Merge remote-tracking branch 'origin/staging' into codex/astrologer-166-rulings

# Conflicts:
#	docs/BUG_HISTORY.md

docs/BUG_HISTORY.md

@@ -21583,6 +21583,23 @@
 - 复发自：无
 - 修复版本：分支 `codex/audit-shadbala-rest-20261104`（未推送）。
 
+## BUG-1580 | 年纪很大的人问父母，回答仍默认父母健在（服务端直接给一句事实，统一盲评通过后上线）
+
+- 状态：fixed-pending-verify（staging；真机未看）
+- 首次发现 / 最近更新：2026-11-04（BUG-1540 回测：80–119 岁的人被劝「趁早沟通」、被写成仍和父母来往）/ 2026-11-04
+- 影响面：`frontend/src/lib/consultation-evidence-card.ts`、`frontend/src/mastra/consultation-tools.ts`、`frontend/src/app/api/consult/route.ts`、`frontend/src/mastra/product-voice.ts`、`frontend/scripts/research/consult-biography-backtest.mts`。
+- 现象：普通本命咨询里模型看不到出生日期，年纪大的人问父母时，回答照样写和父母现在的来往，或给以父母健在为前提的建议。
+- 触发条件：本人 70 岁以上，问父母、家庭类问题。
+- 根因：模型拿不到出生日期，自己无从判断年纪；卡上给了 `age_years`（a957edb9，ac7d25db 撤回）也不用。只给事实时，模型更倾向于遵守口吻里「在不在世盘分不出、不预设」的红线，写「也不说在不在世」。
+- 修复（产品 2026-11-04：服务端判断，模型不算）：本人在请求时（UTC+8 日历日）满 `ELDER_PARENTS_AGE_THRESHOLD = 65` 岁时，卡上 parents / family / general 三个领域段在 label 下加一句 `parents_life_stage`：「本人已 N 岁，父母多半已不在世：不默认父母健在，不写和父母现在的来往，不给以此为前提的建议（如趁早沟通、多陪伴）。」不满 65 岁或没有出生日期时不加。出生日期、时间、地点仍不进模型。VISIBLE VOICE 在传记事实红线后加一句「卡上有 parents_life_stage 时照它写，不当父母还健在」。回测脚本改为读 golden 里的真实出生日期（原写死 1950-01-01）。同批推上 K 按 staging 71d38b0d 重新生成的证据卡、名人与 card-facts 捕获。
+- 验证：合同测试 6 条（64 / 65 门槛、生日前后、2 月 29 日、没有出生日期、模型视图里没有出生资料、只出现在这三个领域）；VISIBLE VOICE 合同测试加一条。统一配对盲评（同一份评分说明，按题配对，基线与 L / M / N 三版各 4 轮 × 53 题 = 212 份，两名评分员重评 84 份，判定一致 93%，bootstrap 10 000 次）：本改动（M）忽视年纪 67 → 8（差 −59，95% CI [−93, −29]），卡罗、梦露、泰勒、布什、乔布斯父母题归零；严重 32 → 39（+7，CI [−7, +21]，不显著）；误报 6 → 9（+3，CI [0, +8]，临界）；可读性 3.79 → 3.84（不显著）。
+- 已知副作用（上线后盯）：卡罗父亲一侧严重 0 → 4、梦露婚姻题 0 → 4、布什原题误报 +2。先前两轮小样本回测（每版两轮、各自评分员）严重 22–25 对基线 15，按当时规则撤回过一次（788725fb）；统一评法下差别不显著，产品 2026-11-04 决定上线。
+- 未解决（与本改动无关，四版都有）：泰勒婚姻题 32 份里 30 份写成「定下来就稳」；梦露、奥巴马缺席父母被写出分量或温度；布什父母题每版 4–7 个误报。
+- 防复发：服务端给事实时，同时检查口吻里有没有与之冲突的红线；提示词改动用统一配对盲评判，不用不同评分员的小样本对比。
+- 关联记录：BUG-1540、BUG-1558（a957edb9 / ac7d25db）、BUG-1164、BUG-1165。
+- 复发自：无
+- 修复版本：分支 `codex/ship-m-golden-20261104`，推 staging。
+
 ## BUG-1581 | 月亮盈亏力（Paksha）亏月取 60 − P：梵文 27.11 把月亮与水、金、木并列取 P，「亏月归凶再取反」只在英译注释里
 
