Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
53 changes: 53 additions & 0 deletions DELIVERY.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,53 @@
# 交付声明(2026-09-07,D-017)

> **按执行方案 §8 时间盒规则:deadline 提前至今日 24:00,交付目标诚实降级为技术切片。本文档是唯一权威的完成状态清单——请以此为准,勿以演示效果推断未完成的能力。**

## 已完成并验证(有自动化测试与实跑证据)

| 能力 | 验证方式 |
|---|---|
| 项目创建 / 证据上传(presign → 服务端 magic bytes + sha256 校验 → READY/QUARANTINED) | Go 集成测试 + 真实 MinIO 实跑 |
| 组织隔离与审计日志(跨租户读写全部 fail closed) | 真实 PG 集成测试 |
| 可恢复审计任务(SKIP LOCKED 领取 / 租约 / 崩溃恢复 / 取消晚到丢弃 / 原子落库) | 真实 PG 集成测试 8 项 |
| StepFun 真实模型评估(D-016)+ 证据内联 base64 传输 | 真实调用冒烟 + 全流程 E2E(38–44s) |
| 前端工作流:登录 → 创建项目 → 上传 → 审计 → Evidence Viewer(原图 + bbox 定位 + 观察/推断 + 待人审标识) | Playwright E2E 2/2 |
| **人审四操作**(confirm / edit / reject / NA 需理由)与不可覆盖语义 | API + 集成路径 |
| **确定性评分与报告冻结**(ADR-0002 / D-018,手算样例单测钉死:62.50 / 83.33-partial / 83.33-NA) | 单元测试 + finalize 实跑 |
| **整改任务状态机**(OPEN→…→RESOLVED / ACCEPTED_RISK 需理由 / 非法迁移拒绝) | API 实跑 |
| **复测 + Before/After Diff**(按 rule key 对齐;缺席≠PASS) | API 实跑 + E2E |
| 离线 Eval runner(5 类 fixture:合法/缺证据/伪造引用/注入/供应商失败) | CI contracts 作业 |
| 离线 CI(Go/AI/Web/contracts 四作业,无模型 key 可跑) | GitHub Actions 全绿 |

**真实模型全流程演示结论(2026-09-07 实录)**:上传自制英文菜单 fixture 后,10 条规则评估落库——有证据项判定准确(语言可达 PASS、价格透明 PASS、支付标识 PASS),**无证据项全部诚实 UNKNOWN**(门头/扫码页/人工支付确认),未编造任何结论(No Fake Certainty 落地)。

## 未完成(显式声明,不冒充完成)

| PRD P0 项 | 状态 | 原因 |
|---|---|---|
| P0-2 URL 自动抓取 | **未实现**(D-021:已批准,2–3 人日工作量超出时间盒) | SSRF 隔离执行器未交付;URL 来源登记已具备 |
| P0-2 PDF / 文本解析 | 未实现(B13 未进入) | 图片链路优先 |
| 20 例 Golden Set + 评测门槛 | 未完成 | 真实素材与访谈未执行(T-005/T-006) |
| 完整比赛 MVP 声明 | **不成立** | 以上任一未达即不满足 PRD §26 完整验收 |

P1 未实现:Guest Page、QR、报告导出、多语言 Profile、团队协作。

## 已知限制(如实登记)

- 真实模型全流程依赖图片内联 base64(≤10MB);云端部署时建议切回预签名 URL + 公网存储(契约已支持两种,schema 注释说明)
- 本机 Docker Desktop 引擎故障,compose 探活验证由 ephemeral PG/MinIO 进程等价替代;T-003 尾项待 Docker 恢复后补验
- E2E 的 CI job 未接线(本地全栈已验证;后续补 CI 编排)
- OIDC 为开发测试身份模式(真实 IdP 配置待接入;生产前必须切换,见 AGENTS.md §7)
- E2E 里 fake 与 real 两次说明:fake 路径 CI 可复现(离线绿),real 路径依赖真实 key 与网络(已单独标注,两次实跑通过)

## 启动指南(评委 / 新成员)

```powershell
# 前提:Windows + Go 1.27 + Node 24/pnpm + Python 3.13/uv + PostgreSQL + MinIO
# 1. 配置模型 key(已由 PO 配置)
# services/ai/.env → MODEL_API_KEY / MODEL_BASE_URL / MODEL_ID=step-1o-turbo-vision
# 2. 启动全栈
powershell -File scripts\dev-e2e.ps1
# 3. 打开 http://localhost:3000 → 开发者登录 → 创建项目 → 上传菜单图 → 启动审计
```

