet, _language: markdown)
+    # 原值：拿掉六章后，二级标题与完整报告只差这六章
+    # 新值：基准同时拿掉六章和独立年度方法参考；完整报告在六章之后还有这一章
+    # 原因：#93 的年度方法参考也是后加的二级标题
+    monkeypatch.setattr(export, "_append_annual_method_reference", lambda markdown, _packet, _language: markdown)
     base_zh = render_full_data_markdown(copy.deepcopy(packet))
     base_en = _english_edition(copy.deepcopy(packet), True).get("markdown_en") or ""
     assert base_en
-    for full, base, chapters in ((full_zh, base_zh, zh_chapters), (full_en, base_en, en_chapters)):
+    for full, base, chapters, extra in (
+        (full_zh, base_zh, zh_chapters, "独立年度方法参考"),
+        (full_en, base_en, en_chapters, "Independent Annual Method Reference"),
+    ):
         h1 = _heading_lines(full, 1)
         assert len(h1) == 1, h1
         assert full.startswith("# " + h1[0])
         assert _heading_lines(base, 1) == h1
         full_h2 = _heading_lines(full, 2)
         base_h2 = _heading_lines(base, 2)
-        assert len(full_h2) == len(base_h2) + len(chapters)
+        assert len(full_h2) == len(base_h2) + len(chapters) + 1
         assert full_h2[: len(base_h2)] == base_h2
-        assert full_h2[len(base_h2):] == chapters
+        assert full_h2[len(base_h2): len(base_h2) + len(chapters)] == chapters
+        assert full_h2[-1] == extra
         tail = full.split("\n## " + chapters[0] + "\n", 1)[1]
+        tail = tail.split("\n## " + extra + "\n", 1)[0]
         assert _heading_lines(tail, 3)[: len(_chapter_sections()[0])] == [
             (dict(SECTIONS)[title] if chapters is zh_chapters else title) for title in _chapter_sections()[0]
         ]
diff --git a/tests/test_pyjhora_annual_method_reference.py b/tests/test_pyjhora_annual_method_reference.py
new file mode 100644
index 00000000..4e374139
--- /dev/null
+++ b/tests/test_pyjhora_annual_method_reference.py
@@ -0,0 +1,318 @@
+"""Issue #93: PyJHora annual method reference stays in one child process."""
+
+from __future__ import annotations
+
+import json
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+
+from scripts.diagnose_pyjhora_adapter import build_product_report
+from scripts.pl9_full_data_export import read_full_data_domain, render_full_data_markdown
+from scripts.rectification_three_engine_packet import _pyjhora_d1
+from scripts.report_render.annual_method_reference import (
+    CHAPTER_EN,
+    CHAPTER_ZH,
+    build_pyjhora_annual_method_reference,
+    normalize_worker_payload,
+    render_annual_method_reference,
+    reset_reference_cache,
+    set_reference_runner,
+    strip_annual_method_chapter,
+)
+from tests.report_english_fictional import fictional_case
+
+ROOT = Path(__file__).resolve().parents[1]
+
+
+def setup_function() -> None:
+    reset_reference_cache()
+    set_reference_runner(None)
+
+
+def test_result_hash_ignores_the_annual_method_chapter() -> None:
+    from scripts.jyotish_engine import _bind_result_without_annual_method_reference
+
+    profile = {"input_hash": "ab" * 32}
+
+    def packet(marker: str) -> dict:
+        return {
+            "value": 1,
+            "worksheets": {
+                "timing_and_predictive_systems": {
+                    "kept": "same",
+                    "pyjhora_annual_method_reference": {"marker": marker},
+                }
+            },
+        }
+
+    installed = _bind_result_without_annual_method_reference(packet("installed"), profile)
+    blocked = _bind_result_without_annual_method_reference(packet("blocked"), profile)
+    assert installed["result_hash"] == blocked["result_hash"]
+    assert installed["worksheets"]["timing_and_predictive_systems"]["pyjhora_annual_method_reference"]["marker"] == "installed"