 - 状态：resolved（引擎 + 回归测试；未部署）

docs/research/pre_work_error_ledger.md

@@ -25,6 +25,7 @@ Purpose: read this file before substantial project work. It exists to stop repea
 - 数据核对 J：Sthana / Dig / Drik（worktree `/workspace/wt-j`，`codex/audit-shadbala-rest-20261104`）：`pre_work_check.py` exit 0。本仓与 PyJHora 在 19 张盘加乔布斯盘上 Dig 140 / 140「一致」，但两边用的是同一个没折回的函数（PyJHora `_dig_bala`；它自己的 `_dig_bala_another` 是折回的）。这是第二次出现「两家一致 = 两家同错」（BUG-1546 之后）。BUG-1270 已按 BPHS 27.7 和 Raman GBB 例 14–15 修好。同一份经度两次复算，Drik 差了 0.01：求和顺序跟着 PYTHONHASHSEED 变，记为 BUG-1576，已修。两个修复一起重冻为 `audit_j_2026_11_04`。比较前后两次运行时，要固定 PYTHONHASHSEED，或者先确认被比较的函数本身是确定的。教训：对照某个外部引擎时，先查这个引擎有没有同一分量的多种写法，几种都跑一遍，再拿原书的例题核对。Sthana 和 Drik 用 Raman 标准盘例 13、例 55 逐项吻合。jyotishganit 在本机跑需要 Hipparcos 星表，下载被 SSL 拦截，改用 Spica 的 J2000 星表值替代；七星经度与 Lahiri 相差不超过 0.022°。jyotishganit 的上升点比我们晚约 1.08°，Piaf 和 Kahlo 两张盘的上升换了星座，不要把它的 Kendra / Dig 当对照。`jyotish_engine.MOOLATRIKONA` 里月亮写成金牛 0–3°，记为 BUG-1575，未修。全量 Shadbala 相关测试在负载 69 的机器上很慢，基线放在独立的干净 worktree（`/workspace/wt-j-base`）里跑，不占用改动中的 worktree。
 - 数据核对第二波合并（worktree `/workspace/wt-int2`，`codex/audit-wave2-integration-20261104`，基于 staging 8f15745c）：G、H、I、J 依次 `--no-ff` 合并。G / H / J 各自的 `audit_g` / `audit_h` / `audit_j` 冻结记录只对各自分支的代码成立，合并后的代码上哪一份都不是当前记录；合并时合同、FREEZE / REPORT 常量与两份门测试先保持 G 的状态，最后按 ERR-110 只重冻一次为 `audit_wave2_2026_11_04`，合同上一份指回 `moon_paksha_2026_11_04`，三份分支记录逐字节保留为中间记录。与 `moon_paksha_2026_11_04` 逐条逐字段比：密封 20 / 20、报时 900 / 900 全同，`metrics`、`summary` 不变，changed_trial_count 仍为 8 / 581。冲突只在文档（BUG_HISTORY、台账、CHANGELOG 全部保留）、冻结合同与 `run_chara_dasha_compare.py`（G、H 都改了岁差传参，取 H 的写法）。注意：`tests/test_narayana_rath_book_tables.py::test_rath_profile_is_display_only` 在 staging 8f15745c 上就失败（`scripts/annual_service/package.py` 也引用 `narayana_rath`），不是本轮造成。
 - BUG-1560 D30 + BUG-1570 金星功能凶入算（worktree `/workspace/wt-sc`，`codex/d30-venus-roles-scoring-20261104`，基于 staging 607a48b6；产品 2026-11-04 决定）：`pre_work_check.py` exit 0。两项一起升 `rectification-v5-matrix-scoring-13`、记忆化 golden v6（v5 逐字节保留）、跨午夜 scoring-13 golden、dated-transitions scan 重生成，ERR-110 只重冻一次为 `d30_venus_2026_11_04`。对 `audit_wave2_2026_11_04`：密封 3 / 20 条变（头名率 0.50 → 0.45，全部来自 D30），报时 250 / 900 条变（头名命中 46 → 7，全部来自金星；覆盖不变）。归因用 `git archive` 副本各改一项各自冻结回放（冻结记录放副本内、传绝对路径）。教训：①`midnight_date_anchor_regression.py --golden` 先断言 baseline 与当前同日 A/B 不变，分数变化的升版里 baseline 只能传当前树（只证明独立进程可复现），不能传旧树；②记忆化公开请求（1990-01-01 北京）候选分数这次没变，变的是 D30 特征与哈希，「golden 红」不等于「这张盘分数变了」；③两项在同一案例（第 20 例）上有交互，合并结果不是单项变化条数之和（166 + 108 − 15 重叠 ≠ 250），归因要逐条比对。前端捕获 golden（咨询卡、名人回测）仍是改前的金星中性快照，未重生成。
+- 前端捕获 golden 重生成（worktree `/workspace/wt-k`，`codex/backtest-golden-refresh-20261104`，基于 staging 71d38b0d）：`pre_work_check.py` exit 0。`consult-evidence-card-golden.json`、`consult-biography-backtest-golden.json`、`consult-card-facts-20261009.json` 按各自脚本（`JYOTISH_API_CHART_CACHE_TTL_SECONDS=0 PYTHONHASHSEED=0`）重生成。注意：两份捕获 golden 实际是 10-04 / 10-07 的快照，不是「今天改前」；用 `git archive db2fc3fc`（今天引擎修复之前）副本再捕获一次做归因：db2fc3fc → 71d38b0d 的差异全部落在 BUG-1541～1572 / 1270 / 1576 范围（天蝎金星功能凶：泰勒、琵雅芙；卡罗 Vimshottari 第二轮；Narayana 当前运；Chara 双主星 / 小运；瑜伽；Shadbala；D16–D45；婚姻点昼夜公式）。旧 golden → db2fc3fc 的 10 月漂移另有来源、不在今天的修复清单里：BUG-1275 章动进恒星系（岁差 ±0.005°，全部经度、星宿、Vimshottari 起止日差几天、高次分盘个别星座）、aa87c0f1 Argala 平局规则与新增字段、上升星宿字段、`vedastro_cross_check` 由 `official_verified` 变 `blocked`（研究替身现在真正生效；旧值的来源未核实）。前端读这三份 fixture 的 18 个测试文件 118 条中只有 2 条要改（BUG-1570、BUG-1547），没有断言卡在 10 月漂移的值上。另：回测脚本默认不带 `consult_card_facts`（判断表 / participation），产品每次都带；要贴近产品需设 `CONSULT_CARD_FACTS_FIXTURE=<freeze_consult_card_facts_fixture.py --biography-out 的输出>`，设与不设的回测不可直接比较。
 
