wangsli/hyb-oa
1
0
派生 0
代码 工单 合并请求 工作流 5 软件包 项目 发布 百科 活动

P2OCRS 收口图片 OCR 样票治理

浏览代码
这个提交包含在：
wangsli
2026-05-16 10:20:25 +08:00
父节点 53251010c2
当前提交 0b0cc69897
共修改 4 个文件，包含 19 行新增和 3 行删除
下载 Patch 文件 下载 Diff 文件 展开所有文件 折叠所有文件
docs/trialrun/人工试运行报告.md
+3
查看文件
@@ -20,6 +20,7 @@ DCL7 已删除本文件中 R70-R91 的长流水历史，避免旧环境、旧开
- 2026-05-16 P2D 旧口径已修正：live smoke 凭据、员工 OCR 凭据和授权 PDF 样票已具备，不再作为 P2-FIN/P2-OCR 的 HOLD 原因。P2-ORG 已完成只读巡检：当前连接数据源 high=0、medium=0、low=9，建议 `NO_WRITE_FIX_NEEDED`；低风险项为 CEO/root 无直属主管观察 1 和 active 员工无企业微信绑定 8。
- 2026-05-16 P2-FIN-DECISION 已完成付款批次启用决策准备：只读核对代码/flag/空态、角色账号、导出字段、审计事件、公网 readiness 和 flag off 回滚口径，结论为 `GO_TO_IMPLEMENT`，后续可另建 `P2-FIN-V1`；当前仍不启用 `payment_batch_v1`、不写付款批次业务数据、不部署。
- 2026-05-16 P2-FIN-V1 已完成付款批次启用前本地/隔离硬化：后端 focused 回归、前端 typecheck/build、隔离付款批次闭环、权限直连审计、公网 readiness 和只读 review 均通过；本轮只修补 E2E 对当前状态胶囊和二次确认 UI 的断言，不改业务 API/schema/UI，不部署，公网 `payment_batch_v1` 继续关闭。
- 2026-05-16 P2-OCR-SAMPLE 已完成图片 OCR 样票治理和实现前置确认：当前 PDF-only 口径、`SAMPLE-IMG-*` 授权登记、脱敏证据、provider 分层和回滚设计已锁定；结论为 `HOLD_FOR_SAMPLE`，授权图片样票和真实图片 provider 回归证据形成前不进入 `P2-OCR-V1`，不开放图片 OCR。
## 关键证据索引
@@ -47,6 +48,7 @@ DCL7 已删除本文件中 R70-R91 的长流水历史，避免旧环境、旧开
| P2-ORG | 已收口：组织账号主数据只读巡检完成，高/中风险为 0，低风险观察 9，不需要写修复合同 | `issues/2026-05-16_02-04-27-p2-org-readonly-master-data-audit.csv`、`artifacts/qa-flow/p2-org-readonly-audit-20260516-020427/report.json` |
| P2-FIN-DECISION | 已收口：付款批次决策准备完成，结论 `GO_TO_IMPLEMENT`；后续进入单独 `P2-FIN-V1` 合同前仍需方案批准 | `issues/2026-05-16_09-02-42-p2-fin-payment-batch-decision.csv` |
| P2-FIN-V1 | 已收口：付款批次启用前本地/隔离硬化完成；隔离付款批次闭环 1 passed、权限直连审计 8 passed、公网 readiness ready | `issues/2026-05-16_09-42-43-p2-fin-payment-batch-v1.csv`、`artifacts/qa-flow/business-flow-20260516-095719/summary.txt`、`artifacts/qa-flow/business-flow-20260516-095900/summary.txt` |
| P2-OCR-SAMPLE | 已收口：图片 OCR 样票治理前置完成，结论 `HOLD_FOR_SAMPLE`；PDF live OCR 基线继续有效但不代表图片 OCR 通过 | `issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv` |
## 当前门禁
@@ -64,6 +66,7 @@ DCL7 已删除本文件中 R70-R91 的长流水历史，避免旧环境、旧开
| P2-ORG 只读巡检 | 通过，含低风险观察项 | `server/src/scripts/audit-org-master-data.js` 在 READ ONLY 事务中执行；当前连接数据源 high=0、medium=0、low=9，未输出手机号、邮箱、token 或企业微信真实标识，报告 `artifacts/qa-flow/p2-org-readonly-audit-20260516-020427/` |
| P2-FIN 决策准备 | 通过 | `LIVE_SMOKE_BASE_URL=https://oa.hyinda.com node scripts/qa/check-live-smoke-env.mjs` 显示 publicReady/liveSmokeReady/liveOcrReady 均为 true；只读核对付款批次 flag、权限、导出、审计和回滚口径；未执行付款批次业务写入 |
| P2-FIN-V1 启用前硬化 | 通过，未部署 | 后端 focused 回归 8 tests passed；`client` typecheck/build 通过；隔离临时库 `reimbursement-payment-batch-flow` 1 passed，覆盖创建批次、复核、二次确认执行、报销 paid、票据 archived、AP 台账和 ERP outbox；权限直连审计 8 passed；公网只做 readiness，不启用 `payment_batch_v1` |