+    assert installed["worksheets"]["timing_and_predictive_systems"]["kept"] == "same"
+
+
+def test_product_report_does_not_probe(monkeypatch) -> None:
+    def _boom(*_args, **_kwargs):
+        raise AssertionError("product diagnostics must not probe jhora")
+
+    monkeypatch.setattr("scripts.diagnose_pyjhora_adapter.importlib.util.find_spec", _boom)
+    report = build_product_report()
+    assert report["status"] == "missing_dependency"
+    assert report["dependency_available"] is False
+    assert report["missing_dependency"] == "jhora"
+
+
+def test_pyjhora_d1_stays_on_the_missing_install_row() -> None:
+    source = (ROOT / "scripts" / "rectification_three_engine_packet.py").read_text(encoding="utf-8")
+    assert "import_module" not in source
+    assert "import jhora" not in source
+    with pytest.raises(ModuleNotFoundError):
+        _pyjhora_d1({"year": 1990})
+
+
+def test_deploy_does_not_set_year_lord_pyjhora_reference() -> None:
+    hits = []
+    for path in (ROOT / "deploy").rglob("*"):
+        if not path.is_file():
+            continue
+        text = path.read_text(encoding="utf-8", errors="ignore")
+        if "JYOTISH_YEAR_LORD_PYJHORA_REFERENCE" in text:
+            hits.append(path.relative_to(ROOT).as_posix())
+    assert hits == []
+
+
+def test_timeout_and_nonzero_exit_leave_every_table_unavailable() -> None:
+    def _timeout(_payload, timeout):
+        raise subprocess.TimeoutExpired(cmd="worker", timeout=timeout)
+
+    set_reference_runner(_timeout)
+    reference = build_pyjhora_annual_method_reference(_args(0), {"birth_info": {}})
+    assert _all_unavailable(reference)
+
+    def _exit(_payload, _timeout):
+        return {"ok": False, "reason": "exit_3"}
+
+    set_reference_runner(_exit)
+    reference = build_pyjhora_annual_method_reference(_args(0), {"birth_info": {}})
+    assert _all_unavailable(reference)
+
+
+def test_normalize_keeps_nine_yogas_and_two_mudda_tables() -> None:
+    raw = {
+        "ok": True,
+        "ayanamsa_ok": True,
+        "ayanamsa_readback": "RAMAN",
+        "node_header": "mean",
+        "years": [{
+            "requested_year": 2026,
+            "anchor": {"date": [2026, 6, 15], "time": "16:35:04"},
+            "mudda_birth": [[[0, 1], [2026, 1, 2, 3.5], 30]],
+            "mudda_annual": [[[2, "L"], [2026, 6, 15, 1], 10]],
+            "patyayini": [
+                [[3, 4], [2026, 6, 15, 0], 0.4],
+                [[5, 6], [2026, 11, 1, 0], 0.6],
+            ],
+            "tajika": {
+                "yogas": {name: [] for name in (
+                    "ithasala", "eesarpha", "manahoo", "kamboola", "nakta", "yamaya",
+                )},
+                "failures": {"radda": "NameError"},
+            },
+        }],
+    }
+    raw["years"][0]["tajika"]["yogas"]["gairi_kamboola"] = None
+    raw["years"][0]["tajika"]["yogas"]["khallasara"] = None
+    second = json.loads(json.dumps(raw["years"][0]))
+    third = json.loads(json.dumps(raw["years"][0]))
+    second["requested_year"] = 2027
+    third["requested_year"] = 2028
+    raw["years"].extend([second, third])
+    reference = normalize_worker_payload(raw, 2026)
+    tables = reference["years"][0]["tables"]
+    assert tables["mudda_birth"]["rows"][0]["main_lord"] == "Sun"
+    assert tables["mudda_annual"]["rows"][0]["sub_lord"] == "Lagna"
+    assert tables["patyayini"]["rows"][0]["end"] == "2026-11-01 00:00:00"
+    assert tables["patyayini"]["rows"][1]["end"] is None
+    states = {row["yoga"]: row["state"] for row in tables["tajika"]["rows"]}
+    assert len(states) == 9
+    assert states["Radda"] == "unavailable"