 ## 2026-11-03 · 输入框单行起步（方案 2，worktree `/workspace/wt-composer`）
 

frontend/scripts/research/consult-biography-backtest.mts

@@ -243,7 +243,8 @@ function serverChart(figure: GoldenFigure): ServerChartConsultation {
     name: "backtest",
     toolInput,
     truth: {
-      birthDate: "1950-01-01",
+      // BUG-1580: the card's parents_life_stage reads the age from this date; the model never sees the date itself.
+      birthDate: typeof birth.date === "string" ? birth.date : "1950-01-01",
       reportedBirthTime: null,
       activeBirthTime: null,
       selectedTimeKind: "reported",
@@ -308,6 +309,7 @@ function agentContext(job: Job, state: ConsultationRuntimeState, executed: strin
     theme: job.domain as ConsultationDomain,
     followUpTurn: Boolean(job.followUp),
     serverChart: serverChart(job.figure),
+    requestTime: REQUEST_TIME,
     state,
     runWorkflow: stubRunWorkflow(job.figure, executed) as never,
   });

frontend/src/app/api/consult/route.ts

@@ -1561,6 +1561,7 @@ export async function POST(request: Request) {
       abortSignal: agentAbortSignal,
       state,
       rectificationOffer,
+      requestTime,
     });
     const agent = getJyotishAgent(selectedModel, agentContext);
     const result = await streamWithOverflowRetry(agent, natalStreamOptions);