回滚:`git revert` 对应批次 PR 的 merge commit;数据库迁移向前修复为默认策略(执行方案 §7.3)。
50 changes: 50 additions & 0 deletions apps/web/app/audits/[id]/page.tsx
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,14 @@ import { useEffect, useState } from "react";

import { ApiError, api, FindingSummary, getToken } from "@/lib/api";

type ScoreReport = {
total_score: number | null;
partial: boolean;
coverage_pct: number;
dimensions: { dimension: string; score: number; applicable_weight?: number }[];
blocking: { rule_id: string; severity: string }[];
};

const TERMINAL = new Set(["COMPLETED", "FAILED", "CANCELLED"]);
const STATUS_LABEL: Record<string, string> = {
QUEUED: "排队中",
Expand All @@ -35,6 +43,7 @@ export default function AuditPage() {
const [run, setRun] = useState<AuditPageRun | null>(null);
const [findings, setFindings] = useState<FindingSummary[]>([]);
const [error, setError] = useState<string | null>(null);
const [report, setReport] = useState<ScoreReport | null>(null);

useEffect(() => {
if (!getToken()) {
Expand Down Expand Up @@ -95,6 +104,47 @@ export default function AuditPage() {
{!TERMINAL.has(run.status) && (
<p className="mb-4 rounded bg-blue-50 p-3 text-blue-700">AI 分析进行中,页面自动刷新…</p>
)}
{run.status === "REVIEW_REQUIRED" && (
<section className="mb-6 rounded border p-4">
<h2 className="mb-2 font-medium">报告</h2>
<button
className="rounded bg-black px-4 py-2 font-medium text-white"
onClick={async () => {
try {
const r = await api.finalizeAudit(run.id);
setReport(r);
} catch (e) {
setError(e instanceof ApiError ? e.message : "finalize 失败");
}
}}
>
生成报告(冻结评分)
</button>
{report && (
<div className="mt-3 text-sm">
<p className="font-semibold">
总分:
{report.total_score === null
? "部分评估(未出总分)"
: `${report.total_score} / 100`}
<span className="ml-2 text-gray-500">覆盖率 {report.coverage_pct}%</span>
</p>
<ul className="mt-2 space-y-1">
{report.dimensions.map((d) => (
<li key={d.dimension}>
{d.dimension}: {d.score}
</li>
))}
</ul>
{report.blocking.length > 0 && (
<p className="mt-2 text-red-700">
阻断/关键问题:{report.blocking.map((b) => b.rule_id).join("、")}
</p>
)}
</div>
)}
</section>
)}

{TERMINAL.has(run.status) && findings.length === 0 && (
<p className="text-gray-500">本次审计没有产出候选结论。</p>
Expand Down
99 changes: 96 additions & 3 deletions apps/web/app/findings/[id]/page.tsx
Original file line number Diff line number Diff line change
Expand Up @@ -10,9 +10,9 @@

import Link from "next/link";
import { useParams } from "next/navigation";
import { useEffect, useState } from "react";
import { useCallback, useEffect, useState } from "react";

import { ApiError, api, getToken } from "@/lib/api";
import { ApiError, api, FindingSummary, getToken } from "@/lib/api";

type Finding = {
id: string;
Expand All @@ -34,9 +34,57 @@ type EvidenceImage = { id: string; url: string | null; status: string };

export default function FindingPage() {
const params = useParams<{ id: string }>();
const [finding, setFinding] = useState<Finding | null>(null);
const [finding, setFinding] = useState<FindingSummary | null>(null);
const [images, setImages] = useState<EvidenceImage[]>([]);
const [error, setError] = useState<string | null>(null);
const [reviewError, setReviewError] = useState<string | null>(null);
const [busy, setBusy] = useState(false);

const submitReview = useCallback(
async (decision: string, edits?: { assessment_status?: string }, naReason?: string) => {
if (!finding) return;
setBusy(true);
setReviewError(null);
try {
const updated = (await api.submitReview(finding.id, {
decision,
edits,
na_reason: naReason,
})) as FindingSummary;
setFinding(updated);
} catch (e) {
setReviewError(e instanceof ApiError ? e.message : "提交失败");
} finally {
setBusy(false);
}
},
[finding],
);

const askNa = useCallback(async () => {
if (!finding) return;
const reason = window.prompt("NA 必须留理由(为什么此项不适用):");
if (reason === null) return;
if (!reason.trim()) {
setReviewError("NA 必须填写理由");
return;
}
await submitReview("na", undefined, reason);
}, [finding, submitReview]);

const askEdit = useCallback(async () => {
if (!finding) return;
const status = window.prompt(
"修订结论(PASS / WARN / FAIL / UNKNOWN):",
finding.assessment_status,
);
if (status === null) return;
if (!["PASS", "WARN", "FAIL", "UNKNOWN"].includes(status)) {
setReviewError("结论必须是 PASS/WARN/FAIL/UNKNOWN");
return;
}
await submitReview("edit", { assessment_status: status });
}, [finding, submitReview]);

useEffect(() => {
if (!getToken()) {
Expand Down Expand Up @@ -100,6 +148,51 @@ export default function FindingPage() {
</span>
</div>

