merge: preserve the latest Wabi research in the document database migration

README.md

@@ -30,11 +30,11 @@ sudo /opt/endorphin-tabularium-docs/docs save < /private/work/purpose.json
 
 路径仅为操作示例，应选择实际存在且受保护的目录。运行成功后刷新网页即可看到正文、导航和搜索更新。版本冲突时重新读取并合并修改；不要把旧 revision 改成新数字强行覆盖。草案是现有展示标记，不是按分支或用户隔离的访问控制。
 
-新增正文不提交回 Git。奇思妙想这类内容同样通过数据库写入；代码分支与数据发布是不同动作，不能因为迁移程序合并就自动发布其他未合并分支的内容。
+新增正文不提交回 Git。奇思妙想这类内容同样通过数据库写入；本次用户明确要求汇总四个 PR，因此该草案已纳入首次迁移清单，保留草案状态。代码分支与数据发布是不同动作，其他分支的内容不会自动发布。
 
 ## 首次迁移与备份
 
-迁移前主线的 50 篇正文及目录可从固定历史提交 `25c5e83d08eae4277fe429a692955bd3bb2e717f` 完整导出：
+本次汇总先合入 PR #14、#15、#16，再应用 PR #17 的数据库迁移。最新研究正文及奇思妙想草案合计 51 篇，连同目录可从固定历史提交 `cffca6942b1bc0005321e1685f38593fb0b853d5` 完整导出；其中 AI 公司研究三篇采用汇总期间追加的 `74af205b2959e1bb6a1fe9f567bc87069ac6d12c` 历史版本，保留 Wabi 更新：
 
 ```sh
 python3 scripts/export-legacy-documents.py --output .artifacts/documents-migration.json
@@ -104,11 +104,16 @@ sudo /opt/endorphin-tabularium-docs/auth verify
 同一版本内容变化会拒绝导入，避免静默读取旧数据。回退代码时继续读取对应旧版本。
 所有修改都经过工作分支和 PR，部署仍等待人工合并主线。
 
-当前资料包含 1,015 条公司/产品记录、10 类业务及待分类组、1,091 条证据记录和 183 条经营与技术观测。
+当前资料包含 1,021 条公司/产品记录、10 类业务及待分类组、1,201 条证据记录和 187 条经营与技术观测。
-其中保留 86 家个案资料、913 条目录基础资料，并新增 16 个平台产品/业务样本；目录项不等于完成财务和技术核验。
+其中保留 93 家个案资料、912 条目录基础资料，并新增 16 个平台产品/业务样本；目录项不等于完成财务和技术核验。
+Product Hunt两年档案各首批19张、合计38张发布/推广卡片以`candidateBatches`保存；公司成立年待核者不自动新增公司。其中18条仍待核成立年；排除依据区分正式成立字段与2025前已运营证据，精确注册日期未核者明确标注。候选元数据与其他研究记录同库持久化，覆盖报告逐条展示状态、缺项和公司关联。Kilo Code与KiloClaw两项映射一个已被收购的2025公司个案，采用规模、token吞吐及同名健康公司的收入不作为其ARR。
+机器之心首批4条公司线索以独立候选批次记录：直达正文未读，经公司公开主页、官方代码或发起方公告补核PrismShadow与Stargate；SSI用2024官方历史更新排除，Utopai仍待核。不是两年中文来源全量。
+TechCrunch栏目首批3条公司候选已追溯原始材料；Wabi据两个公开名录2025年份入表，精确注册日未核。官方条款称当前免费，收入仍未知，不填零。DreamHQ及Thrive Holdings成立年仍待核，后者2025公开发布公告另存。DreamHQ与名录同名摄像头公司区分，ADT拟收购尚待交割，融资、估值和目标公司业绩不当新AI公司的ARR。分类接口与RSS访问403及两年档案缺项保留。
 补充ThinkWheel与中国NaiveAI；MiroMind成立年未核，公开研发年份单列。保留灵初智能成立年排除、深度原理日期待核和Naive近名实体复核记录。
-新增7条Speedrun个案，保留标题/正文冲突、七位数数量级、币种不明、合同管线与履历归属排除记录；相对日期不充当同期收入时点。核查页优先按人工确认的公司ID关联明细，旧版本保留按证据归属回退；54条核查中43条指向具体公司、3条属于全局方法、8条是未纳入候选。明细反向列出关联核查；币种不明的声明保留原值，不绘制美元收入长度条。
+新增7条Speedrun个案，保留标题/正文冲突、七位数数量级、币种不明、合同管线与履历归属排除记录；相对日期不充当同期收入时点。核查页优先按人工确认的公司ID关联明细，旧版本保留按证据归属回退；68条核查中52条指向具体公司、4条属于全局方法、12条是未纳入候选。明细反向列出关联核查；币种不明的声明保留原值，不绘制美元收入长度条。
+Lightfield按Magical Tome主体与2025前历史业务列为CRM转型对照，成立年未知，不计新成立公司；Fundraisly成立年待核，其客户融资和创始人历史ARR不转移为公司收入。
 Bud保留为2025成立且YC标记已收购的案例；Rocket.new按前身2021成立、2025产品上线的转型对照记录，历史ARR不混入新成立公司同期统计。
+原候选36088 EveryApp从AI不明经补证入表，`directoryRechecks`保存后续复核，原1,428条台账及928条原始新增计数不改。OpenSEO与EveryApp不重复计公司；定价、星标、客户数和数据成本加价均不转为ARR。
 验收按2026-10-30确认的公开来源清单边界逐条核对并标明缺项；YC候选池有完整读取台账，其他来源目前为定向补充，不能把来源出现次数当作覆盖完成。收入披露不足且统计时点不统一，不代表全部 AI 公司。
 界面将公司自述、第三方估算、融资、客户收入和履历收入分开，未知收入不记为零。
 保留被撤回的声明作为核查记录，并排除收入比较；并购后业务、管理层预测与历史里程碑单列。