frontend/src/lib/consultation-evidence-card.ts

@@ -29,7 +29,10 @@ import { recordOrEmpty as record } from "@/lib/guards";
  * card-authored strings are `EVIDENCE_CARD_LABELS` (and domain ids that point
  * at another section); selections (which houses, which conclusions) are made
  * here, values never. One label is read from a value: each domain house row's
- * `sav_strength` (强 / 中 / 弱 from the engine's raw SAV, savStrength).
+ * `sav_strength` (强 / 中 / 弱 from the engine's raw SAV, savStrength). One
+ * sentence is computed from the profile, not the engine: `parents_life_stage`
+ * (BUG-1580, evidenceCardParentsLifeStage), and only the age in it reaches
+ * the model, never the birth date.
  *
  * Pure functions, no React.
  */
@@ -120,6 +123,8 @@ export type EvidenceCardDomainSpec = Readonly<{
   vivahSaham: boolean;
   /** The section carries the current subject's gender (「性别未知」 when not filled). */
   gender: boolean;
+  /** BUG-1580: the section carries `parents_life_stage` when the native is elderly. */
+  parentsLifeStage: boolean;
   /** The annual Tajika chart (annual route only). */
   annualChart: boolean;
   /** Houses of the running and next AD lords and of the year's Jupiter / Saturn transits. */
@@ -164,6 +169,7 @@ const spec = (value: Partial<EvidenceCardDomainSpec> & Pick<EvidenceCardDomainSp
   dayNight: false,
   vivahSaham: false,
   gender: false,
+  parentsLifeStage: false,
   annualChart: false,
   annualHouses: false,
   followVargas: false,
@@ -207,9 +213,9 @@ export const EVIDENCE_CARD_SPECS: Readonly<Record<ConsultationDomain, EvidenceCa
   health: spec({ vargas: ["D6", "D8", "D30"], houses: [1, 6, 8, 12], transits: true, shadbala: true }),
   education: spec({ vargas: ["D5", "D24"], houses: [4, 5, 9], planets: ["Mercury", "Jupiter"] }),
   migration: spec({ vargas: ["D4", "D12"], houses: [4, 12] }),
-  family: spec({ vargas: ["D7", "D12"], houses: [4, 5, 9] }),
+  family: spec({ vargas: ["D7", "D12"], houses: [4, 5, 9], parentsLifeStage: true }),
   parents: spec({
-    vargas: ["D12"], houses: [4, 9], planets: ["Sun", "Moon"], karakas: ["MK"], yogas: true, yogaTags: ["parents"],
+    vargas: ["D12"], houses: [4, 9], planets: ["Sun", "Moon"], karakas: ["MK"], yogas: true, yogaTags: ["parents"], parentsLifeStage: true,
     roles: [
       { key: "mother", label: EVIDENCE_CARD_LABELS.roleMother, house: 4, planets: ["Moon"], karakas: ["MK"], optionalKarakas: [], varga: "D12" },
       { key: "father", label: EVIDENCE_CARD_LABELS.roleFather, house: 9, planets: ["Sun"], karakas: [], optionalKarakas: ["PiK"], varga: "D12" },
@@ -223,7 +229,7 @@ export const EVIDENCE_CARD_SPECS: Readonly<Record<ConsultationDomain, EvidenceCa
   }),
   general: spec({
     vargas: ["D9", "D10", "D2"], houses: [1, 10, 7, 2], arudha: ["A10"], yogas: true, yogaTags: ["general"], traditionalYogas: true, transits: true,
-    ashtakavarga: true, shadbala: true,
+    ashtakavarga: true, shadbala: true, parentsLifeStage: true,
   }),
 };
 
@@ -319,6 +325,8 @@ export type EvidenceCardDomainSection = {
   vargas_follow?: ConsultationDomain;
   /** Per-person index into this section (parents: mother / father). */
   roles?: JsonRecord;
+  /** BUG-1580: an elderly native's parents are probably no longer living (evidenceCardParentsLifeStage). */
+  parents_life_stage?: string;
   /**
    * Present only when this domain's own base differs from the card's base
    * (the first domain's). The same chart gives the same base; a difference is
@@ -378,8 +386,52 @@ export function evidenceCardGenderLabel(gender: ProfileGender | null | undefined
 export type EvidenceCardOptions = Readonly<{
   /** The current subject's optional gender; null / undefined = not filled. */
   gender?: ProfileGender | null;
+  /**
+   * BUG-1580: the native's age in whole years at the request time
+   * (`evidenceCardAgeYears`). The card never carries it as a field; at
+   * ELDER_PARENTS_AGE_THRESHOLD or above it only picks the
+   * `parents_life_stage` sentence. null / undefined = unknown, nothing added.
+   */
+  ageYears?: number | null;
 }>;
 
+/**
+ * BUG-1580: from this age on, the parents / family / general sections say the
+ * native's parents are probably no longer living. Parents are typically 25–35
+ * years older than their child, so at 65 they would be about 90–100. Tunable.
+ */
+export const ELDER_PARENTS_AGE_THRESHOLD = 65;
+
+/**
+ * BUG-1580: the native's current age in whole years, from the profile's birth
+ * date (YYYY-MM-DD, local calendar date) and the request's authoritative
+ * server time read as a China Standard Time (UTC+8) calendar date, the same
+ * date the user turn's 「服务端当前时间」 carries. Born on 29 February: the
+ * birthday counts from 1 March in other years. null when the date is missing,
+ * not a real date, or after the request.
+ */
+export function evidenceCardAgeYears(birthDate: string | null | undefined, at: Date): number | null {
+  const match = typeof birthDate === "string" ? /^(\d{4})-(\d{2})-(\d{2})$/.exec(birthDate) : null;
+  if (!match || Number.isNaN(at.getTime())) return null;
+  const [year, month, day] = match.slice(1).map(Number) as [number, number, number];
+  const born = new Date(Date.UTC(year, month - 1, day));
+  if (born.getUTCFullYear() !== year || born.getUTCMonth() !== month - 1 || born.getUTCDate() !== day) return null;
+  const today = new Date(at.getTime() + 8 * 60 * 60 * 1000);
+  const [ty, tm, td] = [today.getUTCFullYear(), today.getUTCMonth() + 1, today.getUTCDate()];
+  const age = ty - year - (tm < month || (tm === month && td < day) ? 1 : 0);
+  return age >= 0 ? age : null;
+}
+
+/**
+ * BUG-1580: the plain fact an elderly native's sections carry, or null below
+ * the threshold / when the age is unknown. The model is never asked to work
+ * the age out; it reads this sentence.
+ */
+export function evidenceCardParentsLifeStage(ageYears: number | null | undefined): string | null {
+  if (typeof ageYears !== "number" || !Number.isFinite(ageYears) || ageYears < ELDER_PARENTS_AGE_THRESHOLD) return null;
+  return `本人已 ${ageYears} 岁，父母多半已不在世：不默认父母健在，不写和父母现在的来往，不给以此为前提的建议（如趁早沟通、多陪伴）。`;
+}
+
 /** One domain's inputs: the projected packet and the agent context behind it. */
 export type EvidenceCardSource = Readonly<{
   domain: ConsultationDomain;
@@ -702,8 +754,11 @@ function domainSection(
   const localLayers = record(context.local_layers);
   const chart = record(context.chart);
   const formal = record(record(natal.varga_spectrum).formal);
+  const lifeStage = rule.parentsLifeStage ? evidenceCardParentsLifeStage(options.ageYears) : null;
   const section: EvidenceCardDomainSection = {
     label: consultationDomainDefinition(domain).label,
+    // BUG-1580: right under the label, before any chart data.
+    ...(lifeStage ? { parents_life_stage: lifeStage } : {}),
     vargas: {},
     houses: {},
   };