| P2-OCR-SAMPLE 样票治理 | HOLD，未部署 | 已确认当前发票 OCR 仍为 PDF-only，图片仅作为普通附件；腾讯云 `image_native` 只证明技术路径存在，正式图片 OCR 仍需授权图片样票、脱敏确认和真实图片 provider 回归；未调用真实图片 OCR |
## R7EP / OPS8 / QAF8 / UX8 / P2T 预检推进
docs/文档事实源状态矩阵.md
+3 -2
查看文件
@@ -41,6 +41,7 @@
| `P2-ORG` | `issues/2026-05-16_02-04-27-p2-org-readonly-master-data-audit.csv` | `EVIDENCE` | 组织账号主数据只读巡检已完成；当前连接数据源高/中风险为 0，仅有低风险观察项，不需要写修复合同 |
| `P2-FIN-DECISION` | `issues/2026-05-16_09-02-42-p2-fin-payment-batch-decision.csv` | `EVIDENCE` | 付款批次启用决策准备已完成，结论 `GO_TO_IMPLEMENT`；后续必须另建 `P2-FIN-V1` 实现合同，当前不启用 `payment_batch_v1` |
| `P2-FIN-V1` | `issues/2026-05-16_09-42-43-p2-fin-payment-batch-v1.csv` | `EVIDENCE` | 付款批次启用前本地/隔离硬化已完成；后端 focused 回归、前端构建、隔离闭环、权限审计和公网 readiness 通过，公网仍不启用 `payment_batch_v1` |
| `P2-OCR-SAMPLE` | `issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv` | `EVIDENCE` | 图片 OCR 样票治理和实现前置确认已完成；结论 `HOLD_FOR_SAMPLE`，当前保持 PDF-only，不开放图片 OCR |
旧 Issue CSV 保留为历史审计证据：不删除、不批量回写、不移动到 `docs/archive/` 或其他归档目录。新连续合同如替代旧合同，只在新合同 `Notes` 或文档索引中说明 supersedes。
@@ -48,8 +49,8 @@
| Doc_ID | Path | Category | Authority | Status | Owner | Related_Issues | Evidence | Supersedes | Last_Audited | Audit_Method | Risk | Notes |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| `IDX` | `docs/文档索引.md` | 入口索引 | 最高文档入口 | `CURRENT` | product/dev | `GOV8,QAOPS9,P2D,P2-ORG,P2-FIN-DECISION,P2-FIN-V1` | `rg` | none | 2026-05-16 | `node scripts/qa/audit-docs.mjs --check-registry` | medium | 判断当前该读什么 |
| `REG` | `docs/文档事实源状态矩阵.md` | 文档台账 | 文档分类和状态事实源 | `CURRENT` | product/dev | `GOV8,QAOPS9,P2D,P2-ORG,P2-FIN-DECISION,P2-FIN-V1` | `audit-docs` | none | 2026-05-16 | `audit-docs --check-registry` | medium | 判断文档权威性 |
| `IDX` | `docs/文档索引.md` | 入口索引 | 最高文档入口 | `CURRENT` | product/dev | `GOV8,QAOPS9,P2D,P2-ORG,P2-FIN-DECISION,P2-FIN-V1,P2-OCR-SAMPLE` | `rg` | none | 2026-05-16 | `node scripts/qa/audit-docs.mjs --check-registry` | medium | 判断当前该读什么 |
| `REG` | `docs/文档事实源状态矩阵.md` | 文档台账 | 文档分类和状态事实源 | `CURRENT` | product/dev | `GOV8,QAOPS9,P2D,P2-ORG,P2-FIN-DECISION,P2-FIN-V1,P2-OCR-SAMPLE` | `audit-docs` | none | 2026-05-16 | `audit-docs --check-registry` | medium | 判断文档权威性 |
| `PRD-MVP` | `docs/海音达OA-MVP需求规格说明书.md` | 需求事实源 | PRD 主事实源 | `CURRENT` | product | `REQ8,P2T,GOV8` | Issue CSV | `docs/requirements-specification.md` | 2026-05-13 | `rg` | high | 定义做什么、不做什么和验收口径 |
| `PRODUCT-SCOPE` | `docs/产品口径.md` | 需求事实源 | 产品边界事实源 | `CURRENT` | product | `REQ8,P2T` | `rg` | none | 2026-05-13 | `rg` | high | 记录已拍板产品边界 |