@@ -137,6 +142,7 @@ MRR×12仅作为显示单位转换，不改写为公司ARR；公司ARR、估算
 逐条读取对应 YC 官方公开页面，以页面中的成立年份、团队字段和业务描述筛选。
 962 条初筛匹配，其中 34 条与原表 URL 重合，实际新增 928 条；268 条年份不符或缺失、
 192 条 AI 业务待确认、6 条成立年份晚于批次年份未纳入本批新增。
+`content/research/ai-companies.sources.json` 保留最初12项来源与后续补充渠道、用途、覆盖边界及下一步；覆盖报告逐渠道复算证据、公司及观测数量。背景报告、入口可读和候选池逐条读取分别标记，零计数不表示没有公司。
 `content/research/ai-companies.coverage.json` 保留逐条筛选结果、来源与响应摘要，
 数据快照中的 `coverage` 也随版本进入同一 SQLite 数据库。只覆盖该候选池，不能称全球全量。
 
@@ -149,7 +155,7 @@ MRR×12仅作为显示单位转换，不改写为公司ARR；公司ARR、估算
 仍有3条因简介不足保留待分类。人工复核分类不代表独立验证技术水平、客户效果或财务数据。
 AI 关键词可能出现在业务背景中，收入、买家和交付状态仍须逐家深查，不能把目录项的未知收入当作未披露。
 技术专项评审用独立字段记录层级、日期、摘要与证据：未评审、已读技术自述、公开代码/模型材料可访问、有第三方独立评测记录。
-当前7条完成该字段的专项记录，其中3条确认材料可访问、1条记录第三方已发表评测（Gradium TTS Sep 2026）；本轮未自行运行模型测试。缺字段不表示无技术能力，代码或模型卡可读不证明性能。
+当前13条完成该字段的专项记录，其中6条确认材料可访问、1条记录第三方已发表评测（Gradium TTS Sep 2026）；本轮未自行运行模型测试。缺字段不表示无技术能力，代码或模型卡可读不证明性能。
 界面支持按资料深度、记录团队规模和技术验证证据筛选；证据列表每页 30 条，避免一次渲染全部来源。
 
 `pnpm test:research` 在远程 Runner 上验证同库持久化、重复读取、版本导入与回退、

apps/docs/scripts/auth-cli.ts

@@ -181,10 +181,10 @@ async function main() {
   assert.equal(researchResponse.status, 200, "authenticated research API");
   assert.ok(researchResponse.headers.get("cache-control")?.includes("no-store"));
   const research = await researchResponse.json();
-  assert.equal(research.companies.length, 1015);
+  assert.equal(research.companies.length, 1021);
   assert.equal(research.categories.length, 11);
-  assert.equal(research.evidence.length, 1091);
+  assert.equal(research.evidence.length, 1201);
-  assert.equal(research.observations.length, 183);
+  assert.equal(research.observations.length, 187);
   assert.equal((await request("/api/research/ai-companies", { method: "POST", headers })).status, 405);
   const researchPage = await request("/docs/other-research/ai-company-research/overview", { headers });
   assert.equal(researchPage.status, 200);
@@ -221,7 +221,8 @@ async function main() {
       saveDocument({ ...first, slug: "endor/ci-live-document-renamed", content: "# Runtime document\n\nUpdated without rebuilding" }, first.revision);
     }
     const stored = readDocument(sourceSlug);
-    assert.ok(stored?.content.includes("Updated without rebuilding"), "DB content survives restart");
+    assert.ok(stored, "Smoke document must exist after creation or restart");
+    assert.ok(stored.content.includes("Updated without rebuilding"), "DB content survives restart");
     const live = await request(`/docs/${stored.slug}`, { headers });
     assert.equal(live.status, 200);
     assert.ok((await live.text()).includes("Updated without rebuilding"));

apps/docs/src/components/IncomeDistribution.tsx

@@ -20,6 +20,9 @@ export default function IncomeDistribution({ data, onSelect }: {
   }), [data, dateWindow, year, annualize]);
