feat(sync/friday): 病历摄入 Phase 1 — EMR/诊断/治疗/影像/治疗计划 五资源接入

FRIDAY 病历链落地(直连勘查 2026-07-18 定案,零新算子,纯配置+导出脚本):
- export.sh:host 等价单表 dump(MySQL 7 表 CSV + Mongo med_emr_info status∈{3,4}
  平铺 JSON;id 字段字符串化对齐 CSV 契约;数组字段双形态:原生数组给 split,
  JSON 字符串给 emr 自由文本位)。凭据走 env 不入 git。
- manifest:EMR 链 transforms —— diag[] 拆行→std_diag lookup(linkCode→std_code,
  K 开头截 K 码,非 K/未命中自然落 code=null 自由文本)→normalize/coalesce;
  treat[] 拆行(=本次治疗 actual,病历无计划字段已用全字段枚举证实);
  med_check lookup 类型字典;customer_treat_plan 行 join 头表(无诊所行显式 filter 丢弃)。
- assemblers ×5:emr/diagnosis/treatment_actual/treatment_planned/image,
  全部 dict_includes 共享字典(diagnosis-kcode / treatment-category-core+
  actual/planned-rules / image-modality)—— 分层重构的首个第二宿主消费者。
- 锚点:临床事实 sourceEncounterExternalId=emr_sub_id(同诊聚合);
  emr 挂 appointment_id(56.5%);影像挂 emr_id 病历号。

全量实跑验证(本地):107,227 患者 / 288,383 txns / 285,986 facts,failed=0;
旧 450 样本幂等去重 ✓。诊断 83.5% 拿到 K 码(共享名称字典生效),治疗 11 类闭集
分布合理。周元策(987d743a)persona v1 + 召回计划与其 K04 诊断临床吻合,闭环通。

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

apps/pac-service/data/friday/.gitignore

@@ -3,5 +3,6 @@
 *
 !.gitignore
 !manifest.yaml
+!export.sh
 !assemblers
 !assemblers/*.yaml

apps/pac-service/data/friday/assemblers/diagnosis.yaml

+# diagnosis — FRIDAY 诊断 fact(med_emr_info.diag[] 拆行)
+# 消费 transforms 输出表 diagnosis_rows(split + std_diag lookup + normalize/substring/coalesce)
+#
+# code 解析规则(用户定,2026-07-18):
+#   linkCode 非空 → std_diag.std_code K 开头 → 截 K 大类(std_code_k 直通下方共享字典 passthrough);
+#   非 K(DA0E.50 等 ICD-11)→ substring 产物("DA0")不在字典 → _default → code=null(自由文本);
+#   linkCode 空(99.7%)→ 归一中文名 → 共享 diagnosis-kcode 字典救码(与 jvs-dw 完全同一套)。
+# 任何情况下 name=原文保留(展示 + LLM 输入),不丢 fact。
+canonical: diagnosis
+
+emits:
+  action: diagnosis_recorded
+  subjectType: diagnosis
+  occurredAtField: occurredAt
+
+primary:
+  table: diagnosis_rows
+  key: diag_external_id
+  dedup_by: diag_external_id
+
+field_mapping:
+  externalId:                   diag_external_id
+  updatedAt:                    updated_gmt_at
+  createdAt:                    created_gmt_at
+  patientExternalId:            patient_id
+  clinicId:                     organization_id
+  occurredAt:                   clinic_time      # 就诊时刻(100% 填充)
+  sourceEncounterExternalId:    emr_sub_id       # 同诊聚合锚(病历不普遍挂预约,用小病历号)
+  code:                         diag_code_src    # K 码 或 归一中文名 → 共享字典翻译
+  stdCodeRaw:                   std_code         # 溯源:字典 join 到的原始 std_code(可 null)
+  name:                         value            # 原始中文名(展示 + LLM 输入)
+  toothPosition:                tooth_position   # "41;42" / "42 L;43 D"(FDI+牙面,分号分隔)
+  doctorId:                     user_id
+  doctorName:                   user_name
+
+# ── 字典分层:临床通用层(K 直通 + 中文术语→K 码 + keyword 兜底)全在共享文件 ──
+dict_includes:
+  - ../../_shared/dict/diagnosis-kcode.yaml
+
+enum_mapping:
+  code:
+    # 兜底:共享 enum/keyword 都翻不出 → 空 → code=null(不丢 fact,name 保留)
+    _default: ""

