OINV1 修正OCR批量小样本轮询
这个提交包含在:
@@ -1,4 +1,5 @@
|
||||
import { expect, type Page } from "@playwright/test";
|
||||
import { readFileSync } from "node:fs";
|
||||
import { expect, type APIRequestContext, type Page } from "@playwright/test";
|
||||
|
||||
import type { E2EUser } from "./auth";
|
||||
|
||||
@@ -9,6 +10,9 @@ export type LiveInvoiceOcrResult = {
|
||||
detailText: string;
|
||||
bannerText: string;
|
||||
durationMs: number;
|
||||
uploadMs?: number;
|
||||
terminalWaitMs?: number;
|
||||
transport?: "ui" | "api";
|
||||
invoiceId: string | null;
|
||||
invoiceCode: string | null;
|
||||
invoiceNumber: string | null;
|
||||
@@ -21,6 +25,11 @@ export type SafeLiveInvoiceOcrEvidence = {
|
||||
alias: string;
|
||||
status: LiveInvoiceOcrStatus;
|
||||
durationMs: number;
|
||||
transport?: "ui" | "api";
|
||||
timings?: {
|
||||
uploadMs?: number;
|
||||
terminalWaitMs?: number;
|
||||
};
|
||||
reasonCategory: "available" | "duplicate_or_same_invoice" | "already_used" | "ocr_failed" | "timeout_or_pending" | "unknown";
|
||||
fieldHits: {
|
||||
invoiceId: boolean;
|
||||
@@ -42,6 +51,32 @@ type UploadInvoiceResponse = {
|
||||
availabilityMessage?: string | null;
|
||||
};
|
||||
};
|
||||
type UploadInvoiceData = NonNullable<UploadInvoiceResponse["data"]>;
|
||||
|
||||
type ApiEnvelope<T> = {
|
||||
data?: T;
|
||||
error?: {
|
||||
code?: string;
|
||||
message?: string;
|
||||
};
|
||||
};
|
||||
|
||||
type InvoiceStatusResponse = {
|
||||
items?: InvoiceStatusItem[];
|
||||
};
|
||||
|
||||
type InvoiceStatusItem = {
|
||||
status?: string | null;
|
||||
reason?: string | null;
|
||||
message?: string | null;
|
||||
availabilityReason?: string | null;
|
||||
availabilityMessage?: string | null;
|
||||
invoiceCode?: string | null;
|
||||
invoiceNumber?: string | null;
|
||||
invoiceDate?: string | null;
|
||||
sellerName?: string | null;
|
||||
totalAmount?: string | null;
|
||||
};
|
||||
|
||||
function extractField(detailText: string, label: string) {
|
||||
const escaped = label.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
|
||||
@@ -67,6 +102,11 @@ export function createSafeLiveInvoiceOcrEvidence(result: LiveInvoiceOcrResult, a
|
||||
alias,
|
||||
status: result.status,
|
||||
durationMs: result.durationMs,
|
||||
transport: result.transport,
|
||||
timings: {
|
||||
uploadMs: result.uploadMs,
|
||||
terminalWaitMs: result.terminalWaitMs,
|
||||
},
|
||||
reasonCategory: categorizeLiveInvoiceOcrResult(result),
|
||||
fieldHits: {
|
||||
invoiceId: Boolean(result.invoiceId),
|
||||
@@ -78,6 +118,155 @@ export function createSafeLiveInvoiceOcrEvidence(result: LiveInvoiceOcrResult, a
|
||||
};
|
||||
}
|
||||
|
||||
async function readApiData<T>(response: { ok(): boolean; status(): number; text(): Promise<string> }, context: string) {
|
||||
const rawText = await response.text();
|
||||
const parsed = rawText ? (JSON.parse(rawText) as ApiEnvelope<T>) : {};
|
||||
if (!response.ok()) {
|
||||
const code = parsed.error?.code ?? `HTTP_${response.status()}`;
|
||||
throw new Error(`${context}:${code}`);
|
||||
}
|
||||
return parsed.data as T;
|
||||
}
|
||||
|
||||
function mapApiStatusToResultStatus(status: string | null | undefined): LiveInvoiceOcrStatus {
|
||||
return status === "available" ? "available" : "ocr_failed";
|
||||
}
|
||||
|
||||
function buildApiDetailText(input: {
|
||||
status?: string | null;
|
||||
reason?: string | null;
|
||||
message?: string | null;
|
||||
invoiceCode?: string | null;
|
||||
invoiceNumber?: string | null;
|
||||
invoiceDate?: string | null;
|
||||
sellerName?: string | null;
|
||||
totalAmount?: string | null;
|
||||
}) {
|
||||
return [
|
||||
"票据上传结果",
|
||||
`状态:${input.status ?? "-"}`,
|
||||
`原因:${input.reason ?? "-"}`,
|
||||
`提示:${input.message ?? "-"}`,
|
||||
`发票号:${[input.invoiceCode, input.invoiceNumber].filter(Boolean).join("-") || "-"}`,
|
||||
`销售方:${input.sellerName ?? "-"}`,
|
||||
`总金额:${input.totalAmount ?? "-"}`,
|
||||
`开票日期:${input.invoiceDate ?? "-"}`,
|
||||
].join("\n");
|
||||
}
|
||||
|
||||
export async function uploadLiveInvoiceViaApiAndWait(
|
||||
api: APIRequestContext,
|
||||
invoiceFilePath: string,
|
||||
alias: string,
|
||||
timeout = 60_000,
|
||||
): Promise<LiveInvoiceOcrResult> {
|
||||
const startedAt = Date.now();
|
||||
const uploadStartedAt = Date.now();
|
||||
const uploadResponse = await api.post("/api/invoice-pool/upload", {
|
||||
multipart: {
|
||||
ocr_type: "vat",
|
||||
file: {
|
||||
name: `${alias}.pdf`,
|
||||
mimeType: "application/pdf",
|
||||
buffer: readFileSync(invoiceFilePath),
|
||||
},
|
||||
},
|
||||
});
|
||||
const uploadMs = Date.now() - uploadStartedAt;
|
||||
const uploadData = await readApiData<UploadInvoiceData>(uploadResponse, "upload_invoice");
|
||||
const uploadedInvoiceId = uploadData?.invoicePoolId ?? null;
|
||||
|
||||
if (!uploadedInvoiceId) {
|
||||
const message =
|
||||
uploadData?.unavailableMessage ??