   const groups = useMemo(() => groupIncome(sample.points, splitModels).sort((a, b) =>
     a.category.localeCompare(b.category) || a.provider.localeCompare(b.provider) || a.model.localeCompare(b.model)), [sample.points, splitModels]);
+  const categoryGroups = [...new Set(groups.map((group) => group.category))].map((category) => ({
+    category, rows: groups.filter((group) => group.category === category),
+  }));
   const categories = new Map(data.categories.map((category) => [category.id, category.name]));
   const metric = annualize ? "MRR×12参考年化" : "平台MRR";
   const values = sample.points.map((point) => point.value);
@@ -41,6 +44,7 @@ export default function IncomeDistribution({ data, onSelect }: {
       <label>显示单位<select value={annualize ? "annualized" : "mrr"} onChange={(event) => setAnnualize(event.target.value === "annualized")}><option value="mrr">MRR（美元/月）</option><option value="annualized">MRR×12参考年化（美元/年）</option></select></label>
     </div>
     <div className="income-options"><label><input type="checkbox" checked={splitModels} onChange={(event) => setSplitModels(event.target.checked)} /> 按收费与交付模式细分</label><label><input type="checkbox" checked={logScale} onChange={(event) => setLogScale(event.target.checked)} /> 对数横轴</label></div>
+    <p>{splitModels ? "分组：业务类别 × 支付平台 × 收费与交付模式。同一类别下的多行是不同统计组，各自计算均值和中位数。" : "分组：业务类别 × 支付平台。同一平台内的收费模式已合并，不同支付平台仍分别统计。"}</p>
     <p className="research-notice">资料截至 {data.asOf}，纳入 {sample.points.length} 个产品，另有 {sample.excluded.length} 个 MRR 观测未纳入。{annualize ? "乘12只改变显示单位，不证明全年收入、续约或ARR；CV与相对敏感性不变。" : "MRR不等于近30日收款；平台数据连接也不等于财务审计。"}{dateWindow === "all" && " 当前包含不同时间的历史观测，不能称为同期对照。"}</p>
     {sample.points.length ? <>
       <figure>
@@ -48,9 +52,16 @@ export default function IncomeDistribution({ data, onSelect }: {
           {ticks.map((tick) => <g key={tick}><line x1={x(tick)} x2={x(tick)} y1={15} y2={groups.length * 70 + 10} stroke="#dfe6e0" /><text x={x(tick)} y={groups.length * 70 + 35} textAnchor="middle" fontSize="11" fill="#637066">{usd(tick)}</text></g>)}
           {groups.map((group, index) => {
             const y = index * 70 + 40;
-            return <g key={group.key}>
+            const firstInCategory = index === 0 || groups[index - 1].category !== group.category;
-              <text x={8} y={y - 8} fontSize="12" fill="#202724">{categories.get(group.category)}</text>
+            const categoryRowCount = categoryGroups.find((entry) => entry.category === group.category)!.rows.length;
-              <text x={8} y={y + 10} fontSize="10" fill="#637066">{group.provider} · {group.model} · n={group.stats.n}</text>
+            return <g key={group.key} role="group" aria-label={`${categories.get(group.category)}，${group.provider}，${group.model}，${group.stats.n}个产品`}>