frontend/src/mastra/consultation-tools.ts

@@ -29,6 +29,7 @@ import { CONSULTATION_QUEUE_LABEL, retryOnConsultationRateLimit } from "../lib/c
 import { consultDomainConcurrency } from "../lib/consult/prep-flags.ts";
 import {
   buildEvidenceCard,
+  evidenceCardAgeYears,
   citedEvidenceCardFields,
   EVIDENCE_LOOKUP_SECTIONS,
   evidenceCardTokenEstimate,
@@ -600,6 +601,8 @@ export type ConsultationAgentContext = Readonly<{
   // The domain loop budgets itself against this clock, so tests can drive the
   // deadline without waiting for it.
   now?: () => number;
+  /** BUG-1580: the request's authoritative server time (the user turn's 服务端当前时间); the native's age is read at it. */
+  requestTime?: Date;
 }>;
 
 export type WindowConsultationCalculationCache = {
@@ -1505,7 +1508,11 @@ export function createConsultationTools(ctx: ConsultationAgentContext) {
             domain: execution.domain,
             packet: execution.modelOutput,
             context: execution.context,
-          })), { gender: subjectGender });
+          })), {
+            gender: subjectGender,
+            // BUG-1580: only the age reaches the model (in parents_life_stage); the birth date stays on the server.
+            ageYears: evidenceCardAgeYears(ctx.serverChart.truth?.birthDate, ctx.requestTime ?? new Date()),
+          });
           const modelView = toModelEvidenceView(modelContext, card);
           ctx.state.evidenceCard = card;
           ctx.state.evidenceCardChars = JSON.stringify({

frontend/src/mastra/product-voice.ts

@@ -23,7 +23,7 @@ export const productConversationVoice = `VISIBLE VOICE
 
 红线（这几条之外靠你的判断）：
 - 只用卡上和补取到的盘面事实，不编位置、不编时间；不允许精确应期时不写月份。分析清单（受冲信号的编号和条数、层名、规则原句）只用来判断，不念给用户；不写「从某宫数起第几位」这类数法，只说结论（「这颗星落在对这个宫不利的位置」）。
-- 用户自己更清楚的传记事实——有没有伴侣、是否已婚、几段感情、父母在不在身边、在不在世、有没有病、在不在上班、做哪一行、孩子性别——盘分不出，不预设、不替他断定；性别未知时不用「他」「她」称伴侣和孩子。某个人这条线压力重时，盘分不出这是聚少离多还是人在但相处吃力：用自己的话给出这个范围，全篇只说一次（几个人都重时合成一句），不套固定句式；这一侧不写这个人实际怎么对你——照顾、管、在意、有温度、有分量都不写，「善意是存在的」「这条线不是空的」这类安慰也不写；可以写你从这一侧带走的东西。这一侧压力不重时才写他的方式。只有回答确实取决于这类事实时，才问一句，全篇最多一次。
+- 用户自己更清楚的传记事实——有没有伴侣、是否已婚、几段感情、父母在不在身边、在不在世、有没有病、在不在上班、做哪一行、孩子性别——盘分不出，不预设、不替他断定；卡上有 parents_life_stage 时照它写，不当父母还健在；性别未知时不用「他」「她」称伴侣和孩子。某个人这条线压力重时，盘分不出这是聚少离多还是人在但相处吃力：用自己的话给出这个范围，全篇只说一次（几个人都重时合成一句），不套固定句式；这一侧不写这个人实际怎么对你——照顾、管、在意、有温度、有分量都不写，「善意是存在的」「这条线不是空的」这类安慰也不写；可以写你从这一侧带走的东西。这一侧压力不重时才写他的方式。只有回答确实取决于这类事实时，才问一句，全篇最多一次。
 - 不断言死亡、离婚、疾病诊断、流产、不孕；不给医疗、法律、投资指令。
 - 不起「外热内冷」这类四字格局名，不把大运写成推着谁、替谁收尾的角色，空宫不单独下结论（说它的宫主落在哪）。
 - 不以「根据计算结果」「综合来看」「从星盘来看」开头；不写内部字段、英文状态码、JSON、技法审计表、统一参数。