|
||||
uploadData?.availabilityMessage ??
|
||||
uploadData?.unavailableReason ??
|
||||
uploadData?.availabilityReason ??
|
||||
"发票上传未生成新票据";
|
||||
const reason = uploadData?.unavailableReason ?? uploadData?.availabilityReason ?? uploadData?.uploadAction ?? "-";
|
||||
const detailText = buildApiDetailText({
|
||||
status: uploadData?.status ?? "ocr_failed",
|
||||
reason,
|
||||
message,
|
||||
});
|
||||
return {
|
||||
status: "ocr_failed",
|
||||
detailText,
|
||||
bannerText: message,
|
||||
durationMs: Date.now() - startedAt,
|
||||
uploadMs,
|
||||
terminalWaitMs: 0,
|
||||
transport: "api",
|
||||
invoiceId: null,
|
||||
invoiceCode: null,
|
||||
invoiceNumber: null,
|
||||
sellerName: null,
|
||||
totalAmount: null,
|
||||
invoiceDate: null,
|
||||
};
|
||||
}
|
||||
|
||||
const pollStartedAt = Date.now();
|
||||
let latest: InvoiceStatusItem | null = null;
|
||||
while (Date.now() - pollStartedAt <= timeout) {
|
||||
const statusResponse = await api.get(`/api/invoice-pool/status?ids=${encodeURIComponent(uploadedInvoiceId)}`);
|
||||
const statusPayload = await readApiData<InvoiceStatusResponse>(statusResponse, "invoice_status");
|
||||
latest = statusPayload?.items?.[0] ?? null;
|
||||
if (latest && ["available", "ocr_failed", "bound", "archived"].includes(String(latest.status ?? ""))) {
|
||||
const reason = latest.availabilityReason ?? latest.reason ?? null;
|
||||
const message = latest.availabilityMessage ?? latest.message ?? null;
|
||||
const detailText = buildApiDetailText({
|
||||
status: latest.status,
|
||||
reason,
|
||||
message,
|
||||
invoiceCode: latest.invoiceCode,
|
||||
invoiceNumber: latest.invoiceNumber,
|
||||
invoiceDate: latest.invoiceDate,
|
||||
sellerName: latest.sellerName,
|
||||
totalAmount: latest.totalAmount,
|
||||
});
|
||||
return {
|
||||
status: mapApiStatusToResultStatus(latest.status),
|
||||
detailText,
|
||||
bannerText: message ?? "",
|
||||
durationMs: Date.now() - startedAt,
|
||||
uploadMs,
|
||||
terminalWaitMs: Date.now() - pollStartedAt,
|
||||
transport: "api",
|
||||
invoiceId: uploadedInvoiceId,
|
||||
invoiceCode: latest.invoiceCode ?? null,
|
||||
invoiceNumber: latest.invoiceNumber ?? null,
|
||||
sellerName: latest.sellerName ?? null,
|
||||
totalAmount: latest.totalAmount ?? null,
|
||||
invoiceDate: latest.invoiceDate ?? null,
|
||||
};
|
||||
}
|
||||
await new Promise((resolve) => {
|
||||
setTimeout(resolve, 1_000);
|
||||
});
|
||||
}
|
||||
|
||||
const detailText = buildApiDetailText({
|
||||
status: latest?.status ?? "pending_ocr",
|
||||
reason: latest?.availabilityReason ?? latest?.reason ?? "timeout_or_pending",
|
||||
message: latest?.availabilityMessage ?? latest?.message ?? "OCR 仍在处理中",
|
||||
invoiceCode: latest?.invoiceCode ?? null,
|
||||
invoiceNumber: latest?.invoiceNumber ?? null,
|
||||
invoiceDate: latest?.invoiceDate ?? null,
|
||||
sellerName: latest?.sellerName ?? null,
|
||||
totalAmount: latest?.totalAmount ?? null,
|
||||
});
|
||||
throw new Error(
|
||||
[
|
||||
"live 腾讯 OCR API 等待终态超时",
|
||||
`invoiceIdPresent=${Boolean(uploadedInvoiceId)}`,
|
||||
`timeoutMs=${timeout}`,
|
||||
`reasonCategory=${categorizeText(detailText)}`,
|
||||
`uploadMs=${uploadMs}`,
|
||||
].join("\n"),
|
||||
);
|
||||
}
|
||||
|
||||
export function resolveLiveOcrUser(fallback: E2EUser): E2EUser {
|
||||
const identifier =
|
||||