apps/pac-service/data/friday/assemblers/emr.yaml

+# emr — FRIDAY 病历自由文本(诊断/治疗已拆独立 fact,本资源只装自由文本)
+# 直接消费 med_emr_info(数组列不映射即不入 canonical;dispose_text/examine_text 是导出侧
+# 序列化的 JSON 字符串形态,与 jvs-dw examine/dispose "JSON 原文" 同语义)
+canonical: emr
+
+emits:
+  action: emr_submitted
+  subjectType: emr
+  # occurred = clinic_time(就诊时刻,100% 填充);created_gmt_at 是录入时点,留幂等键用
+  occurredAtField: submittedAt
+
+primary:
+  table: med_emr_info
+  key: emr_sub_id      # 小病历号 uuid(两侧连接键 + 临床事实锚点)
+  dedup_by: emr_sub_id
+
+field_mapping:
+  externalId:           emr_sub_id
+  # updatedAt 进 source_event_id 幂等键,host UPDATE → 新 transaction(同 jvs-dw 纪律)
+  updatedAt:            updated_gmt_at
+  createdAt:            created_gmt_at
+  patientExternalId:    patient_id
+  clinicId:             organization_id
+  submittedAt:          clinic_time       # 就诊时刻(100%)
+  encounterExternalId:  appointment_id    # 56.5% 有值 → 挂预约/接诊链;临床事实锚用 emr_sub_id
+  doctorId:             user_id
+  doctorName:           user_name
+  # SOAP 自由文本段落
+  illnessDesc:          illness_desc      # 主诉
+  preIllness:           pre_illness       # 现病史
+  pastHistory:          past_hist         # 既往史
+  generalCondition:     gen_cond          # 全身情况
+  examFindings:         examine_text      # 检查所见(JSON 原文,带牙位)
+  auxExamFindings:      auxi_exam         # 辅助检查(自由文本)
+  disposal:             dispose_text      # 处置叙述(98% 填充,JSON 原文带牙位;actual 治疗主源是 treat[])
+  doctorAdvice:         doc_order         # 医嘱

apps/pac-service/data/friday/assemblers/image.yaml

+# image — FRIDAY 影像 metadata(MySQL emr.med_check;PAC 不持文件本体)
+# 影像不在 Mongo 病历文档内(全字段枚举证实),挂 emr_id(10 位病历号)回病历。
+# 消费 transforms 输出表 image_rows(med_check lookup std_check_class 附 class_name)
+canonical: image
+
+emits:
+  action: image_uploaded
+  subjectType: image
+  occurredAtField: uploadedAt
+
+primary:
+  table: image_rows
+  key: id
+  dedup_by: id
+
+field_mapping:
+  externalId:           id
+  updatedAt:            updated_gmt_at
+  createdAt:            created_gmt_at
+  patientExternalId:    patient_id
+  clinicId:             organization_id
+  uploadedAt:           shooting_time     # 拍摄时间
+  encounterExternalId:  emr_id            # 病历号级挂载(= med_emr_info.emr_id,非 emr_sub_id)
+  modality:             class_name        # lookup 自 std_check_class 字典
+  fileUrl:              file_url          # OSS 相对路径(passthrough)
+  fileName:             file_name
+
+# ── 字典分层:真影像类型名共享(与 jvs-dw 实测同名);FRIDAY 特有名目留本地 ──
+dict_includes:
+  - ../../_shared/dict/image-modality.yaml
+
+enum_mapping:
+  modality:
+    # FRIDAY 特有:胸片(987d743a 字典)→ 非口腔影像归 other;
+    # 3a9d9c68 字典乱录("测试"/"1"/"11")与 lookup 未命中(class_name=null)→ _default
+    胸片:                         other
+    _default:                     other