| `DESIGN-UI` | `docs/产品设计规则.md` | 设计事实源 | PC/移动/企业微信设计事实源 | `CURRENT` | design/frontend | `MIA7,MIF7,GOV8,DG8,DG9` | Playwright/视觉审查/DG8/DG9 | none | 2026-05-13 | `rg` | high | 页面改动必须读取；DG8 后飞书为主 UI/布局参考；DG9 后按页面分层和样板选型 |
docs/文档索引.md
+3 -1
查看文件
@@ -20,7 +20,7 @@
## 最近事实源链
R6 -> PC6 -> PA6 -> FL6 -> AP6 -> RI6 -> LP6 -> ME6/QG6 -> DG7/MRF7 -> MIA7 -> MIF7/MIF7T(公网部署与带凭据公网复检已收口) -> DG8 -> DG9 -> R7E -> FBA -> QAOPS9/P2D -> P2-ORG -> P2-FIN-DECISION -> P2-FIN-V1。
R6 -> PC6 -> PA6 -> FL6 -> AP6 -> RI6 -> LP6 -> ME6/QG6 -> DG7/MRF7 -> MIA7 -> MIF7/MIF7T(公网部署与带凭据公网复检已收口) -> DG8 -> DG9 -> R7E -> FBA -> QAOPS9/P2D -> P2-ORG -> P2-FIN-DECISION -> P2-FIN-V1 -> P2-OCR-SAMPLE。
- R6、PC6、PA6、FL6、AP6、RI6、LP6 已完成并部署。
- ME6 自动化、公网视觉矩阵和 `ME6-3` 代表性企业微信真机门禁已完成。
@@ -35,6 +35,7 @@ R6 -> PC6 -> PA6 -> FL6 -> AP6 -> RI6 -> LP6 -> ME6/QG6 -> DG7/MRF7 -> MIA7 -> M
- P2-ORG 是 P2D 后第一个执行合同，已完成组织账号主数据只读巡检；当前连接数据源下高/中风险为 0，低风险观察项为 CEO/root 无直属主管 1 和 active 员工无企业微信绑定 8；本轮不需要写修复合同。
- P2-FIN-DECISION 已完成付款批次启用决策准备，结论为 `GO_TO_IMPLEMENT`：可进入单独 `P2-FIN-V1` 实现合同；当前仍不启用 `payment_batch_v1`、不写付款批次业务数据、不部署。
- P2-FIN-V1 已完成付款批次启用前本地/隔离硬化：后端 focused 回归、前端 typecheck/build、隔离付款批次闭环、权限直连审计和公网 readiness 均通过；本轮只硬化 E2E 断言，不新增 API/schema/feature flag/UI，不部署，公网 `payment_batch_v1` 继续关闭。
- P2-OCR-SAMPLE 已完成图片 OCR 样票治理和实现前置确认：当前 PDF-only 口径、样票脱敏、provider 回归分层和回滚条件已锁定；结论为 `HOLD_FOR_SAMPLE`，需授权图片样票和真实图片 provider 回归证据后才可进入 `P2-OCR-V1`。
## 当前活跃治理流
@@ -56,6 +57,7 @@ R6 -> PC6 -> PA6 -> FL6 -> AP6 -> RI6 -> LP6 -> ME6/QG6 -> DG7/MRF7 -> MIA7 -> M
| `P2-ORG` | `issues/2026-05-16_02-04-27-p2-org-readonly-master-data-audit.csv` | 已收口：组织主数据只读巡检完成 | 主管链、停用审批人、角色持有人、admin-only、财务/admin 混授、企业微信绑定漂移的脱敏只读审计 | 不写生产账号数据、不执行 P2-ORG-FIX、不改 API/schema/UI、不部署 |
| `P2-FIN-DECISION` | `issues/2026-05-16_09-02-42-p2-fin-payment-batch-decision.csv` | 已收口：付款批次决策准备完成，结论 `GO_TO_IMPLEMENT` | 付款批次代码/flag/空态、角色账号矩阵、导出字段、审计事件、公网验收窗口和 flag off 回滚口径 | 不启用 `payment_batch_v1`、不写付款批次、不接真实支付、不写 ERP 实时凭证、不部署 |
| `P2-FIN-V1` | `issues/2026-05-16_09-42-43-p2-fin-payment-batch-v1.csv` | 已收口：付款批次启用前本地/隔离硬化完成 | 后端付款批次/AP/导出 focused 回归、前端构建、隔离闭环、权限审计、公网 readiness、flag off 回滚口径 | 不公网启用 `payment_batch_v1`、不部署、不接真实支付、不写 ERP 实时凭证 |
| `P2-OCR-SAMPLE` | `issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv` | 已收口：图片 OCR 样票治理前置完成，结论 `HOLD_FOR_SAMPLE` | PDF-only 当前边界、`SAMPLE-IMG-*` 授权登记口径、脱敏证据规则、mock/PDF/真实图片 provider 分层、feature flag 和回滚设计 | 不开放图片 OCR、不新增入口或 feature flag、不调用真实图片 provider、不保存真实票据、不部署 |
旧 Issue CSV 不删除、不批量回写、不移动到归档目录。若新治理流替代旧合同，只在新 CSV 的 `Notes` 或本索引说明 supersedes，旧文件保留历史审计语义。
issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv
+10
查看文件
@@ -0,0 +1,10 @@
ID,Title,Description,Acceptance,Test_Method,Tools,Dev_Status,Review1_Status,Regression_Status,Files,Dependencies,Notes