process.env.PLAYWRIGHT_OCR_LIVE_IDENTIFIER?.trim() ||
|
||||
@@ -143,6 +332,9 @@ export async function uploadLiveInvoiceAndWait(page: Page, invoiceFilePath: stri
|
||||
detailText,
|
||||
bannerText: message,
|
||||
durationMs: Date.now() - startedAt,
|
||||
uploadMs: Date.now() - startedAt,
|
||||
terminalWaitMs: 0,
|
||||
transport: "ui",
|
||||
invoiceId: null,
|
||||
invoiceCode: null,
|
||||
invoiceNumber: null,
|
||||
@@ -192,6 +384,7 @@ export async function uploadLiveInvoiceAndWait(page: Page, invoiceFilePath: stri
|
||||
detailText,
|
||||
bannerText,
|
||||
durationMs: Date.now() - startedAt,
|
||||
transport: "ui",
|
||||
invoiceId: uploadedInvoiceId ?? extractField(detailText, "票据ID:"),
|
||||
invoiceCode,
|
||||
invoiceNumber,
|
||||
|
||||
@@ -2,12 +2,13 @@ import { existsSync, mkdirSync, readdirSync, statSync, writeFileSync } from "nod
|
||||
import path from "node:path";
|
||||
|
||||
import { expect, test } from "../helpers/fixtures";
|
||||
import { EMPLOYEE, login } from "../helpers/auth";
|
||||
import { createAuthenticatedApiClient, EMPLOYEE, login, type AuthenticatedApiClient } from "../helpers/auth";
|
||||
import {
|
||||
createSafeLiveInvoiceOcrEvidence,
|
||||
openInvoicePool,
|
||||
resolveLiveOcrUser,
|
||||
uploadLiveInvoiceAndWait,
|
||||
uploadLiveInvoiceViaApiAndWait,
|
||||
type SafeLiveInvoiceOcrEvidence,
|
||||
} from "../helpers/liveInvoiceOcr";
|
||||
|
||||
@@ -16,12 +17,23 @@ type BatchSample = {
|
||||
alias: string;
|
||||
};
|
||||
|
||||
type BatchCoverage = {
|
||||
hasAvailable: boolean;
|
||||
hasDuplicateOrUsed: boolean;
|
||||
};
|
||||
|
||||
const rootDir = path.resolve(process.cwd(), "..");
|
||||
const reportDir = process.env.PLAYWRIGHT_OCR_BATCH_REPORT_DIR?.trim()
|
||||
|| path.join(rootDir, "artifacts/qa-flow", `live-ocr-batch-${new Date().toISOString().replace(/[-:]/g, "").replace(/\.\d{3}Z$/, "Z")}`);
|
||||
const markdownReportPath = path.join(reportDir, "summary.md");
|
||||
const jsonReportPath = path.join(reportDir, "summary.json");
|
||||
|
||||
function parsePositiveInteger(value: string | undefined, fallback: number | null = null) {
|
||||
if (!value?.trim()) return fallback;
|
||||
const parsed = Number.parseInt(value.trim(), 10);
|
||||
return Number.isFinite(parsed) && parsed > 0 ? parsed : fallback;
|
||||
}
|
||||
|
||||
function resolveInvoiceFiles() {
|
||||
const files = new Set<string>();
|
||||
const directFile = process.env.PLAYWRIGHT_OCR_LIVE_FILE?.trim();
|
||||
@@ -49,13 +61,26 @@ function resolveInvoiceFiles() {
|
||||
return [...files]
|
||||
.filter((filePath) => existsSync(filePath))
|
||||
.sort((left, right) => left.localeCompare(right, "zh-CN"))
|
||||
.slice(0, parsePositiveInteger(process.env.PLAYWRIGHT_OCR_BATCH_MAX_SAMPLES, Number.POSITIVE_INFINITY) ?? undefined)
|
||||
.map((filePath, index): BatchSample => ({
|
||||
filePath,
|
||||
alias: `SAMPLE-PDF-${String(index + 1).padStart(3, "0")}`,
|
||||
}));
|
||||
}
|
||||
|
||||
function writeReport(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
function getCoverage(results: SafeLiveInvoiceOcrEvidence[]): BatchCoverage {
|
||||
return {
|
||||
hasAvailable: results.some((item) => item.status === "available"),
|
||||
hasDuplicateOrUsed: results.some((item) => ["duplicate_or_same_invoice", "already_used"].includes(item.reasonCategory)),
|
||||
};
|
||||
}
|
||||
|
||||