+              {firstInCategory && <>
+                {index > 0 && <line x1={8} x2={right} y1={y - 34} y2={y - 34} stroke="#dfe6e0" />}
+                <text x={8} y={y - 8} fontSize="12" fontWeight="600" fill="#202724">{categories.get(group.category)}（{categoryRowCount}组）</text>
+                {categoryRowCount > 1 && <line x1={10} x2={10} y1={y + 3} y2={y + (categoryRowCount - 1) * 70 + 12} stroke="#95b4a6" strokeWidth={2} />}
+              </>}
+              <text x={20} y={y + 10} fontSize="10" fill="#637066">{group.provider} · {group.model} · n={group.stats.n}</text>
+              {group.stats.n < 3 && <text x={20} y={y + 26} fontSize="9" fill="#637066">{group.stats.n === 1 ? "仅1个产品，无法判断组内离散度" : "仅2个产品，不代表领域典型收入"}</text>}
               <line x1={x(group.stats.min)} x2={x(group.stats.max)} y1={y} y2={y} stroke="#95b4a6" strokeWidth={2} />
               <line x1={x(group.stats.mean)} x2={x(group.stats.mean)} y1={y - 17} y2={y + 17} stroke="#b17725" strokeWidth={3}><title>样本均值 {usd(group.stats.mean)}</title></line>
               <path d={`M${x(group.stats.median)},${y - 7} l7,7 l-7,7 l-7,-7 Z`} fill="#8a78ad"><title>样本中位数 {usd(group.stats.median)}</title></path>
@@ -62,12 +73,13 @@ export default function IncomeDistribution({ data, onSelect }: {
             </g>;
           })}
         </svg></div>
-        <figcaption>圆点：产品观测　竖线：样本均值　菱形：样本中位数。按业务类别与支付来源分组，点击点或下表产品名可看原始证据。</figcaption>
+        <figcaption>圆点：产品观测　竖线：样本均值　菱形：样本中位数。业务类别标题下各行按支付平台{splitModels && "及收费与交付模式"}分别统计；n为该组产品数。点击点或下表产品名可看原始证据。</figcaption>
       </figure>
-      <div className="research-table-wrap"><table><caption>{metric}样本统计；金额按显示值计算，CV使用样本标准差</caption><thead><tr><th>类别 / 来源 / 模式</th><th>n</th><th>均值</th><th>中位数</th><th>均值/中位数</th><th>CV</th><th>去除最大值后的均值 / 下降幅度</th><th>产品与限制</th></tr></thead><tbody>{groups.map((group) => <tr key={group.key}>
+      <div className="research-table-wrap"><table><caption>{metric}样本统计；每行是类别下的一个统计组，CV使用样本标准差</caption><thead><tr><th scope="col">业务类别</th><th scope="col">支付平台 / 收费模式</th><th scope="col">n</th><th scope="col">均值</th><th scope="col">中位数</th><th scope="col">均值/中位数</th><th scope="col">CV</th><th scope="col">去除最大值后的均值 / 下降幅度</th><th scope="col">产品与限制</th></tr></thead>{categoryGroups.map(({ category, rows }) => <tbody key={category}>{rows.map((group, index) => <tr key={group.key}>
-        <td>{categories.get(group.category)}<small>{group.provider} · {group.model}</small></td><td>{group.stats.n}</td><td>{usd(group.stats.mean)}</td><td>{usd(group.stats.median)}</td><td>{group.stats.meanMedianRatio?.toFixed(2) ?? "—"}</td><td>{group.stats.cv?.toFixed(2) ?? "—"}</td><td>{usd(group.stats.meanWithoutLargest)}<small>{percent(group.stats.largestRemovalDrop)}</small></td>