apps/pac-service/data/friday/assemblers/treatment_actual.yaml

+# treatment_actual — FRIDAY 本次治疗(med_emr_info.treat[] 拆行)
+# 语义(用户确认,2026-07-18):treat[] = 本次治疗(actual);病历无计划字段(全字段枚举证实),
+# 治疗计划走独立表 customer_treat_plan(见 treatment_planned.yaml)。
+# 覆盖率注:treat 填充 25%(dispose 98% 是处置叙述 → emr_record.disposal 全文保留不丢);
+# 若实践发现漏召,Phase 2 评估 dispose keyword 路由补 actual。
+canonical: treatment
+
+emits:
+  action: treatment_completed
+  subjectType: treatment
+  occurredAtField: occurredAt
+  kind: actual
+
+primary:
+  table: treatment_actual_rows
+  key: treat_external_id
+  dedup_by: treat_external_id
+
+field_mapping:
+  externalId:                  treat_external_id
+  updatedAt:                   updated_gmt_at
+  createdAt:                   created_gmt_at
+  patientExternalId:           patient_id
+  clinicId:                    organization_id
+  occurredAt:                  clinic_time      # 就诊时刻(100% 填充)
+  sourceEncounterExternalId:   emr_sub_id       # 同诊聚合锚
+  category:                    category_raw     # 归一治疗名 → 共享字典翻 11 类闭集
+  subtype:                     value            # 原始 treat 文本留作 subtype(chain milestone fallback)
+  toothPosition:               tooth_position
+  doctorId:                    user_id
+  doctorName:                  user_name
+
+# ── 字典分层:临床核心词 + keyword/strip(actual 语义变体)全在共享文件;FRIDAY 尚无宿主特有词 ──
+# (测试库 treat 一半键盘乱敲 → 高频宿主词等生产数据回来后按 mapping-miss 巡检补这里)
+dict_includes:
+  - ../../_shared/dict/treatment-category-core.yaml
+  - ../../_shared/dict/treatment-category-actual-rules.yaml
+
+enum_mapping:
+  category:
+    # 长尾兜底:未命中 → category='',parser 跳过该 fact(数据损失换质量,jvs-dw 同款纪律)
+    _default: ""

apps/pac-service/data/friday/assemblers/treatment_planned.yaml

+# treatment_planned — FRIDAY 治疗计划(customer_treat_plan + _item 独立表;用户确认此表为治疗计划)
+# 消费 transforms 输出表 treatment_planned_rows(行表 lookup 头表拿 organization_id/plan_name)
+# 行粒度 = 计划行项(牙位 + 术式 mode_name + 价格区间);externalId = 行表主键(host 稳定)
+canonical: treatment
+
+emits:
+  action: treatment_planned
+  subjectType: treatment
+  occurredAtField: occurredAt
+  kind: planned
+
+primary:
+  table: treatment_planned_rows
+  key: id
+  dedup_by: id
+
+field_mapping:
+  externalId:                  id
+  updatedAt:                   updated_gmt_at
+  createdAt:                   created_gmt_at
+  patientExternalId:           customer_id
+  clinicId:                    organization_id   # lookup 自头表(行表无诊所)
+  occurredAt:                  created_gmt_at    # 计划制定时点
+  category:                    mode_norm         # 归一术式名 → 共享字典翻 11 类闭集
+  subtype:                     mode_name         # 原始术式名
+  toothPosition:               tooth_position
+  planName:                    plan_name         # 方案名(passthrough,展示用)
+
+# ── 字典分层:planned 语义变体(不剥"建议/拟"措辞,只剥条件词)──
+dict_includes:
+  - ../../_shared/dict/treatment-category-core.yaml
+  - ../../_shared/dict/treatment-category-planned-rules.yaml
+
+enum_mapping:
+  category:
+    _default: ""