function isCoverageSatisfied(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
const coverage = getCoverage(results);
|
||||
return coverage.hasAvailable && coverage.hasDuplicateOrUsed;
|
||||
}
|
||||
|
||||
function writeReport(results: SafeLiveInvoiceOcrEvidence[], options: { plannedSamples: number; stopReason: string }) {
|
||||
mkdirSync(reportDir, { recursive: true });
|
||||
const counts = {
|
||||
total: results.length,
|
||||
@@ -76,6 +101,9 @@ function writeReport(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
const summary = {
|
||||
mode: "live-ocr-batch-redacted",
|
||||
checkedAt: new Date().toISOString(),
|
||||
plannedSamples: options.plannedSamples,
|
||||
stopReason: options.stopReason,
|
||||
coverage: getCoverage(results),
|
||||
counts,
|
||||
fieldHitCounts,
|
||||
privacy: "redacted aliases and boolean field hits only; no invoice number, seller, amount, full filename, full path, raw OCR payload, invoice image, token, or password",
|
||||
@@ -89,6 +117,8 @@ function writeReport(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
"",
|
||||
"## 结果",
|
||||
"",
|
||||
`- 计划样本数:${options.plannedSamples}`,
|
||||
`- 停止原因:${options.stopReason}`,
|
||||
`- 总文件数:${counts.total}`,
|
||||
`- 可用:${counts.available}`,
|
||||
`- 识别失败:${counts.ocrFailed}`,
|
||||
@@ -99,10 +129,10 @@ function writeReport(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
"",
|
||||
"## 明细",
|
||||
"",
|
||||
"| 样本 | 状态 | 耗时(ms) | 原因分类 | 票据ID命中 | 发票号命中 | 销售方命中 | 金额命中 | 日期命中 |",
|
||||
"| --- | --- | ---: | --- | --- | --- | --- | --- | --- |",
|
||||
"| 样本 | 通道 | 状态 | 总耗时(ms) | 上传(ms) | 终态等待(ms) | 原因分类 | 票据ID命中 | 发票号命中 | 销售方命中 | 金额命中 | 日期命中 |",
|
||||
"| --- | --- | --- | ---: | ---: | ---: | --- | --- | --- | --- | --- | --- |",
|
||||
...results.map((item) => (
|
||||
`| ${item.alias} | ${item.status} | ${item.durationMs} | ${item.reasonCategory} | ${item.fieldHits.invoiceId ? "Y" : "N"} | ${item.fieldHits.invoiceCodeOrNumber ? "Y" : "N"} | ${item.fieldHits.sellerName ? "Y" : "N"} | ${item.fieldHits.totalAmount ? "Y" : "N"} | ${item.fieldHits.invoiceDate ? "Y" : "N"} |`
|
||||
`| ${item.alias} | ${item.transport ?? "-"} | ${item.status} | ${item.durationMs} | ${item.timings?.uploadMs ?? "-"} | ${item.timings?.terminalWaitMs ?? "-"} | ${item.reasonCategory} | ${item.fieldHits.invoiceId ? "Y" : "N"} | ${item.fieldHits.invoiceCodeOrNumber ? "Y" : "N"} | ${item.fieldHits.sellerName ? "Y" : "N"} | ${item.fieldHits.totalAmount ? "Y" : "N"} | ${item.fieldHits.invoiceDate ? "Y" : "N"} |`
|
||||
)),
|
||||
"",
|
||||
"## 说明",
|
||||
@@ -111,6 +141,8 @@ function writeReport(results: SafeLiveInvoiceOcrEvidence[]) {
|
||||
"- 本报告只记录样本别名、状态、耗时、字段命中布尔值和原因分类。",
|
||||
"- 本报告不得记录发票号码、销售方、金额、完整文件名、完整路径、原始 OCR payload 或票面原图。",
|
||||
"- 如某张票失败,会继续处理后续文件,并在本报告统一汇总。",
|
||||
"- 如开启覆盖即停止,达到可用终态和重复/已用终态覆盖后会停止继续上传。",
|
||||
"- 批量模式默认使用 API 状态轮询,避免 UI 详情面板刷新造成 OCR 耗时误判;如需 UI 复核可设置 PLAYWRIGHT_OCR_BATCH_TRANSPORT=ui。",
|
||||
];
|
||||
|
||||
writeFileSync(markdownReportPath, `${lines.join("\n")}\n`, "utf-8");
|
||||
@@ -125,20 +157,44 @@ test("手工 live 腾讯 OCR 批量辅助复核:真实票样逐张上传并输
|
||||
test.skip(!invoiceFiles.length, "需要提供 PDF 格式的 PLAYWRIGHT_OCR_LIVE_FILE / PLAYWRIGHT_OCR_LIVE_FILES / PLAYWRIGHT_OCR_LIVE_DIR");
|
||||
|
||||
const results: SafeLiveInvoiceOcrEvidence[] = [];
|
||||
let stopReason = "completed_all_samples";
|
||||