"P2OCRS-0","P2-OCR-SAMPLE 合同建档","建立图片 OCR 样票治理和实现前置确认合同；范围限定 Issue CSV、既有事实源状态回填和只读核对，不开放图片 OCR，不改业务 API/schema/UI，不部署。","1) CSV 校验通过；2) 本轮不新增业务 API、schema、feature flag、前端入口或 OCR 能力；3) 发票/OCR 正式入口继续 PDF-only；4) 不保存真实票据或敏感字段；5) 后续 P2-OCR-V1 必须另建实现合同。","受影响流程：员工上传发票、附件图片上传、真实 OCR 样票回归和后续 P2-OCR 派工；验证步骤：python3 /home/shunli/.codex/skills/plan/scripts/validate_issues_csv.py issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv；人工核对本 CSV 不包含业务代码、API/schema、部署、生产恢复、数据库重置或 secret/env 打印。","python3|manual","DONE","DONE","DONE","issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv","[]","ScopeLock=DONE; CSV initial validation passed; scope limited to issues/docs read-only evidence; no API/schema/UI/feature flag/runtime/deploy/prod restore/db reset/secret or env print; rollback=delete this CSV and doc status updates before commit."
"P2OCRS-1","当前 OCR 边界审计","只读核对当前 OCR 边界：PDF-only 上传入口、附件图片规则、腾讯云 image_native 分支、live OCR 脚本和批量样票回归口径。","1) PDF-only 当前验收口径清楚；2) 图片仍作为普通附件，不参与 OCR 或查重；3) 代码存在 image_native 不等于产品已开放；4) live OCR 批量脚本当前仅采集 PDF；5) 结论写入 Notes。","受影响流程：PC/移动发票上传 -> OCR -> 查重 -> 报销添加；附件图片上传 -> 仅作为材料；验证步骤：rg -n '图片 OCR|PDF-only|SAMPLE-IMG|授权图片|image_native|PLAYWRIGHT_OCR_LIVE|真实 OCR|SKIP' docs issues server/src client/tests scripts；人工核对 docs/需求-报销发票OCR规约.md、docs/报销发票附件规则.md、client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts 和 server/src/services/ocr/tencent.js。","rg|manual","DONE","DONE","DONE","docs/需求-报销发票OCR规约.md|docs/报销发票附件规则.md|server/src/services/ocr/tencent.js|client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts|scripts/qa","[""P2OCRS-0""]","Audit=DONE; docs keep PDF-only current acceptance and images as normal attachments; server Tencent transport has image_native branch for non-PDF but product入口未开放; live-invoice-ocr and batch scripts require/collect PDF files; no provider call was made."
"P2OCRS-2","授权图片样票登记口径","确认图片样票治理登记口径：使用 SAMPLE-IMG-* 稳定别名，仓库外保存真实文件名映射、来源授权、用途、授权人和有效期；仓库内只写摘要。","1) 样票别名规则明确；2) 真实文件名和原件不进入 git；3) Issue/报告只记录授权摘要或外部登记位置；4) 无授权图片样票时真实图片 OCR 只能 SKIP；5) 不读取或输出真实票面内容。","受影响流程：后续 P2-OCR 图片样票准备 -> 真实 provider 回归 -> 证据归档；验证步骤：人工核对 docs/报销发票附件规则.md 的样票授权规则和 P2D-3 前置条件；如仓库内无授权图片样票登记，记录 HOLD_FOR_SAMPLE，不伪造 PASS。","manual|rg","DONE","DONE","DONE","docs/报销发票附件规则.md|issues/2026-05-16_00-19-21-p2-decision-followup.csv","[""P2OCRS-1""]","SampleAlias=DONE; rule confirmed: use SAMPLE-IMG-* alias and keep real filename/source/authorization mapping outside repo; repo-visible evidence has authorized PDF sample baseline but no authorized image sample registry; do not read or print invoice content."