apps/pac-service/data/friday/export.sh

+#!/usr/bin/env bash
+# FRIDAY SaaS 测试环境 → PAC cold-import 导出(等价"host 侧单表 dump")
+#
+# 纪律(同 jvs-dw SQL 朴素化):每条导出 = 列选 + WHERE,零 join/零变换;
+#   唯一例外是 tenant_id AS brand_id 别名(避免与 PAC tenant 概念撞名,与既有 CSV 一致)。
+#   形态改造(拆 JSON 数组/字典 join/K 码截取)全部在 PAC transforms 层做。
+#
+# 凭据走 env(不入 git):
+#   export FRIDAY_MYSQL_HOST=… FRIDAY_MYSQL_PORT=3309 FRIDAY_MYSQL_USER=… FRIDAY_MYSQL_PASSWORD=…
+#   export FRIDAY_MONGO_URI='mongodb://user:pass@host:27017/emr_market'   # 密码含 & 等须 URL 编码
+#
+# 依赖:docker(mysql:8 / mongo:7 客户端镜像),产物落本目录(.gitignore 已挡,不入 git)。
+set -euo pipefail
+cd "$(dirname "$0")"
+
+: "${FRIDAY_MYSQL_HOST:?need FRIDAY_MYSQL_HOST}"
+: "${FRIDAY_MYSQL_PORT:=3309}"
+: "${FRIDAY_MYSQL_USER:?need FRIDAY_MYSQL_USER}"
+: "${FRIDAY_MYSQL_PASSWORD:?need FRIDAY_MYSQL_PASSWORD}"
+: "${FRIDAY_MONGO_URI:?need FRIDAY_MONGO_URI}"
+
+# ── MySQL → CSV(TSV 中转,node 处理 mysql --batch 的 \t \n \\ 转义再按 RFC4180 引号)──
+mysql_csv() {
+  local sql="$1" out="$2"
+  docker run --rm mysql:8 mysql \
+    -h"$FRIDAY_MYSQL_HOST" -P"$FRIDAY_MYSQL_PORT" -u"$FRIDAY_MYSQL_USER" -p"$FRIDAY_MYSQL_PASSWORD" \
+    --default-character-set=utf8mb4 --batch --raw=FALSE --connect-timeout=20 -e "$sql" \
+    | node -e '
+      const lines = require("fs").readFileSync(0, "utf-8").split("\n");
+      const unesc = (s) => { let r = ""; for (let i = 0; i < s.length; i++) { const c = s[i];
+        if (c === "\\" && i + 1 < s.length) { const n = s[++i];
+          r += n === "n" ? "\n" : n === "t" ? "\t" : n === "0" ? "\0" : n === "N" ? "" : n; }
+        else r += c; } return r; };
+      const q = (s) => /[",\n\r]/.test(s) ? `"${s.replace(/"/g, `""`)}"` : s;
+      const out = [];
+      for (const line of lines) { if (line === "") continue;
+        out.push(line.split("\t").map((f) => q(f === "NULL" ? "" : unesc(f))).join(",")); }
+      process.stdout.write(out.join("\n") + "\n");
+    ' > "$out"
+  echo "  $out: $(($(wc -l < "$out") - 1)) rows"
+}
+
+echo "── MySQL 导出 ──"
+# 患者主档(全量;既有 450 行样本 → 扩全量,EMR 覆盖 ~1.65 万患者的前置)
+mysql_csv "SELECT id,name,sex,birthday,file_number,tenant_id AS brand_id,organization_id,created_gmt_at,updated_gmt_at FROM customer.customer_basic_info" customer_basic_info.csv
+# 联系方式 1:N(lookup 挑默认号 → patient.phone)
+mysql_csv "SELECT id,customer_id,contacts_tel,is_default,contacts_type,tel_type,relationship FROM customer.customer_contacts" customer_contacts.csv
+# 预约(全状态;transforms 丢草稿 10)
+mysql_csv "SELECT id,patient_appointment_id,organization_id,tenant_id AS brand_id,appointment_date,appointment_start,appointment_status,doctor_user_id,appointment_time_length,in_time,created_gmt_at,updated_gmt_at FROM \`arrail-appointment-server\`.appointment_base" appointment_base.csv
+# 影像档案(挂 emr_id 病历号;modality 经 std_check_class 字典翻)
+mysql_csv "SELECT id,tenant_id AS brand_id,organization_id,patient_id,emr_id,class_code,file_name,file_type,file_url,files_size,shooting_time,created_gmt_at,updated_gmt_at FROM emr.med_check" med_check.csv