const stopWhenCovered = process.env.PLAYWRIGHT_OCR_BATCH_STOP_WHEN_COVERED === "1";
|
||||
const transport = process.env.PLAYWRIGHT_OCR_BATCH_TRANSPORT?.trim() || "api";
|
||||
const liveOcrUser = resolveLiveOcrUser(EMPLOYEE);
|
||||
await login(page, liveOcrUser);
|
||||
let apiClient: AuthenticatedApiClient | null = null;
|
||||
|
||||
if (transport === "ui") {
|
||||
await login(page, liveOcrUser);
|
||||
} else {
|
||||
apiClient = await createAuthenticatedApiClient(liveOcrUser);
|
||||
}
|
||||
|
||||
try {
|
||||
await openInvoicePool(page);
|
||||
if (transport === "ui") {
|
||||
await openInvoicePool(page);
|
||||
}
|
||||
|
||||
for (const sample of invoiceFiles) {
|
||||
await openInvoicePool(page);
|
||||
const result = await uploadLiveInvoiceAndWait(page, sample.filePath);
|
||||
let result;
|
||||
if (transport === "ui") {
|
||||
await openInvoicePool(page);
|
||||
result = await uploadLiveInvoiceAndWait(page, sample.filePath);
|
||||
} else {
|
||||
if (!apiClient) throw new Error("api_client_not_initialized");
|
||||
result = await uploadLiveInvoiceViaApiAndWait(apiClient.api, sample.filePath, sample.alias);
|
||||
}
|
||||
results.push(createSafeLiveInvoiceOcrEvidence(result, sample.alias));
|
||||
if (stopWhenCovered && isCoverageSatisfied(results)) {
|
||||
stopReason = "coverage_satisfied";
|
||||
break;
|
||||
}
|
||||
}
|
||||
} finally {
|
||||
writeReport(results);
|
||||
await page.context().clearCookies().catch(() => undefined);
|
||||
writeReport(results, { plannedSamples: invoiceFiles.length, stopReason });
|
||||
await apiClient?.dispose().catch(() => undefined);
|
||||
if (transport === "ui") {
|
||||
await page.context().clearCookies().catch(() => undefined);
|
||||
}
|
||||
}
|
||||
|
||||
const allowDuplicate = process.env.PLAYWRIGHT_OCR_ALLOW_DUPLICATE === "1";
|
||||
|
||||
@@ -3,7 +3,7 @@
|
||||
"OINV1-1","真实发票样本脱敏盘点","盘点 artifacts 和私有 env 中配置的真实 PDF 样本,只记录数量、类型、大小区间和可用性,不记录完整文件名、销售方、发票号码、金额、完整本地路径或票面原图。","生成或记录脱敏样本清单;至少确认 PDF 样本数量、文件存在性、文件类型和是否可供 OCR;报告隐私扫描通过。","受影响流程:样本准备 -> 自动化输入 -> 隐私边界;验证步骤:rg --files | rg -i 'pdf$|发票|invoice';node scripts/qa/check-ocr-provider-samples.mjs --json 使用别名或 env 样本,不输出原始票面值。","rg|node","DONE","DONE","DONE","artifacts;scripts/qa/check-ocr-provider-samples.mjs;artifacts/qa-flow/oinv1-provider-baseline-20260523-202701/summary.json","[""OINV1-0""]","已用 /tmp SAMPLE-PDF-xxx 别名盘点根部 PDF 样本 2 个;summary 只含 alias、mime、size、transport/status,不含完整文件名、本地路径或票面值。"
|
||||
"OINV1-2","OCR provider 样本只读验证","使用现有 check-ocr-provider-samples.mjs 直接调用 OCR provider 或在缺凭据时给出 SKIP,验证 PDF native transport、必要字段命中情况和图片 OCR 是否继续冻结。","脚本输出 status/decision/counts/fieldHits,不输出发票号码、销售方、金额或原始 OCR payload;缺腾讯凭据时记录 SKIP,不误判业务失败;如 provider 返回 HOLD,进入修复分析。","受影响流程:真实 PDF -> OCR provider -> 归一化字段 -> 决策;验证步骤:PLAYWRIGHT_OCR_LIVE_FILES=<样本列表> node scripts/qa/check-ocr-provider-samples.mjs --json。","node","DONE","DONE","DONE","scripts/qa/check-ocr-provider-samples.mjs;server/src/services/ocr/tencent.js;artifacts/qa-flow/oinv1-provider-baseline-20260523-202701/summary.json","[""OINV1-1""]","provider 基线 status=SKIP,decision=FREEZE_BY_DEFAULT,counts total=2/pdf=2/skip=2,原因 missing_tencent_credentials;这是本地缺凭据,不是产品失败。不启用 OCR_IMAGE_OCR_EVALUATION;图片 OCR 继续冻结。回滚/清理:provider 只读验证不产生 OA 数据;只需删除本地脱敏报告或环境变量即可。"
|
||||