"P2OCRS-3","脱敏与证据规则","固定图片 OCR 样票脱敏和证据输出规则，禁止提交真实票面、手机号、税号明文、二维码、完整截图或敏感业务备注。","1) 脱敏字段最小化规则清楚；2) 报告只允许记录别名、格式、大小、类别、OCR 状态、字段命中摘要和失败原因；3) artifacts/output 不得包含未脱敏票面；4) 自动截图如含敏感信息必须移出仓库外；5) 规则写入 Notes。","受影响流程：样票脱敏 -> OCR 回归 -> 报告归档；验证步骤：人工核对 docs/报销发票附件规则.md 样票脱敏和授权章节；rg -n '脱敏|SAMPLE-IMG|真实票据|敏感字段|不得提交' docs/报销发票附件规则.md docs/需求-报销发票OCR规约.md。","manual|rg","DONE","DONE","DONE","docs/报销发票附件规则.md|docs/需求-报销发票OCR规约.md","[""P2OCRS-2""]","EvidencePolicy=DONE; reports may contain alias/format/size/category/status/field-hit summary/failure reason only; no raw invoice, sensitive field plaintext, QR code or full invoice screenshot; no new sample artifact in this round."
"P2OCRS-4","provider 回归设计","设计图片 OCR provider 回归口径，区分 mock、腾讯云真实 provider、PDF 基线和图片样票结论；本轮不调用真实图片 OCR。","1) mock 与真实 provider 结论分离；2) PDF live OCR 基线不代表图片 OCR 通过；3) provider 图片回归必须使用授权图片样票；4) 无授权图片样票记 SKIP；5) 字段归一化、失败、重复和人工复核均有后续验收口径。","受影响流程：图片样票 -> provider 识别 -> 字段归一化 -> 查重 -> 人工复核；验证步骤：人工核对 docs/试运行与自动化验证手册.md 批量样票 OCR 回归设计和 docs/需求-报销发票OCR规约.md BIZ8 图片 OCR 可行性设计；不执行真实 provider 图片调用。","manual|rg","DONE","DONE","DONE","docs/试运行与自动化验证手册.md|docs/需求-报销发票OCR规约.md|client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts","[""P2OCRS-3""]","ProviderPlan=DONE; mock/PDF live/image real provider conclusions must be separated; current PDF live baseline remains valid but does not prove image OCR; real image provider regression requires authorized image sample and must be SKIP without it; no image provider call was made."
"P2OCRS-5","图片 OCR flag 与回滚方案","仅设计图片 OCR 入口控制、feature flag 和回滚口径；本轮不创建或启用新 flag。","1) 当前 PDF-only allowlist 保持不变；2) 后续实现必须能关闭图片入口和图片 OCR flag；3) 回滚后图片按识别失败或人工复核处理，不自动转为附件；4) 不恢复拍照直传或 capture 默认入口；5) 不影响现有附件图片规则。","受影响流程：后续图片 OCR 实现 -> 开关启停 -> 回滚 -> PDF-only 恢复；验证步骤：人工核对 OCR 规约中入口控制和回滚建议；rg -n 'feature flag|PDF-only|capture|回滚|图片入口' docs/需求-报销发票OCR规约.md docs/报销发票附件规则.md client/src server/src。","manual|rg","DONE","DONE","DONE","docs/需求-报销发票OCR规约.md|docs/报销发票附件规则.md|client/src|server/src","[""P2OCRS-4""]","FlagPlan=DONE; keep PDF-only allowlist; future implementation needs controlled image invoice entry, feature flag and rollback to PDF-only; no capture default restoration; no new flag or runtime behavior in this round."
"P2OCRS-6","PDF-only 兼容验收","确认图片样票治理不改变现有 PDF-only 上传、查重、OCR 失败、重复票据和人工复核口径。","1) 正式发票上传仍只接受 PDF；2) 图片附件不参与 OCR/查重；3) PDF live OCR 基线继续独立；4) 未运行真实图片 OCR；5) 如后续实现触及代码，必须补前后端回归。","受影响流程：PDF 发票上传 -> OCR -> 查重 -> 添加报销；图片附件上传 -> 材料查看；验证步骤：rg -n 'pdf_invoice_required|accept|application/pdf|image/png|附件不参与 OCR|file_sha256' server/src client/src client/tests docs；本轮不跑会写公网数据的图片 OCR。","rg|manual","DONE","DONE","DONE","server/src|client/src|client/tests|docs","[""P2OCRS-5""]","Compatibility=DONE; server invoicePool keeps application/pdf and pdf_invoice_required; client invoice inputs use PDF_FILE_ACCEPT while attachment inputs still allow images; docs state attachment images do not participate in OCR/dedup; no code change and no public write."