+    assert states["Gairi-Kamboola"] == "unknown"
+    assert states["Ithasala"] == "empty"
+
+
+def test_chapter_is_last_and_domain_view_omits_the_key() -> None:
+    packet = {
+        "report_language": "zh",
+        "worksheets": {"timing_and_predictive_systems": {"pyjhora_annual_method_reference": {
+            "years": [_shell(2026), _shell(2027), _shell(2028)],
+        }}},
+    }
+    chinese = render_full_data_markdown(packet)
+    english_packet = dict(packet)
+    english_packet["report_language"] = "en"
+    english = render_full_data_markdown(english_packet)
+    assert chinese.rstrip().splitlines()[-1] != ""
+    assert _h2(chinese)[-1] == CHAPTER_ZH
+    assert _h2(english)[-1] == CHAPTER_EN
+    assert chinese.count("来源：PyJHora") == 3
+    assert english.count("Source: PyJHora") == 3
+    assert "未获得结果" in chinese and "Result unavailable" in english
+    assert strip_annual_method_chapter(chinese) == strip_annual_method_chapter(strip_annual_method_chapter(chinese))
+    assert CHAPTER_ZH not in strip_annual_method_chapter(chinese)
+    domain = read_full_data_domain(packet, "career")
+    timing = domain["worksheets"]["timing_and_predictive_systems"]
+    assert "pyjhora_annual_method_reference" not in timing
+
+
+def test_three_fictional_charts_match_a_second_annual_chart_call() -> None:
+    """Real PyJHora, three fictional charts. Birth fields stay out of the assertion text."""
+    script = r"""
+import json, sys
+sys.path[:0] = [sys.argv[1], sys.argv[1] + "\\scripts"]
+from tests.report_english_fictional import fictional_case
+from scripts.report_render.annual_method_reference import _invoke_worker, _payload_from
+class Args:
+    pass
+out = []
+for index in (0, 1, 9):
+    birth = fictional_case(index)
+    args = Args()
+    for key, value in birth.items():
+        setattr(args, key, value)
+    payload = _payload_from(args, {})
+    if index == 0:
+        payload["prove_reuse"] = True
+        payload["include_upstream_diff"] = True
+    raw = _invoke_worker(payload, 120)
+    years = raw.get("years") or []
+    diff = (years[0].get("upstream_diff") if years else None) or {}
+    out.append({
+        "i": index,
+        "ok": bool(raw.get("ok")),
+        "ayanamsa_ok": bool(raw.get("ayanamsa_ok")),
+        "readback": raw.get("ayanamsa_readback"),
+        "node": raw.get("node_header"),
+        "reuse": [bool(year.get("reuse_matches")) for year in years],
+        "documented_anchor": (diff.get("documented_anchor") or {}).get("date"),
+        "upstream_anchor": (diff.get("upstream_anchor") or {}).get("date"),
+        "documented_mudda": diff.get("documented_mudda_start"),
+        "upstream_mudda": diff.get("upstream_mudda_start"),
+        "yoga_keys": sorted(((years[0].get("tajika") or {}).get("yogas") or {}).keys()) if years else [],
+        "radda": ((years[0].get("tajika") or {}).get("failures") or {}).get("radda"),
+        "jhora_in_parent": any(name == "jhora" or name.startswith("jhora.") for name in sys.modules),
+    })
+print(json.dumps(out, ensure_ascii=False))
+"""
+    completed = subprocess.run(
+        [sys.executable, "-c", script, str(ROOT)],
+        cwd=ROOT,
+        text=True,
+        capture_output=True,
+        timeout=400,
+        check=False,
+    )
+    assert completed.returncode == 0, completed.stderr[-1000:]
+    rows = json.loads(completed.stdout)
+    assert [row["i"] for row in rows] == [0, 1, 9]
+    for row in rows:
+        assert row["ok"] and row["ayanamsa_ok"], row
+        assert row["readback"]
+        if row["i"] == 0:
+            assert row["reuse"] == [True, True, True]