+# 影像类型字典(class_code UUID 全局唯一,lookup 不需 tenant 限定)
+mysql_csv "SELECT class_code,class_name,tenant_id AS brand_id FROM emr.std_check_class" std_check_class.csv
+# 诊断字典(linkCode → std_code;K 码用,非 K 落自由文本)
+mysql_csv "SELECT diag_code,diag_name,std_code FROM emr.std_diag" std_diag.csv
+# 治疗计划(头:方案名/期望;行:牙位/术式/价格区间)→ treatment_planned
+mysql_csv "SELECT id,tenant_id AS brand_id,organization_id,customer_id,plan_group_id,plan_name,desired_effect,emergency_urgency,expect_cost,created_gmt_at,updated_gmt_at FROM customer.customer_treat_plan" customer_treat_plan.csv
+mysql_csv "SELECT id,tenant_id AS brand_id,customer_id,treat_plan_id,tooth_position,mode_name,charge_min,charge_max,created_gmt_at,updated_gmt_at FROM customer.customer_treat_plan_item" customer_treat_plan_item.csv
+
+echo "── Mongo 导出(med_emr_info,status∈{3,4} 正式病历)──"
+# 不用 mongoexport:Extended JSON 会把 Date/Long 包成 {$date}/{$numberLong} 对象,破坏平铺契约。
+# mongosh 手动平铺:Date→ISO 字符串,Long→字符串,数组(diag/treat/dispose/examine)原样保留
+# (PAC split_json_array 原生支持已 parse 数组)。产出 JSON 数组文件(cold-import json 格式)。
+docker run --rm mongo:7 mongosh "$FRIDAY_MONGO_URI" --quiet --eval '
+  const flat = (v) => v instanceof Date ? v.toISOString()
+    : (v && v.constructor && v.constructor.name === "Long") ? String(v)
+    : v === undefined ? null : v;
+  // id 类字段字符串化:Mongo 存的是 number/Long 混型,canonical 契约要 string
+  // (CSV 源天然全字符串;JSON 导出须显式对齐,否则 patientExternalId 等校验失败)
+  const sid = (v) => (v === undefined || v === null || v === "") ? null : String(flat(v));
+  let first = true;
+  print("[");
+  db.med_emr_info.find({ status: { $in: [3, 4] } }).forEach((d) => {
+    const o = {
+      emr_sub_id: sid(d.emrSubId), emr_id: sid(d.emrId),
+      appointment_id: sid(d.appointmentId),
+      brand_id: sid(d.tenantId), organization_id: sid(d.organizationId),
+      patient_id: sid(d.patientId), user_id: sid(d.userId), user_name: flat(d.userName),
+      clinic_time: flat(d.clinicTime), visit_indicator: flat(d.visitIndicator), status: flat(d.status),
+      diag: d.diag ?? [], treat: d.treat ?? [], dispose: d.dispose ?? [], examine: d.examine ?? [],
+      // emr_record 自由文本位要字符串(fact 层 zod nullableString);数组原样给 split_json_array 拆行
+      dispose_text: (d.dispose ?? []).length ? JSON.stringify(d.dispose) : null,
+      examine_text: (d.examine ?? []).length ? JSON.stringify(d.examine) : null,
+      illness_desc: flat(d.illnessDesc), pre_illness: flat(d.preIllness), past_hist: flat(d.pastHist),
+      gen_cond: flat(d.genCond), auxi_exam: flat(d.auxiExam), doc_order: flat(d.docOrder),
+      created_gmt_at: flat(d.createdGmtAt), updated_gmt_at: flat(d.updatedGmtAt),
+    };
+    print((first ? "" : ",") + JSON.stringify(o)); first = false;
+  });
+  print("]");
+' > med_emr_info.json
+node -e "const a=JSON.parse(require('fs').readFileSync('med_emr_info.json','utf-8')); console.log('  med_emr_info.json:', a.length, 'docs')"
+
+echo "── 导出完成 ──"
+ls -la *.csv *.json | awk '{print "  " $NF, $5}'