"P2OCRS-7","样票治理决策输出","输出 P2-OCR-SAMPLE 结论：GO_TO_IMPLEMENT、HOLD_FOR_SAMPLE 或 FREEZE；只有授权图片样票、脱敏、provider 回归和回滚方案都具备时才能进入 P2-OCR-V1。","1) 结论使用统一枚举；2) GO_TO_IMPLEMENT 必须列出已具备前置；3) HOLD_FOR_SAMPLE 必须列出缺口；4) FREEZE 必须说明业务冻结理由；5) 结论不得自动启动实现。","受影响流程：P2-OCR 后续派工；验证步骤：人工核对 P2OCRS-1..6 结论，按 Decision=<GO_TO_IMPLEMENT|HOLD_FOR_SAMPLE|FREEZE>; Rationale=<原因>; Prereq=<前置>; NextContract=<P2-OCR-V1|N/A>; Rollback=<回滚口径>; TestScope=<验收口径> 写入 Notes。","manual","DONE","DONE","DONE","issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv","[""P2OCRS-1"",""P2OCRS-2"",""P2OCRS-3"",""P2OCRS-4"",""P2OCRS-5"",""P2OCRS-6""]","Decision=HOLD_FOR_SAMPLE; Rationale=PDF live OCR baseline, employee OCR credentials and Tencent image_native technical path are available, but authorized image sample registry and real image provider regression evidence are not yet present in repo-safe evidence; Prereq=authorized SAMPLE-IMG alias, desensitization confirmation, provider image regression account/window, field normalization evidence, image OCR flag proposal and PDF-only regression; NextContract=P2-OCR-V1 only after prereq pass; Rollback=do not implement/enable image OCR now, keep PDF-only; TestScope=PDF no regression, image mock and real provider separated, no authorized image sample means SKIP not PASS."
"P2OCRS-8","收口回填与中文提交","回填 P2-OCR-SAMPLE CSV、文档索引、事实源矩阵和人工试运行报告，执行最终校验并提交中文 commit。","1) CSV 状态真实回填；2) docs/文档索引.md、docs/文档事实源状态矩阵.md、docs/trialrun/人工试运行报告.md 必要状态已更新；3) validate、audit-docs、rg 和 git diff --check 通过；4) 中文 commit 引用 P2OCRS；5) 明确未部署。","流程级验证：python3 /home/shunli/.codex/skills/plan/scripts/validate_issues_csv.py issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv；node scripts/qa/audit-docs.mjs --check-registry；rg -n 'P2OCRS|P2-OCR-SAMPLE|图片 OCR|PDF-only|SAMPLE-IMG|HOLD_FOR_SAMPLE|GO_TO_IMPLEMENT|FREEZE|image_native|真实 OCR|SKIP' docs issues server/src client/tests scripts；git diff --check；git status --short。","python3|node|rg|git|manual","DONE","DONE","DONE","issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv|docs/文档索引.md|docs/文档事实源状态矩阵.md|docs/trialrun/人工试运行报告.md","[""P2OCRS-7""]","Closeout=PASS; docs index/matrix/trial report updated; validate_issues_csv passed; audit-docs errors=0 warnings=1 with existing unregistered docs files=24; rg keywords可解释; git diff --check passed; not deployed; no public runtime change."
1 ID Title Description Acceptance Test_Method Tools Dev_Status Review1_Status Regression_Status Files Dependencies Notes
2 P2OCRS-0 P2-OCR-SAMPLE 合同建档 建立图片 OCR 样票治理和实现前置确认合同；范围限定 Issue CSV、既有事实源状态回填和只读核对，不开放图片 OCR，不改业务 API/schema/UI，不部署。 1) CSV 校验通过；2) 本轮不新增业务 API、schema、feature flag、前端入口或 OCR 能力；3) 发票/OCR 正式入口继续 PDF-only；4) 不保存真实票据或敏感字段；5) 后续 P2-OCR-V1 必须另建实现合同。 受影响流程：员工上传发票、附件图片上传、真实 OCR 样票回归和后续 P2-OCR 派工；验证步骤：python3 /home/shunli/.codex/skills/plan/scripts/validate_issues_csv.py issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv；人工核对本 CSV 不包含业务代码、API/schema、部署、生产恢复、数据库重置或 secret/env 打印。 python3|manual DONE DONE DONE issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv [] ScopeLock=DONE; CSV initial validation passed; scope limited to issues/docs read-only evidence; no API/schema/UI/feature flag/runtime/deploy/prod restore/db reset/secret or env print; rollback=delete this CSV and doc status updates before commit.
3 P2OCRS-1 当前 OCR 边界审计 只读核对当前 OCR 边界：PDF-only 上传入口、附件图片规则、腾讯云 image_native 分支、live OCR 脚本和批量样票回归口径。 1) PDF-only 当前验收口径清楚；2) 图片仍作为普通附件，不参与 OCR 或查重；3) 代码存在 image_native 不等于产品已开放；4) live OCR 批量脚本当前仅采集 PDF；5) 结论写入 Notes。 受影响流程：PC/移动发票上传 -> OCR -> 查重 -> 报销添加；附件图片上传 -> 仅作为材料；验证步骤：rg -n '图片 OCR|PDF-only|SAMPLE-IMG|授权图片|image_native|PLAYWRIGHT_OCR_LIVE|真实 OCR|SKIP' docs issues server/src client/tests scripts；人工核对 docs/需求-报销发票OCR规约.md、docs/报销发票附件规则.md、client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts 和 server/src/services/ocr/tencent.js。 rg|manual DONE DONE DONE docs/需求-报销发票OCR规约.md|docs/报销发票附件规则.md|server/src/services/ocr/tencent.js|client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts|scripts/qa ["P2OCRS-0"] Audit=DONE; docs keep PDF-only current acceptance and images as normal attachments; server Tencent transport has image_native branch for non-PDF but product入口未开放; live-invoice-ocr and batch scripts require/collect PDF files; no provider call was made.