+        {index === 0 && <th scope="rowgroup" rowSpan={rows.length}>{categories.get(category)}<small>{rows.length}个统计组</small></th>}
+        <td>{group.provider}<small>{group.model}</small></td><td>{group.stats.n}</td><td>{usd(group.stats.mean)}</td><td>{usd(group.stats.median)}</td><td>{group.stats.meanMedianRatio?.toFixed(2) ?? "—"}</td><td>{group.stats.cv?.toFixed(2) ?? "—"}</td><td>{usd(group.stats.meanWithoutLargest)}<small>{percent(group.stats.largestRemovalDrop)}</small></td>
         <td>{group.points.map((point) => <span key={point.company.id}><button className="research-company-name" onClick={() => onSelect(point.company.id)}>{point.company.name}</button>{" "}</span>)}<small>{group.stats.n < 3 ? "样本少于3，不能据此判断领域典型收入。" : "自选的小样本，须结合产品年龄、成本与留存评估。"}</small></td>
-      </tr>)}</tbody></table></div>
+      </tr>)}</tbody>)}</table></div>
     </> : <p className="research-message">当前窗口和创立年份没有符合条件的 MRR 记录。</p>}
     <details><summary>统计规则与解释</summary><p>每个产品只取符合条件的最新一条平台 MRR；币种为美元，要求有效数据日期、确定数值及支付来源。公司 ARR、估算、融资、目标和近30日收入均不进入这里。未知值不补零。</p><p>CV＝样本标准差÷均值，标准差分母为 n−1；n=1 不算 CV。n=2 时均值必然等于中位数，不能据此认为均值有代表性。较大的均值/中位数差距、CV或移除最大值的变化，仅提示进一步核查样本构成。</p><p>支付商与平台的退款、试用、年付、币种及指标计算可能不同（<a href="https://trustmrr.com/faq" target="_blank" rel="noopener noreferrer">平台方法说明</a>）。不能将 MRR 与收款差异、单个高值或高离散度直接解释为财务造假。比较商业模式还需要毛利、获客成本、留存、人工交付和客户集中度。</p></details>
     {sample.excluded.length > 0 && <details><summary>未纳入的 MRR 记录（{sample.excluded.length}）</summary><ul>{sample.excluded.map(({ observation, reason }) => <li key={observation.id}><button onClick={() => onSelect(observation.company)}>{observation.name}</button>：{usd(observation.value)} / 月，{observation.observedAt ?? "日期未知"}；{reason}</li>)}</ul></details>}

apps/docs/src/lib/ai-research.ts

@@ -43,11 +43,24 @@ export type ResearchCoverage = {
   otherOrMissingYear: number; unclearAI: number; dateConflicts: number; failedProfiles: number;
   discoverySha256: string;
 };
+export type ResearchCandidateBatch = {
+  id: string; sourceId: string; url: string; accessedAt: string; scope: string; complete: boolean;
+  records: {
+    id: string; name: string; url: string | null; categoryId: string;
+    kind: "product" | "company" | "promotion"; profileStatus: "readable" | "unavailable" | "archive-only";
+    status: "pending-year" | "linked" | "existing-outside-cohort" | "outside-cohort" | "promotion";
+    companyId?: string; evidenceIds?: string[]; note: string;
+  }[];
+};
 export type ResearchSnapshot = {
   version: string; asOf: string; scope: string; limitations: string[];
   categories: ResearchCategory[]; companies: ResearchCompany[];
   observations: ResearchObservation[]; evidence: ResearchEvidence[];