+            assert row["documented_anchor"] and row["upstream_anchor"]
+        assert row["jhora_in_parent"] is False
+        assert "radda" not in row["yoga_keys"]
+        assert row["radda"]
+
+
+def test_parent_moon_and_ayanamsa_stay_put() -> None:
+    script = r"""
+import json, sys
+sys.path[:0] = [sys.argv[1], sys.argv[1] + "\\scripts"]
+import swisseph as swe
+from tests.report_english_fictional import fictional_case
+from scripts.report_render.annual_method_reference import build_pyjhora_annual_method_reference
+birth = fictional_case(0)
+class Args:
+    pass
+args = Args()
+for key, value in birth.items():
+    setattr(args, key, value)
+jd = swe.julday(2000, 1, 1, 0)
+before_ay = swe.get_ayanamsa_ut(jd)
+before_moon = swe.calc_ut(jd, swe.MOON)[0][0]
+reference = build_pyjhora_annual_method_reference(args, {})
+after_ay = swe.get_ayanamsa_ut(jd)
+after_moon = swe.calc_ut(jd, swe.MOON)[0][0]
+print(json.dumps({
+    "same_ayanamsa": before_ay == after_ay,
+    "same_moon": before_moon == after_moon,
+    "jhora": any(name == "jhora" or name.startswith("jhora.") for name in sys.modules),
+    "years": len(reference.get("years") or []),
+    "tables": sorted((reference.get("years") or [{}])[0].get("tables", {})),
+}))
+"""
+    completed = subprocess.run(
+        [sys.executable, "-c", script, str(ROOT)],
+        cwd=ROOT,
+        text=True,
+        capture_output=True,
+        timeout=90,
+        check=False,
+    )
+    assert completed.returncode == 0, completed.stderr[-1000:]
+    payload = json.loads(completed.stdout.strip().splitlines()[-1])
+    assert payload["same_ayanamsa"] and payload["same_moon"]
+    assert payload["jhora"] is False
+    assert payload["years"] == 3
+    assert payload["tables"] == ["mudda_annual", "mudda_birth", "patyayini", "tajika"]
+
+
+def _args(index: int):
+    birth = fictional_case(index)
+
+    class Args:
+        pass
+
+    args = Args()
+    for key, value in birth.items():
+        setattr(args, key, value)
+    return args
+
+
+def _shell(year: int) -> dict:
+    return {
+        "requested_year": year,
+        "ayanamsa": "RAMAN",
+        "ayanamsa_ok": False,
+        "node_header": "mean",
+        "patyayini_anchor": None,
+        "tajika_anchor": None,
+        "tables": {name: {"ok": False, "rows": []} for name in ("mudda_birth", "mudda_annual", "patyayini", "tajika")},
+    }
+
+
+def _all_unavailable(reference: dict) -> bool:
+    years = reference.get("years") or []
+    return len(years) == 3 and all(
+        all(not year["tables"][name]["ok"] for name in year["tables"])
+        for year in years
+    )
+
+
+def _h2(markdown: str) -> list[str]:
+    return [line[3:].strip() for line in markdown.splitlines() if line.startswith("## ")]
+
+
+def test_rendered_english_chapter_has_no_han() -> None:
+    text = render_annual_method_reference({
+        "worksheets": {"timing_and_predictive_systems": {"pyjhora_annual_method_reference": {
+            "years": [_shell(2026)],
+        }}},
+    }, "en")
+    assert "Source: PyJHora" in text
+    assert not any("\u4e00" <= char <= "\u9fff" for char in text)
diff --git a/tests/test_report_english_edition.py b/tests/test_report_english_edition.py
index 66b2528b..46d582b6 100644
--- a/tests/test_report_english_edition.py
+++ b/tests/test_report_english_edition.py
@@ -21,6 +21,7 @@ import pytest
 from scripts.jyotish_engine import build_professional_report_reference_packet, cmd_full_reading
 from scripts.pl9_reader_english import HAN, han_leaks
 from scripts.pl9_reader_export import _dignity_label, _pl9_export_markdown_for_edition