"OINV1-3","公网上传型 OCR smoke 脱敏加固方案","审查 live-invoice-ocr 和 batch spec 的输出,避免真实发票成功识别时把发票号、销售方、金额、完整文件名或本地路径写入日志/附件;准备 QA-only 脱敏报告格式。","形成实现方案并经用户确认后再改代码;脱敏报告只保留 alias、状态、耗时、字段命中布尔值、重复/已使用等可解释原因;不输出原始票面值。","受影响流程:公网 OA 上传 -> OCR 等待终态 -> 报告留存;验证步骤:代码审查 client/tests/e2e/manual/live-invoice-ocr*.spec.ts 和 helpers/liveInvoiceOcr.ts;改造后跑 npm/Playwright。","manual|playwright","DONE","DONE","DONE","client/tests/e2e/manual/live-invoice-ocr.spec.ts;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts;client/tests/e2e/helpers/liveInvoiceOcr.ts;artifacts/qa-flow/live-smoke-readiness-20260523-205755/summary.json;artifacts/qa-flow/oinv1-redacted-batch-smoke-20260523-205755/summary.json","[""OINV1-2""]","完成:用户确认 Q1-Q6 全部 A 后实施脱敏加固。单票 live OCR 输出 alias/status/duration/reasonCategory/fieldHits;批量报告写 artifacts/qa-flow/*/summary.md 和 summary.json,仅含 alias、状态、耗时、字段命中布尔值和原因分类;最小公网单票与批量 smoke 均通过。回滚:恢复 3 个 Playwright 测试文件改动;不影响生产 API/schema。"
|
||||
"OINV1-4","公网真实发票 OCR 小样本 smoke","在脱敏加固后,用 OA-only/live smoke 员工账号上传 2-3 张代表 PDF 样票,验证重复票、可用票、识别失败票的用户可理解终态;不提交报销,不创建付款。105 张全量仅保留为后续可选压力/覆盖专项。","生成脱敏 summary;每张样票有 alias、终态、耗时和字段命中;重复/已用于报销可解释时为 PASS_WITH_FINDINGS 或 PASS_DUPLICATE_EXPLAINED;不可解释 OCR 失败记录 finding。","受影响流程:员工登录 -> 票据上传 -> OCR 终态 -> 票据池查看;验证步骤:PLAYWRIGHT_MANUAL=1 PLAYWRIGHT_OCR_LIVE=1 PLAYWRIGHT_OCR_ALLOW_DUPLICATE=1 npx playwright test tests/e2e/manual/live-invoice-ocr-batch.spec.ts --workers=1 --reporter=list;隐私扫描。","playwright","TODO","TODO","TODO","client/tests/e2e/manual/live-invoice-ocr.spec.ts;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts;artifacts/qa-flow","[""OINV1-3""]","写入边界:只允许票据上传 smoke 数据;不提交报销、不审批、不付款、不导出含敏感内容资料;报告不得含完整账号、发票号、销售方、金额、路径或票面原图。2026-05-23 用户纠偏:够用即可,启用前只跑代表小样本;代表样本能覆盖可用终态和重复/已用终态即可。已产生脱敏部分证据:sample2 报告覆盖可用终态和重复/相同发票终态,隐私扫描未命中真实票面值;但多样本 Playwright 因公网 OCR/UI 等待超过测试时限,OINV1-4 暂不置 DONE,后续应增加 max-samples/覆盖即停止后再关闭。"
|
||||
"OINV1-4","公网真实发票 OCR 小样本 smoke","在脱敏加固后,用 OA-only/live smoke 员工账号上传 2-3 张代表 PDF 样票,验证重复票、可用票、识别失败票的用户可理解终态;不提交报销,不创建付款。105 张全量仅保留为后续可选压力/覆盖专项。","生成脱敏 summary;每张样票有 alias、终态、耗时和字段命中;重复/已用于报销可解释时为 PASS_WITH_FINDINGS 或 PASS_DUPLICATE_EXPLAINED;不可解释 OCR 失败记录 finding。","受影响流程:员工登录 -> 票据上传 -> OCR 终态 -> 票据池查看;验证步骤:PLAYWRIGHT_MANUAL=1 PLAYWRIGHT_OCR_LIVE=1 PLAYWRIGHT_OCR_ALLOW_DUPLICATE=1 PLAYWRIGHT_OCR_BATCH_MAX_SAMPLES=3 PLAYWRIGHT_OCR_BATCH_STOP_WHEN_COVERED=1 npx playwright test tests/e2e/manual/live-invoice-ocr-batch.spec.ts --workers=1 --reporter=list;隐私扫描。","playwright","DONE","DONE","DONE","client/tests/e2e/manual/live-invoice-ocr.spec.ts;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts;client/tests/e2e/helpers/liveInvoiceOcr.ts;artifacts/qa-flow/oinv1-redacted-batch-api-coverage-stop-20260523-continue/summary.json","[""OINV1-3""]","完成:按用户“够用即可”纠偏,批量脚本支持 max samples 和覆盖即停止,默认 API 状态轮询避免 UI 详情刷新误判 OCR 耗时。公网小样本 planned=3、actual=2、stopReason=coverage_satisfied,覆盖 available 和 duplicate_or_same_invoice,隐私扫描未命中真实票面值;未提交报销、未审批、未付款。远端脱敏查询显示最近真实 OCR completeLagSec=3,几分钟假象来自 UI 批量等待方式。回滚:恢复 helper/spec 改动;公网 smoke 票据按审计保留。"
|
||||
"OINV1-5","报销闭环测试数据准备","在 OCR smoke 可解释通过后,评估是否用 TEST/SMOKE 标记创建一条报销测试闭环,覆盖绑定票据、提交、审批、付款批次、回单和 AP 导出;真实业务闭环仍需人工最终验收。","若执行自动闭环,必须带 TEST/SMOKE marker,报告脱敏;若不执行,输出人工步骤和数据准备要求;不得执行真实银行付款。","受影响流程:票据池 -> 报销 -> 审批 -> 付款/AP;验证步骤:复用 reimbursement-payment-batch-flow 或 OAOPS1 真实业务门禁,必要时新建专项合同。","playwright|manual","TODO","TODO","TODO","client/tests/e2e/reimbursement-payment-batch-flow.spec.ts;scripts/qa/run-oa-real-use-daily.sh","[""OINV1-4""]","付款仍由线下银行完成;自动化只能创建/验证 OA 测试数据,不能替代真实业务发票合规判断。"
|
||||