4 P2OCRS-2 授权图片样票登记口径 确认图片样票治理登记口径：使用 SAMPLE-IMG-* 稳定别名，仓库外保存真实文件名映射、来源授权、用途、授权人和有效期；仓库内只写摘要。 1) 样票别名规则明确；2) 真实文件名和原件不进入 git；3) Issue/报告只记录授权摘要或外部登记位置；4) 无授权图片样票时真实图片 OCR 只能 SKIP；5) 不读取或输出真实票面内容。 受影响流程：后续 P2-OCR 图片样票准备 -> 真实 provider 回归 -> 证据归档；验证步骤：人工核对 docs/报销发票附件规则.md 的样票授权规则和 P2D-3 前置条件；如仓库内无授权图片样票登记，记录 HOLD_FOR_SAMPLE，不伪造 PASS。 manual|rg DONE DONE DONE docs/报销发票附件规则.md|issues/2026-05-16_00-19-21-p2-decision-followup.csv ["P2OCRS-1"] SampleAlias=DONE; rule confirmed: use SAMPLE-IMG-* alias and keep real filename/source/authorization mapping outside repo; repo-visible evidence has authorized PDF sample baseline but no authorized image sample registry; do not read or print invoice content.
5 P2OCRS-3 脱敏与证据规则 固定图片 OCR 样票脱敏和证据输出规则，禁止提交真实票面、手机号、税号明文、二维码、完整截图或敏感业务备注。 1) 脱敏字段最小化规则清楚；2) 报告只允许记录别名、格式、大小、类别、OCR 状态、字段命中摘要和失败原因；3) artifacts/output 不得包含未脱敏票面；4) 自动截图如含敏感信息必须移出仓库外；5) 规则写入 Notes。 受影响流程：样票脱敏 -> OCR 回归 -> 报告归档；验证步骤：人工核对 docs/报销发票附件规则.md 样票脱敏和授权章节；rg -n '脱敏|SAMPLE-IMG|真实票据|敏感字段|不得提交' docs/报销发票附件规则.md docs/需求-报销发票OCR规约.md。 manual|rg DONE DONE DONE docs/报销发票附件规则.md|docs/需求-报销发票OCR规约.md ["P2OCRS-2"] EvidencePolicy=DONE; reports may contain alias/format/size/category/status/field-hit summary/failure reason only; no raw invoice, sensitive field plaintext, QR code or full invoice screenshot; no new sample artifact in this round.
6 P2OCRS-4 provider 回归设计 设计图片 OCR provider 回归口径，区分 mock、腾讯云真实 provider、PDF 基线和图片样票结论；本轮不调用真实图片 OCR。 1) mock 与真实 provider 结论分离；2) PDF live OCR 基线不代表图片 OCR 通过；3) provider 图片回归必须使用授权图片样票；4) 无授权图片样票记 SKIP；5) 字段归一化、失败、重复和人工复核均有后续验收口径。 受影响流程：图片样票 -> provider 识别 -> 字段归一化 -> 查重 -> 人工复核；验证步骤：人工核对 docs/试运行与自动化验证手册.md 批量样票 OCR 回归设计和 docs/需求-报销发票OCR规约.md BIZ8 图片 OCR 可行性设计；不执行真实 provider 图片调用。 manual|rg DONE DONE DONE docs/试运行与自动化验证手册.md|docs/需求-报销发票OCR规约.md|client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts ["P2OCRS-3"] ProviderPlan=DONE; mock/PDF live/image real provider conclusions must be separated; current PDF live baseline remains valid but does not prove image OCR; real image provider regression requires authorized image sample and must be SKIP without it; no image provider call was made.
7 P2OCRS-5 图片 OCR flag 与回滚方案 仅设计图片 OCR 入口控制、feature flag 和回滚口径；本轮不创建或启用新 flag。 1) 当前 PDF-only allowlist 保持不变；2) 后续实现必须能关闭图片入口和图片 OCR flag；3) 回滚后图片按识别失败或人工复核处理，不自动转为附件；4) 不恢复拍照直传或 capture 默认入口；5) 不影响现有附件图片规则。 受影响流程：后续图片 OCR 实现 -> 开关启停 -> 回滚 -> PDF-only 恢复；验证步骤：人工核对 OCR 规约中入口控制和回滚建议；rg -n 'feature flag|PDF-only|capture|回滚|图片入口' docs/需求-报销发票OCR规约.md docs/报销发票附件规则.md client/src server/src。 manual|rg DONE DONE DONE docs/需求-报销发票OCR规约.md|docs/报销发票附件规则.md|client/src|server/src ["P2OCRS-4"] FlagPlan=DONE; keep PDF-only allowlist; future implementation needs controlled image invoice entry, feature flag and rollback to PDF-only; no capture default restoration; no new flag or runtime behavior in this round.