+from scripts.report_render.annual_method_reference import strip_annual_method_chapter
 from scripts.professional_report_reference import (
     ProfessionalReportReferenceInputError,
     _export_args,
@@ -91,7 +92,10 @@ def test_english_edition_has_no_chinese(rendered) -> None:
 
 def test_english_edition_passes_reader_leak_patterns(rendered) -> None:
     _, _, en, _ = rendered
-    assert _leak_hits(en) == []
+    # 原值：英文全文泄漏词为零，包括 PyJHora
+    # 新值：先去掉 Independent Annual Method Reference 再查
+    # 原因：#93 这一章必须写 Source: PyJHora，清理器会删掉更早出现的同一词
+    assert _leak_hits(strip_annual_method_chapter(en)) == []
     assert en.startswith("# Vedic Astrology Chart Report\n")
     for heading in ("## Birth Data and Charts", "## Strength, Relationships and Ashtakavarga",
                     "## Standard Dasha Tables", "## Saturn and KP",
diff --git a/tests/test_report_reader_main.py b/tests/test_report_reader_main.py
index bc45b872..b5f0f374 100644
--- a/tests/test_report_reader_main.py
+++ b/tests/test_report_reader_main.py
@@ -15,6 +15,7 @@ from scripts.jyotish_engine import (
     render_pl9_markdown,
 )
 from scripts.pl9_reader_export import _pl9_export_markdown_for_edition
+from scripts.report_render.annual_method_reference import strip_annual_method_chapter
 from scripts.pl9_full_data_export import REPORT_VERSION_FULL
 from scripts.professional_report_reference import (
     ProfessionalReportReferenceInputError,
@@ -177,8 +178,21 @@ def test_reader_main_hides_audit_vocabulary(full_reading) -> None:
     reader = _pl9_export_markdown_for_edition(packet, "full_data")
     assert "结论等级规则" not in reference
     assert "This reference export" not in reference
-    hits = _leak_hits(reader)
+    # 原值：完整数据版全文 LEAK_PATTERNS（含 PyJHora）零命中
+    # 新值：切掉「独立年度方法参考」后零命中；本章含「来源：PyJHora」，其余模式在本章内为零
+    # 原因：D1、D2 要求表头注明来源，清理器会删掉更早出现的 PyJHora 行
+    from scripts.report_render.annual_method_reference import CHAPTER_ZH, strip_annual_method_chapter
+
+    chapter = ""
+    marker = f"\n## {CHAPTER_ZH}\n"
+    if marker in reader:
+        chapter = reader.split(marker, 1)[1]
+    body = strip_annual_method_chapter(reader)
+    hits = _leak_hits(body)
     assert hits == [], hits
+    assert "来源：PyJHora" in chapter
+    chapter_hits = [pattern.pattern for pattern in LEAK_PATTERNS if pattern.pattern != "PyJHora" and pattern.search(chapter)]
+    assert chapter_hits == [], chapter_hits
     for chapter in CORE_CHAPTERS:
         assert chapter in reader, chapter
     assert reader.strip().startswith("#")
@@ -291,7 +305,10 @@ def test_reader_annual_cells_use_real_distinct_years_and_keep_return_dates(reade
         return_time = annual["solar_return"]["data"]["dt_local"]
         section = body.split(f"### {year} 年度 Varshaphala / Tajika 细项", 1)[1].split("### ", 1)[0]
         assert f"| Varshapravesha | {return_time} |" in section
-    assert _leak_hits(reader) == []
+    # 原值：全文泄漏词为零，包括 PyJHora
+    # 新值：先去掉「独立年度方法参考」再查
+    # 原因：#93 这一章必须写「来源：PyJHora」，清理器会删掉更早出现的同一词
+    assert _leak_hits(strip_annual_method_chapter(reader)) == []
 
 
 def test_reader_annual_conflict_keeps_both_real_producer_values(reader_packet) -> None:
@@ -318,7 +335,10 @@ def test_reader_annual_conflict_keeps_both_real_producer_values(reader_packet) -
     # 新值：| Muntha 星座 | 白羊座 |
     # 原因：完整数据版中文清理把 sign 写成「星座」；星座值仍是当年真实 Muntha
     assert "| Muntha 星座 | 白羊座 |" in reader
-    assert _leak_hits(reader) == []
+    # 原值：全文泄漏词为零，包括 PyJHora
+    # 新值：先去掉「独立年度方法参考」再查
+    # 原因：#93 这一章必须写「来源：PyJHora」，清理器会删掉更早出现的同一词
+    assert _leak_hits(strip_annual_method_chapter(reader)) == []
 
 
 def test_reader_omits_missing_name_and_panchanga_machine_keys(reader_packet) -> None:
@@ -359,4 +379,7 @@ def test_reader_headings_drop_field_jargon_and_dignity_is_one_name(reader_packet
     assert _dignity_label("极敌(Great Enemy)") == "极敌"
     assert _dignity_label("中性") == "中性"
     assert _dignity_label(None) == ""
-    assert _leak_hits(reader) == []
+    # 原值：全文泄漏词为零，包括 PyJHora
+    # 新值：先去掉「独立年度方法参考」再查
+    # 原因：#93 这一章必须写「来源：PyJHora」，清理器会删掉更早出现的同一词
+    assert _leak_hits(strip_annual_method_chapter(reader)) == []
diff --git a/tests/test_runtime_jhora_import_scan.py b/tests/test_runtime_jhora_import_scan.py
index 9c42c5a7..f741159c 100644
--- a/tests/test_runtime_jhora_import_scan.py
+++ b/tests/test_runtime_jhora_import_scan.py
@@ -19,6 +19,9 @@ ENTRIES = (
     "scripts/professional_report_reference.py",
     "scripts/consultation_workflow_service.py",
     "scripts/consultation_native_layers.py",
+    "scripts/unified_consultation_orchestrator.py",
+    "scripts/rectification_three_engine_packet.py",
+    "scripts/report_render/annual_method_reference.py",
 )
 SELF = "tests/test_runtime_jhora_import_scan.py"
 
@@ -162,9 +165,27 @@ def test_product_entries_do_not_import_jhora():
         "scripts/native_special_lagnas.py",
         "scripts/tajika.py",
         "scripts/pl9_reader_export.py",
+        "scripts/unified_consultation_orchestrator.py",
+        "scripts/rectification_three_engine_packet.py",
+        "scripts/report_render/annual_method_reference.py",
     ):
         assert required in visited
     assert all("pyjhora" not in path and "/research/" not in path for path in visited)
+    worker = "pyjhora_annual_reference_worker"
+    for path in visited:
+        tree = ast.parse((ROOT / path).read_text(encoding="utf-8"))
+        for node in ast.walk(tree):
+            if isinstance(node, ast.Import):
+                names = [alias.name for alias in node.names]
+            elif isinstance(node, ast.ImportFrom):
+                names = [node.module or ""]
+            else:
+                continue
+            assert all(worker not in name for name in names), path
+    source = (ROOT / "scripts/report_render/annual_method_reference.py").read_text(encoding="utf-8")
+    assert f'"-m", {worker.upper()}' not in source
+    assert '"-m", WORKER_MODULE' in source
+    assert "pyjhora_annual_reference_worker" in source
 
 
 def test_scanner_is_on_the_quick_gate():
diff --git a/tests/test_unified_consultation_orchestrator.py b/tests/test_unified_consultation_orchestrator.py
index bca7cb19..6ad02640 100644
--- a/tests/test_unified_consultation_orchestrator.py
+++ b/tests/test_unified_consultation_orchestrator.py
@@ -338,10 +338,10 @@ def test_runtime_evidence_log_exposes_blind_packet_case_and_quality_gate_contrac
     assert engines["VedAstro"]["status"] == "local_fallback"
     assert engines["PyJHora/JHora"]["status"] == "reference_available_not_runtime_invoked"
     assert engines["PyJHora/JHora"]["adapter_command"] == "python3 benchmarks/jyotish/scripts/run_pyjhora_compare.py"
-    assert engines["PyJHora/JHora"]["adapter_status"] in {
-        "available",
-        "blocked_missing_python_module:jhora",
-    }
+    # 原值：adapter_status 可以是 available 或 blocked_missing_python_module:jhora
+    # 新值：只有 blocked_missing_python_module:jhora
+    # 原因：#93 产品路径不再因为装了 PyJHora 就改成 available
+    assert engines["PyJHora/JHora"]["adapter_status"] == "blocked_missing_python_module:jhora"
     assert engines["jyotishganit"]["status"] == "reference_available_not_runtime_invoked"
     assert engines["jyotishganit"]["adapter_path"] == "references/open_source_sources/jyotishganit"
     assert engines["jyotishganit"]["adapter_status"] == "available"