"OINV1-6","本轮门禁和回填","完成合同校验、现有脚本审查、provider 样本基线和隐私检查;未获代码确认前,不改 live OCR spec;未获公网写入确认前,不运行上传型批量测试。","validate_issues_csv、process contract、git diff --check 通过;最终回复明确哪些自动完成、哪些需要用户批准或人工验收。","受影响流程:合同 -> 自动基线 -> 后续实施判断;验证步骤:validate_issues_csv.py <OINV1 csv>;node scripts/qa/check-process-contract.mjs --issue <OINV1 csv> --profile general --json;git diff --check。","python3|node|git","TODO","TODO","TODO","issues/2026-05-23_20-24-42-oa-invoice-automation-hardening-v1.csv","[""OINV1-0"",""OINV1-1"",""OINV1-2""]","本轮可先做只读/本地 provider 基线;公网上传和脚本改造按用户确认继续。"
|
||||
"OINV1-7","现有单票公网 live OCR 基线","在不跑批量真实票的前提下,加载本机私有 live-smoke.env,使用现有单票 live OCR 测试复核公网发票上传/OCR 可解释终态;不提交报销、不审批、不付款。","公网 /ready、live smoke 和单票 OCR 测试通过;若样票重复或已用于报销,需给出可解释终态;报告不得输出完整发票号、销售方、金额、完整本地路径或票面原图。","受影响流程:员工登录 -> 单票上传 -> OCR/重复校验 -> 票据池终态;验证步骤:set -a; source /home/shunli/.hyb-oa/live-smoke.env; set +a; LIVE_SMOKE_BASE_URL=https://oa.hyinda.com bash scripts/qa/run-live-smoke-readiness.sh。","bash|playwright","DONE","DONE","DONE","artifacts/qa-flow/live-smoke-readiness-20260523-202814/summary.json;client/tests/e2e/manual/live-invoice-ocr.spec.ts","[""OINV1-0""]","公网 live smoke 3 passed;真实 OCR 单票 1 passed;样票结果为已用于报销/不能重复添加的可解释终态,输出字段为空值,未打印票面值。回滚/清理:本轮未提交报销或付款;如需清理仅按 OA 票据审计规则处理测试上传记录。"
|
||||
|
||||
|
@@ -2,7 +2,7 @@
|
||||
"OANEXT2-0","建立后续工作细化采纳合同","基于 OINV1 外部发票样本池、OALAUNCH1 最终启用门禁、OANEXT1 采纳路线和 OAOPS1 自动巡检最新状态,建立下一轮后续工作细化合同;本合同只供用户审查采纳,不直接实施代码、部署或公网写入。","Issue CSV 校验通过;明确本合同不替代 OINV1、OALAUNCH1、OAOPS1 或 OANEXT1,只提供后续执行顺序、门禁和人工决策点;除已完成的审视/合同门禁外,执行项默认 TODO。","受影响流程:最新状态审视 -> 后续路线拆分 -> 用户审查采纳;验证步骤:python3 /home/shunli/.codex/skills/plan/scripts/validate_issues_csv.py issues/2026-05-23_20-47-20-oa-next-adoption-detailed-roadmap-v2.csv。","python3|manual","DONE","DONE","DONE","issues/2026-05-23_20-47-20-oa-next-adoption-detailed-roadmap-v2.csv","[]","MAGI 审视:自动巡检和资产/发票样本准备较完整,但最终启用仍缺真实业务、权限、备份、7 天和 Go/No-Go;执行:新建采纳合同;提升:先自动补证据,再进入人工和运维门禁。回滚/清理:删除本规划 CSV 或后续提交即可,不影响公网、数据库或业务数据。"
|
||||
"OANEXT2-1","最新状态基线审视","梳理当前未完成项:OINV1-3/4/5/6、OALAUNCH1-3..7、OANEXT1-2..9、OAOPS1-3..6;识别哪些可自动化、哪些需人工/运维批准。","列出当前最高优先级:1) OINV1 批量 OCR 脱敏与验证;2) 权限矩阵自动负例;3) TEST/SMOKE 业务闭环;4) 真实业务人工闭环;5) 备份恢复;6) 7 天稳定化;7) Go/No-Go。","受影响流程:Issue 状态读取 -> 后续排序 -> 采纳决策;验证步骤:读取 OINV1/OALAUNCH1/OANEXT1/OAOPS1 CSV 中 TODO 行并核对最新提交。","python3|git","DONE","DONE","DONE","issues/2026-05-23_20-24-42-oa-invoice-automation-hardening-v1.csv;issues/2026-05-23_16-49-39-oa-final-launch-readiness-v1.csv;issues/2026-05-23_19-14-44-oa-next-work-adoption-plan-v1.csv;issues/2026-05-22_20-09-22-oa-real-use-stabilization-v1.csv","[""OANEXT2-0""]","当前证据:OAOPS1 最新 PASS_OA_REAL_USE_DAILY_READY;OINV1 找到报销候选 114 个、PDF 105 个;OALAUNCH1 仍 PAUSED_MANUAL_CONFIRMATION。"
|
||||
"OANEXT2-2","第一优先级:批量 OCR 报告脱敏加固","若用户采纳,优先实施 OINV1-3:改造或新增 QA-only 批量 OCR 报告逻辑,避免真实发票号、销售方、金额、完整文件名、完整路径和票面原图进入日志、附件或 summary。","批量 OCR 报告只输出 alias、状态、耗时、字段命中布尔值、重复/已使用/失败可解释原因;隐私扫描通过;不改变生产业务 API/schema。","受影响流程:真实 PDF 样本 -> 公网上传 OCR -> 脱敏报告;验证步骤:node/ts 语法检查;Playwright 单票 smoke;报告隐私扫描;validate OINV1。","playwright|node|git","DONE","DONE","DONE","client/tests/e2e/manual/live-invoice-ocr.spec.ts;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts;client/tests/e2e/helpers/liveInvoiceOcr.ts;artifacts/qa-flow/live-smoke-readiness-20260523-205755;artifacts/qa-flow/oinv1-redacted-batch-smoke-20260523-205755","[""OANEXT2-1""]","用户选择 A;已完成脱敏加固和最小公网验证。回滚:恢复测试脚本改动;不影响生产 API/schema 或既有业务数据。"