frontend/tests/consult-elder-parents-hint-20261104.test.ts

@@ -0,0 +1,135 @@
+// BUG-1580: answers presumed living parents for natives aged 80–119 (BUG-1540
+// backtests). Asking the model to work out the age failed twice (a voice clause
+// with no birth date; an age_years field it ignored). The server now decides:
+// at ELDER_PARENTS_AGE_THRESHOLD or above, the parents / family / general
+// sections carry one plain sentence with the age; below it, nothing. The birth
+// date, time and place never reach the model.
+import assert from "node:assert/strict";
+import { readFileSync } from "node:fs";
+import test from "node:test";
+
+import {
+  buildEvidenceCard,
+  EVIDENCE_CARD_SPECS,
+  ELDER_PARENTS_AGE_THRESHOLD,
+  evidenceCardAgeYears,
+  evidenceCardParentsLifeStage,
+} from "../src/lib/consultation-evidence-card.ts";
+import type { ConsultationDomain } from "../src/lib/consultation-domain-registry.ts";
+import { createConsultationRuntimeState, createConsultationTools } from "../src/mastra/consultation-tools.ts";
+
+type Json = Record<string, unknown>;
+const golden = JSON.parse(readFileSync(
+  new URL("./fixtures/consult-evidence-card-golden.json", import.meta.url),
+  "utf8",
+)) as { charts: { id: string; workflow: Json }[] };
+
+const AT = new Date("2026-09-27T04:00:00.000Z");
+
+test("threshold is 65 and the sentence carries the age; 64 gets nothing", () => {
+  assert.equal(ELDER_PARENTS_AGE_THRESHOLD, 65);
+  assert.equal(evidenceCardParentsLifeStage(64), null);
+  assert.equal(
+    evidenceCardParentsLifeStage(65),
+    "本人已 65 岁，父母多半已不在世：不默认父母健在，不写和父母现在的来往，不给以此为前提的建议（如趁早沟通、多陪伴）。",
+  );
+  assert.match(evidenceCardParentsLifeStage(110)!, /^本人已 110 岁，/);
+  for (const unknown of [null, undefined, Number.NaN]) assert.equal(evidenceCardParentsLifeStage(unknown), null);
+});
+
+test("age is whole years at the request's China calendar date, around the birthday", () => {
+  // 2026-07-05 23:59:59 in UTC+8 is the day before the 65th birthday; 00:00 on 07-06 is the birthday.
+  assert.equal(evidenceCardAgeYears("1961-07-06", new Date("2026-07-05T15:59:59.000Z")), 64);
+  assert.equal(evidenceCardAgeYears("1961-07-06", new Date("2026-07-05T16:00:00.000Z")), 65);
+  // The year turns over in UTC+8 before it does in UTC.
+  assert.equal(evidenceCardAgeYears("2000-01-01", new Date("2025-12-31T16:00:00.000Z")), 26);
+  assert.equal(evidenceCardAgeYears("2000-01-01", new Date("2025-12-31T15:59:59.000Z")), 25);
+  // Born on 29 February: the birthday counts from 1 March in other years.
+  assert.equal(evidenceCardAgeYears("2004-02-29", new Date("2026-02-28T04:00:00.000Z")), 21);
+  assert.equal(evidenceCardAgeYears("2004-02-29", new Date("2026-03-01T04:00:00.000Z")), 22);
+  // The backtest figures at its request time.
+  assert.equal(evidenceCardAgeYears("1946-07-06", AT), 80);
+  assert.equal(evidenceCardAgeYears("1907-07-06", AT), 119);
+  for (const missing of [null, undefined, "", "1946-02-30", "1946/07/06", "2030-01-01"]) {
+    assert.equal(evidenceCardAgeYears(missing, AT), null, String(missing));
+  }
+});
+
+const empty = (domain: ConsultationDomain) => ({ domain, packet: { claim_cards: [] }, context: {} });
+
+test("card: the sentence sits right under the label, only in parents / family / general", () => {
+  const intended = new Set<ConsultationDomain>(["parents", "family", "general"]);
+  for (const domain of Object.keys(EVIDENCE_CARD_SPECS) as ConsultationDomain[]) {
+    const card = buildEvidenceCard([empty(domain)], { ageYears: 80 });
+    const section = card.domain_sections[domain]!;
+    if (intended.has(domain)) {