8 P2OCRS-6 PDF-only 兼容验收 确认图片样票治理不改变现有 PDF-only 上传、查重、OCR 失败、重复票据和人工复核口径。 1) 正式发票上传仍只接受 PDF；2) 图片附件不参与 OCR/查重；3) PDF live OCR 基线继续独立；4) 未运行真实图片 OCR；5) 如后续实现触及代码，必须补前后端回归。 受影响流程：PDF 发票上传 -> OCR -> 查重 -> 添加报销；图片附件上传 -> 材料查看；验证步骤：rg -n 'pdf_invoice_required|accept|application/pdf|image/png|附件不参与 OCR|file_sha256' server/src client/src client/tests docs；本轮不跑会写公网数据的图片 OCR。 rg|manual DONE DONE DONE server/src|client/src|client/tests|docs ["P2OCRS-5"] Compatibility=DONE; server invoicePool keeps application/pdf and pdf_invoice_required; client invoice inputs use PDF_FILE_ACCEPT while attachment inputs still allow images; docs state attachment images do not participate in OCR/dedup; no code change and no public write.
9 P2OCRS-7 样票治理决策输出 输出 P2-OCR-SAMPLE 结论：GO_TO_IMPLEMENT、HOLD_FOR_SAMPLE 或 FREEZE；只有授权图片样票、脱敏、provider 回归和回滚方案都具备时才能进入 P2-OCR-V1。 1) 结论使用统一枚举；2) GO_TO_IMPLEMENT 必须列出已具备前置；3) HOLD_FOR_SAMPLE 必须列出缺口；4) FREEZE 必须说明业务冻结理由；5) 结论不得自动启动实现。 受影响流程：P2-OCR 后续派工；验证步骤：人工核对 P2OCRS-1..6 结论，按 Decision=<GO_TO_IMPLEMENT|HOLD_FOR_SAMPLE|FREEZE>; Rationale=<原因>; Prereq=<前置>; NextContract=<P2-OCR-V1|N/A>; Rollback=<回滚口径>; TestScope=<验收口径> 写入 Notes。 manual DONE DONE DONE issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv ["P2OCRS-1","P2OCRS-2","P2OCRS-3","P2OCRS-4","P2OCRS-5","P2OCRS-6"] Decision=HOLD_FOR_SAMPLE; Rationale=PDF live OCR baseline, employee OCR credentials and Tencent image_native technical path are available, but authorized image sample registry and real image provider regression evidence are not yet present in repo-safe evidence; Prereq=authorized SAMPLE-IMG alias, desensitization confirmation, provider image regression account/window, field normalization evidence, image OCR flag proposal and PDF-only regression; NextContract=P2-OCR-V1 only after prereq pass; Rollback=do not implement/enable image OCR now, keep PDF-only; TestScope=PDF no regression, image mock and real provider separated, no authorized image sample means SKIP not PASS.
10 P2OCRS-8 收口回填与中文提交 回填 P2-OCR-SAMPLE CSV、文档索引、事实源矩阵和人工试运行报告，执行最终校验并提交中文 commit。 1) CSV 状态真实回填；2) docs/文档索引.md、docs/文档事实源状态矩阵.md、docs/trialrun/人工试运行报告.md 必要状态已更新；3) validate、audit-docs、rg 和 git diff --check 通过；4) 中文 commit 引用 P2OCRS；5) 明确未部署。 流程级验证：python3 /home/shunli/.codex/skills/plan/scripts/validate_issues_csv.py issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv；node scripts/qa/audit-docs.mjs --check-registry；rg -n 'P2OCRS|P2-OCR-SAMPLE|图片 OCR|PDF-only|SAMPLE-IMG|HOLD_FOR_SAMPLE|GO_TO_IMPLEMENT|FREEZE|image_native|真实 OCR|SKIP' docs issues server/src client/tests scripts；git diff --check；git status --short。 python3|node|rg|git|manual DONE DONE DONE issues/2026-05-16_10-15-21-p2-ocr-image-sample-governance.csv|docs/文档索引.md|docs/文档事实源状态矩阵.md|docs/trialrun/人工试运行报告.md ["P2OCRS-7"] Closeout=PASS; docs index/matrix/trial report updated; validate_issues_csv passed; audit-docs errors=0 warnings=1 with existing unregistered docs files=24; rg keywords可解释; git diff --check passed; not deployed; no public runtime change.