{finding.review_status !== "UNREVIEWED" ? (
<p className="mb-4 rounded border border-green-200 bg-green-50 p-3 text-sm text-green-800">
已处置:{finding.review_status}(处置记录不可覆盖;更正产生新记录)
</p>
) : (
<section className="mb-4 rounded border p-3 text-sm">
<p className="mb-2 font-medium">人审处置(AI Suggests, Human Owns)</p>
<div className="flex flex-wrap gap-2">
<button
className="rounded border px-3 py-1.5 hover:bg-green-50 disabled:opacity-50"
disabled={busy}
onClick={() => submitReview("confirm")}
>
确认
</button>
<button
className="rounded border px-3 py-1.5 hover:bg-amber-50 disabled:opacity-50"
disabled={busy}
onClick={askEdit}
>
修订结论
</button>
<button
className="rounded border px-3 py-1.5 hover:bg-red-50 disabled:opacity-50"
disabled={busy}
onClick={() => submitReview("reject")}
>
驳回
</button>
<button
className="rounded border px-3 py-1.5 hover:bg-gray-50 disabled:opacity-50"
disabled={busy}
onClick={askNa}
>
标记 NA
</button>
</div>
{reviewError && <p className="mt-2 text-red-700">{reviewError}</p>}
<p className="mt-2 text-xs text-gray-400">
确认=采纳候选;修订=给出人工结论(保留 AI 原始候选);驳回≠通过(无替代结论时保留
UNKNOWN);NA 需留理由并缩小评分范围。
</p>
</section>
)}

{images.map((img, i) => {
const bbox = finding.evidence_refs[i]?.locator.bbox;
return (
Expand Down
50 changes: 34 additions & 16 deletions apps/web/lib/api.ts
Original file line number Diff line number Diff line change
Expand Up @@ -134,6 +134,36 @@ export const api = {
findings: FindingSummary[];
}>(`/audits/${auditID}`),

submitReview: (
findingID: string,
body: {
decision: string;
note?: string;
edits?: { assessment_status?: string };
na_reason?: string;
},
) =>
request<FindingSummary>(`/findings/${findingID}/reviews`, {
method: "POST",
body: JSON.stringify(body),
}),
finalizeAudit: (auditID: string) =>
request<{
total_score: number | null;
partial: boolean;
coverage_pct: number;
dimensions: { dimension: string; score: number }[];
blocking: { rule_id: string; severity: string }[];
}>(`/audits/${auditID}/finalize`, { method: "POST", body: JSON.stringify({}) }),
updateTask: (
findingID: string,
body: { workflow_status: string; version: number; reason?: string },
) =>
request<{ workflow_status: string }>(`/findings/${findingID}/task`, {
method: "PATCH",
body: JSON.stringify(body),
}),

getEvidence: (evidenceID: string) =>
request<{
id: string;
Expand All @@ -142,22 +172,7 @@ export const api = {
download_url: string | null;
sha256: string;
}>(`/evidence/${evidenceID}`),
getFinding: (findingID: string) =>
request<{
id: string;
rule_id: string;
assessment_status: string;
severity: string;
observation: string | null;
reason: string | null;
recommended_fix: string | null;
confidence: number;
review_status: string;
evidence_refs: {
evidence_id: string;
locator: { type: string; bbox?: { x: number; y: number; w: number; h: number } };
}[];
}>(`/findings/${findingID}`),
getFinding: (findingID: string) => request<FindingSummary>(`/findings/${findingID}`),
};

export type FindingSummary = {
Expand All @@ -166,8 +181,11 @@ export type FindingSummary = {
assessment_status: "PASS" | "WARN" | "FAIL" | "UNKNOWN";
severity: string;
observation: string | null;
reason: string | null;
recommended_fix: string | null;
confidence: number;
review_status: string;
original_candidate: unknown;
evidence_refs: {
evidence_id: string;
locator: { type: string; bbox?: { x: number; y: number; w: number; h: number } };
Expand Down
Loading
Loading