apps/pac-service/data/friday/manifest.yaml

@@ -31,6 +31,15 @@ tables:
   - { table: customer_basic_info, file: customer_basic_info.csv }   # 患者主档(原始)
   - { table: customer_contacts,   file: customer_contacts.csv }     # 联系方式 1:N(原始)
   - { table: appointment_base,    file: appointment_base.csv }      # 预约(原始,全状态)
+  # ── EMR 病历链(W FRIDAY-EMR:Mongo med_emr_info status∈{3,4} 正式病历;export.sh 平铺导出)──
+  # 数组字段(diag/treat/dispose/examine)导出保留原生数组(split_json_array 原生支持);
+  # dispose_text/examine_text 是同数据的 JSON 字符串形态(emr_record 自由文本位要 string)。
+  - { table: med_emr_info,            file: med_emr_info.json }     # 病历正文(Mongo,含 diag/treat 数组)
+  - { table: med_check,               file: med_check.csv }         # 影像档案(挂 emr_id 病历号)
+  - { table: std_check_class,         file: std_check_class.csv }   # 影像类型字典(class_code→class_name)
+  - { table: std_diag,                file: std_diag.csv }          # 诊断字典(linkCode→std_code;K 码用)
+  - { table: customer_treat_plan,     file: customer_treat_plan.csv }      # 治疗计划头(方案名/期望)
+  - { table: customer_treat_plan_item, file: customer_treat_plan_item.csv } # 治疗计划行(牙位/术式/价格)
 
 transforms:
   # ── 患者:把 1:N 联系方式归约成 patient.phone 标量(无需源侧 join)──
@@ -73,7 +82,148 @@ transforms:
       appointment_status: { in: ['4', '5', '6'] }
       in_time: { not_empty: true }
 