+      assert.equal(Object.keys(section)[1], "parents_life_stage", domain);
+      assert.match(section.parents_life_stage!, /^本人已 80 岁，父母多半已不在世/);
+    } else {
+      assert.equal("parents_life_stage" in section, false, domain);
+    }
+  }
+  // No other part of the card carries the age or the sentence.
+  const card = buildEvidenceCard([empty("parents"), empty("career")], { ageYears: 80 });
+  assert.equal(JSON.stringify(card.base).includes("80"), false);
+  assert.equal("parents_life_stage" in card.domain_sections.career!, false);
+  assert.equal(card.meta.gaps.some((gap) => gap.includes("age") || gap.includes("life_stage")), false);
+});
+
+test("card: under 65, unknown age or no option -> no field and no gap", () => {
+  for (const options of [{ ageYears: 64 }, { ageYears: null }, {}]) {
+    const card = buildEvidenceCard([empty("parents"), empty("general")], options);
+    assert.equal("parents_life_stage" in card.domain_sections.parents!, false, JSON.stringify(options));
+    assert.equal("parents_life_stage" in card.domain_sections.general!, false, JSON.stringify(options));
+    assert.equal(card.meta.gaps.some((gap) => gap.includes("age") || gap.includes("life_stage")), false);
+  }
+});
+
+async function modelView(truth: Json, requestTime: Date, domains: ConsultationDomain[] = ["parents"]) {
+  const chart = golden.charts[0]!;
+  const tools = createConsultationTools({
+    userId: "u", sessionId: "s", requestId: "elder", consultationMode: "verified_chart",
+    serverChart: {
+      name: "public",
+      toolInput: { year: 1946, month: 7, day: 6, hour: 7, minute: 26, city: "fixture", lat: 0, lon: 0, tz: 0, ayanamsa: "raman" },
+      truth,
+    } as never,
+    state: createConsultationRuntimeState(),
+    runWorkflow: async () => structuredClone(chart.workflow) as never,
+    requestTime,
+  });
+  return await tools["run-jyotish-consultation"].execute!(
+    { question: "q", domains } as never,
+    { writer: { custom: async () => {} } } as never,
+  ) as Json;
+}
+
+function domainSections(view: Json) {
+  return (view.claim_cards as { category: string; evidence: Json }[])
+    .find((card) => card.category === "domain")!.evidence as Record<string, Json>;
+}
+
+test("model view: an 80-year-old's parents section says so; no birth date, time or place reaches the model", async () => {
+  const view = await modelView({
+    birthDate: "1946-07-06",
+    reportedBirthTime: "07:26",
+    placeLabel: "New Haven, CT",
+    birthTimeSource: "reported",
+  }, AT);
+  assert.match(String(domainSections(view).parents!.parents_life_stage), /^本人已 80 岁，/);
+  const text = JSON.stringify(view);
+  for (const leak of ["1946-07-06", "1946", "07:26", "New Haven", "birth_date", "birthDate", "age_years"]) {
+    assert.equal(text.includes(leak), false, `model view carries ${leak}`);
+  }
+});
+
+test("model view: a day before the 65th birthday nothing; on it the sentence; no birth date -> nothing", async () => {
+  const truth = { birthDate: "1961-09-27", birthTimeSource: "reported" };
+  const before = await modelView(truth, new Date("2026-09-26T15:59:59.000Z"));
+  assert.equal("parents_life_stage" in domainSections(before).parents!, false);
+  const on = await modelView(truth, new Date("2026-09-26T16:00:00.000Z"));
+  assert.match(String(domainSections(on).parents!.parents_life_stage), /^本人已 65 岁，/);
+  const missing = await modelView({ birthTimeSource: "reported" }, AT);
+  assert.equal("parents_life_stage" in domainSections(missing).parents!, false);
+  assert.equal(JSON.stringify(missing).includes("本人已"), false);
+});