|
||||
"OANEXT2-3","第二优先级:代表小样本 PDF OCR smoke","在 OANEXT2-2 脱敏加固后,从 /tmp SAMPLE-PDF-xxx 别名池选取 2-3 张代表 PDF 报销候选样本执行公网 OCR smoke;默认不提交报销、不审批、不付款。105 张全量仅作为后续可选压力/覆盖专项,不作为启用前必跑门禁。","输出小样本脱敏 summary:总数、成功数、重复/已用数、失败/超时数、字段命中率和可解释失败分类;报告不含票面值、完整路径或原始 payload;发现不可解释失败进入修复合同。","受影响流程:PDF 样本池 -> 公网票据上传 -> OCR/重复终态 -> 脱敏统计;验证步骤:PLAYWRIGHT_MANUAL=1 PLAYWRIGHT_OCR_LIVE=1 PLAYWRIGHT_OCR_ALLOW_DUPLICATE=1 对代表小样本执行批量 OCR Playwright;隐私扫描。","playwright|bash","TODO","TODO","TODO","artifacts/qa-flow;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts","[""OANEXT2-2""]","公网写入边界:只允许 smoke 员工票据上传测试记录;不提交报销、不审批、不付款。2026-05-23 用户纠偏:够用即可,启动前不追求 105 张全量批跑;代表样本能覆盖可用终态和重复/已用终态即可。"
|
||||
"OANEXT2-3","第二优先级:代表小样本 PDF OCR smoke","在 OANEXT2-2 脱敏加固后,从 /tmp SAMPLE-PDF-xxx 别名池选取 2-3 张代表 PDF 报销候选样本执行公网 OCR smoke;默认不提交报销、不审批、不付款。105 张全量仅作为后续可选压力/覆盖专项,不作为启用前必跑门禁。","输出小样本脱敏 summary:总数、成功数、重复/已用数、失败/超时数、字段命中率和可解释失败分类;报告不含票面值、完整路径或原始 payload;发现不可解释失败进入修复合同。","受影响流程:PDF 样本池 -> 公网票据上传 -> OCR/重复终态 -> 脱敏统计;验证步骤:PLAYWRIGHT_MANUAL=1 PLAYWRIGHT_OCR_LIVE=1 PLAYWRIGHT_OCR_ALLOW_DUPLICATE=1 PLAYWRIGHT_OCR_BATCH_MAX_SAMPLES=3 PLAYWRIGHT_OCR_BATCH_STOP_WHEN_COVERED=1 执行批量 OCR Playwright;隐私扫描。","playwright|bash","DONE","DONE","DONE","artifacts/qa-flow/oinv1-redacted-batch-api-coverage-stop-20260523-continue/summary.json;client/tests/e2e/manual/live-invoice-ocr-batch.spec.ts;client/tests/e2e/helpers/liveInvoiceOcr.ts","[""OANEXT2-2""]","完成:公网小样本 planned=3、actual=2、stopReason=coverage_satisfied,覆盖可用终态和重复/相同发票终态;默认 API 轮询,UI 复核可通过 PLAYWRIGHT_OCR_BATCH_TRANSPORT=ui 单独执行。未提交报销、未审批、未付款。105 张全量继续作为后续可选压力/覆盖专项。"
|
||||
"OANEXT2-4","第三优先级:权限矩阵自动负例","补齐 OALAUNCH1-5/OANEXT1-3:用 OA-only smoke 账号自动验证员工、主管、财务、HR、CEO、管理员的入口可用性和普通员工越权负例。","普通员工不能访问财务/AP、合同敏感附件、他人审批、管理入口、审计导出、高敏或冻结模块;角色必要入口可用;报告脱敏且可作为 OALAUNCH_PERMISSION_EVIDENCE。","受影响流程:多角色登录 -> 页面入口 -> API 直连负例 -> 权限报告;验证步骤:复用 check-oa-only-smoke-readiness、permission-direct-access 或新增 QA-only 权限矩阵脚本。","playwright|node","TODO","TODO","TODO","scripts/qa/check-oa-only-smoke-readiness.mjs;client/tests/e2e/permission-direct-access.spec.ts;docs/trialrun/权限直连审计记录.md","[""OANEXT2-1""]","涉及权限模型修复时必须启用只读 reviewer subagent;本项优先只读验证,不直接改权限代码。"
|
||||
"OANEXT2-5","第四优先级:TEST/SMOKE 业务闭环自动化","在不使用真实付款、不导入银行流水、不写 ERP 的前提下,准备用 TEST/SMOKE 标记自动跑 OA 主办域闭环:报销/OCR、审批、付款批次、银行回单附件、AP 导出、采购/费用、项目/合同关联。","自动化测试数据有 TEST/SMOKE marker;不含真实合同正文、敏感报价、银行流水原件或票面原图;可验证流程技术闭环,但不替代真实业务人工验收。","受影响流程:报销 -> 审批 -> 付款批次 -> 回单 -> AP;采购/费用 -> 项目/合同;验证步骤:复用 reimbursement-payment-batch-flow、run-oa-real-use-daily 或新增 QA-only flow;报告脱敏。","playwright|bash","TODO","TODO","TODO","client/tests/e2e/reimbursement-payment-batch-flow.spec.ts;scripts/qa/run-oa-real-use-daily.sh;artifacts/qa-flow","[""OANEXT2-2"",""OANEXT2-4""]","执行前需要用户确认是否允许公网创建 TEST/SMOKE 业务测试数据。实际付款仍由线下银行渠道完成,自动化不得触发真实支付。"
|
||||
"OANEXT2-6","第五优先级:真实业务人工闭环证据包","补齐 OAOPS1-3/4/5/6 和 OALAUNCH1-3:用 1 条真实报销、1 条采购/费用、1 条合同台账和每日问题登记形成正式启用前人工证据。","真实报销完成 OCR、提交、审批、付款批次、回单、已付款和 AP 导出;采购/费用完成审批并关联项目或合同;合同台账有编号、负责人、归档位置和权限检查;问题登记机制可执行。","受影响流程:真实业务 -> OA 记录 -> 财务/审计 -> 最终启用汇总;验证步骤:人工在公网 OA 操作;运行 run-oa-final-launch-readiness.sh 并提供 OALAUNCH_BUSINESS_EVIDENCE。","manual|bash","TODO","TODO","TODO","issues/2026-05-22_20-09-22-oa-real-use-stabilization-v1.csv;issues/2026-05-23_16-49-39-oa-final-launch-readiness-v1.csv","[""OANEXT2-5""]","真实票据、合同正文、敏感报价、银行流水原件和未脱敏票面不得进入仓库/聊天/自动化报告。"
|
||||
|
||||
|
在新工单中引用
屏蔽一个用户