+  # ═══════════ EMR 病历链(同 jvs-dw 分化模式:一张病历源喂多 canonical)═══════════
+  # med_emr_info → emr_record(自由文本) + diagnosis(diag[] 拆行) + treatment_actual(treat[] 拆行)
+  # 锚点:临床事实 sourceEncounterExternalId=emr_sub_id(同诊聚合);emr 自身挂 appointment_id(56.5%)。
+
+  # ── E.1 病历自由文本行(emr.yaml 直接消费 med_emr_info,无需 project:
+  #        assembler field_mapping 天然只取所需列,数组列不映射即不入 canonical)──
+
+  # ── E.2 diag[] 拆行 → 诊断候选 ──
+  - kind: split_json_array
+    input: med_emr_info
+    output: _diag_raw
+    array_field: diag
+    parent_keys:
+      emr_sub_id: emr_sub_id
+      emr_id: emr_id
+      brand_id: brand_id
+      organization_id: organization_id
+      patient_id: patient_id
+      user_id: user_id
+      user_name: user_name
+      clinic_time: clinic_time
+      created_gmt_at: created_gmt_at
+      updated_gmt_at: updated_gmt_at
+    element_fields:
+      value: value
+      tooth_position: toothPosition
+      link_code: linkCode
+    where:
+      value: { not_empty: true }    # 只要求有诊断名;linkCode 可空(仅 0.3% 有)
+
+  # E.2.1 linkCode → std_diag 字典(diag_code UUID 全局唯一,LEFT JOIN 未命中 → std_code=null)
+  - kind: lookup
+    input: _diag_raw
+    output: _diag_raw            # in-place 附加 std_code
+    from: std_diag
+    left_key: link_code
+    right_key: diag_code
+    select:
+      std_code: std_code
+
+  # E.2.2 归一 + K 码截取 + coalesce
+  # 规则(用户定):std_code K 开头 → 截 K 大类;非 K(DA0E.50 等 ICD-11)/join 不上/linkCode 空
+  #   → 自由文本处理:substring 产物("DA0"等)不在共享 K 码字典 → 落 _default → code=null,name 保留。
+  #   linkCode 空(99.7%)→ coalesce 落到归一中文名 → 共享 diagnosis-kcode 字典救码。
+  - kind: derive
+    input: _diag_raw
+    output: _diag_norm
+    fields:
+      value_norm:
+        op: normalize
+        from: value
+      std_code_k:
+        op: substring
+        from: std_code
+        start: 0
+        end: 3
+  - kind: derive
+    input: _diag_norm
+    output: diagnosis_rows
+    fields:
+      diag_code_src:
+        op: coalesce
+        from: ['std_code_k', 'value_norm']
+      diag_external_id:
+        op: concat
+        parts: ['${emr_sub_id}', '|', '${std_code_k}', '|', '${value_norm}', '|', '${tooth_position}']
+
+  # ── E.3 treat[] 拆行 → 本次治疗(actual;用户确认 treat=本次治疗,病历无计划字段)──
+  # 注:treat 填充仅 25%(dispose 98% 是处置叙述,整段进 emr_record.disposal 不丢);
+  #    若后续漏召,可加 dispose keyword 路由补 actual(Phase 2 评估)。
+  - kind: split_json_array
+    input: med_emr_info
+    output: _treat_raw
+    array_field: treat
+    parent_keys:
+      emr_sub_id: emr_sub_id
+      emr_id: emr_id
+      brand_id: brand_id
+      organization_id: organization_id
+      patient_id: patient_id
+      user_id: user_id
+      user_name: user_name
+      clinic_time: clinic_time
+      created_gmt_at: created_gmt_at
+      updated_gmt_at: updated_gmt_at
+    element_fields:
+      value: value
+      tooth_position: toothPosition
+    where:
+      value: { not_empty: true }
+  - kind: derive
+    input: _treat_raw
+    output: treatment_actual_rows
+    fields:
+      category_raw:
+        op: normalize
+        from: value
+      treat_external_id:
+        op: concat
+        parts: ['${emr_sub_id}', '|treat|', '${value}', '|', '${tooth_position}']
+
+  # ── E.4 影像:med_check + 类型字典 → class_name(modality 经共享字典翻)──
+  - kind: lookup
+    input: med_check
+    output: image_rows
+    from: std_check_class
+    left_key: class_code
+    right_key: class_code
+    select:
+      class_name: class_name
+
+  # ── E.5 治疗计划:行表 join 头表拿 诊所/方案名 → planned 候选 ──
+  - kind: lookup
+    input: customer_treat_plan_item
+    output: _plan_item_joined
+    from: customer_treat_plan
+    left_key: treat_plan_id
+    right_key: id
+    select:
+      organization_id: organization_id
+      plan_name: plan_name
+  # 头表 lookup 未命中 / 头表无诊所 → 计划行无法归属诊所,显式丢弃(canonical clinicId 必填;
+  # 留着只会变 failed 噪音。测试库实测 75/187 行如此)
+  - kind: filter
+    input: _plan_item_joined
+    output: _plan_item_with_org
+    where:
+      organization_id: { not_empty: true }
+  - kind: derive
+    input: _plan_item_with_org
+    output: treatment_planned_rows
+    fields:
+      mode_norm:
+        op: normalize
+        from: mode_name
+
 assemblers:
   - { file: assemblers/patient.yaml }
   - { file: assemblers/appointment.yaml }
   - { file: assemblers/encounter.yaml }
+  - { file: assemblers/emr.yaml }
+  - { file: assemblers/diagnosis.yaml }
+  - { file: assemblers/treatment_actual.yaml }
+  - { file: assemblers/treatment_planned.yaml }
+  - { file: assemblers/image.yaml }