frontend/tests/consult-projection-native-20260927.test.ts

@@ -42,6 +42,16 @@ test("the natal layers are projected value for value: D9 summary, Punarphoo, Viv
     const natal = evidence(chart.workflow, "marriage", "natal_foundation");
     for (const name of ["d9_summary", "punarphoo", "vivah_saham", "day_night"]) {
       assert.equal((engine[name] as Json).status === "blocked", false, `${chart.id} ${name} executed`);
+      // 原值：vivah_saham 与引擎逐字段相等（引擎没有 formula_variant）
+      // 新值：引擎多出 formula_variant（"day" / "night"），投影的白名单没有它；其余字段仍逐字段相等，formula_variant 不进投影
+      // 原因：BUG-1547。本命婚姻点改用共享的昼夜公式，引擎随值写出所用公式；捕获 golden 2026-11-04 重生成。是否把它给模型另议
+      if (name === "vivah_saham") {
+        const { formula_variant: variant, ...projected } = engine[name] as Json;
+        assert.ok(variant === "day" || variant === "night", `${chart.id} vivah_saham formula_variant`);
+        assert.equal("formula_variant" in (natal[name] as Json), false, `${chart.id} vivah_saham formula_variant not projected`);
+        assert.deepEqual(natal[name], projected, `${chart.id} ${name}`);
+        continue;
+      }
       assert.deepEqual(natal[name], engine[name], `${chart.id} ${name}`);
     }
     const d9 = natal.d9_summary as Json;

frontend/tests/consult-upstream-functional-v2-20261002.test.ts

@@ -42,12 +42,15 @@ test("goldens carry functional profile v2 from the real engine (BUG-1158)", () =
   ];
   assert.equal(layers.length, 12);
   for (const layer of layers) assert.equal(layer.profile, "bphs_ch34_general_with_sign_exceptions_v2");
-  // Scorpio (Taylor): explicit exceptions Mars mixed, Sun yoga-supporting; Venus (7/12) no longer malefic by the 12th alone.
+  // Scorpio (Taylor): explicit exceptions Mars mixed, Sun yoga-supporting.
+  // 原值：functional_malefics ["Mercury", "Saturn"]，functional_neutrals ["Mars", "Venus"]（金星 7 / 12 宫主不因 12 宫单独判凶）
+  // 新值：functional_malefics ["Mercury", "Saturn", "Venus"]，functional_neutrals ["Mars"]
+  // 原因：BUG-1570。BPHS 34.35 明文：天蝎上升的金星为凶；引擎补上这条例外，捕获 golden 2026-11-04 重生成
   const taylor = card(golden.charts.find((chart) => chart.id === "elizabeth_taylor")!.workflow, "career");
   const functional = taylor.base.natal.functional_benefic_malefic as Json;
   assert.deepEqual(functional.functional_benefics, ["Jupiter", "Moon", "Sun"]);
-  assert.deepEqual(functional.functional_malefics, ["Mercury", "Saturn"]);
-  assert.deepEqual(functional.functional_neutrals, ["Mars", "Venus"]);
+  assert.deepEqual(functional.functional_malefics, ["Mercury", "Saturn", "Venus"]);
+  assert.deepEqual(functional.functional_neutrals, ["Mars"]);
   assert.deepEqual(functional.yogakarakas, []);
   // Virgo (Jobs): Jupiter challenging by exception, Sun (sole 12th lord) neutral.
   const jobs = card(golden.charts.find((chart) => chart.id === "steve_jobs")!.workflow, "parents");

frontend/tests/consultation-voice-contract.test.ts

@@ -73,6 +73,11 @@ test("the red lines the old rules guarded are all still in VISIBLE VOICE (BUG-15
   assert.match(redLines, /分析清单（受冲信号的编号和条数、层名、规则原句）只用来判断，不念给用户；不写「从某宫数起第几位」这类数法/);
   // BUG-1164 / BUG-1182: never decide a biographical fact the user knows better, never the opposite reassurance.
   assert.match(redLines, /有没有伴侣、是否已婚、几段感情、父母在不在身边、在不在世、有没有病、在不在上班、做哪一行、孩子性别——盘分不出，不预设、不替他断定/);
+  // 原值: 传记事实红线只说「父母在不在身边、在不在世……盘分不出，不预设、不替他断定」
+  // 新值: 同一句后接一个分句「卡上有 parents_life_stage 时照它写，不当父母还健在」
+  // 原因: BUG-1580（2026-11-04）卡上给 65 岁以上的人加了服务端写好的一句（本人已 N 岁，父母多半已不在世）；
+  //       只加字段的两轮回测里模型照「在不在世盘分不出」写「也不说在不在世」，106 份里只有 3 份顾及年纪，红线与卡上事实打架，只加这一句指过去
+  assert.match(redLines, /不预设、不替他断定；卡上有 parents_life_stage 时照它写，不当父母还健在；性别未知时/);
   assert.match(redLines, /性别未知时不用「他」「她」称伴侣和孩子/);
   assert.match(redLines, /某个人这条线压力重时，盘分不出这是聚少离多还是人在但相处吃力：用自己的话给出这个范围，全篇只说一次（几个人都重时合成一句），不套固定句式/);
   assert.match(redLines, /照顾、管、在意、有温度、有分量都不写，「善意是存在的」「这条线不是空的」这类安慰也不写/);