-  reviews: ResearchReview[]; coverage?: ResearchCoverage;
+  reviews: ResearchReview[]; coverage?: ResearchCoverage; candidateBatches?: ResearchCandidateBatch[];
+  directoryRechecks?: {
+    ycId: number; originalStatus: "ai-unclear" | "year-outside-or-missing" | "date-conflict";
+    companyId: string; reviewedAt: string; evidenceIds: string[]; reason: string;
+  }[];
 };
 
 export function evidenceUrl(value: string): string | undefined {
@@ -98,7 +111,40 @@ export function validateResearch(snapshot: ResearchSnapshot) {
         !Number.isFinite(o.value))
       throw new Error(`Invalid research observation: ${o.id}`);
   }
+  unique((snapshot.candidateBatches ?? []).map((b) => b.id), "candidate batches");
+  unique((snapshot.candidateBatches ?? []).flatMap((b) => b.records.map((r) => r.id)), "candidates");
+  for (const batch of snapshot.candidateBatches ?? []) {
+    if (!/^S\d{2}$/.test(batch.sourceId) || !evidenceUrl(batch.url) || !batch.scope.trim() ||
+        typeof batch.complete !== "boolean" || !/^\d{4}-\d{2}-\d{2}$/.test(batch.accessedAt) ||
+        Number.isNaN(Date.parse(batch.accessedAt)) ||
+        new Date(batch.accessedAt).toISOString().slice(0, 10) !== batch.accessedAt ||
+        batch.accessedAt > snapshot.asOf)
+      throw new Error(`Invalid research candidate batch: ${batch.id}`);
+    for (const row of batch.records) {
+      const linked = ["linked", "existing-outside-cohort"].includes(row.status);
+      if (!row.name.trim() || !row.note.trim() || !categories.has(row.categoryId) ||
+          !["product", "company", "promotion"].includes(row.kind) ||
+          !["readable", "unavailable", "archive-only"].includes(row.profileStatus) ||
+          !["pending-year", "linked", "existing-outside-cohort", "outside-cohort", "promotion"].includes(row.status) ||
+          (row.kind !== "promotion" && (!row.url || !evidenceUrl(row.url))) ||
+          (row.url !== null && !evidenceUrl(row.url)) ||
+          linked !== !!row.companyId || (row.companyId && !companies.has(row.companyId)) ||
+          (row.kind === "promotion") !== (row.status === "promotion") ||
+          (row.evidenceIds ?? []).some((id) => !evidence.has(id)) ||
+          (row.status === "outside-cohort" && !row.evidenceIds?.length))
+        throw new Error(`Invalid research candidate: ${row.id}`);
+    }
+  }
   const coverage = snapshot.coverage;
+  unique((snapshot.directoryRechecks ?? []).map((row) => String(row.ycId)), "directory rechecks");
+  for (const row of snapshot.directoryRechecks ?? []) {
+    if (!Number.isSafeInteger(row.ycId) || row.ycId <= 0 || !companies.has(row.companyId) ||
+        !["ai-unclear", "year-outside-or-missing", "date-conflict"].includes(row.originalStatus) ||
+        !/^\d{4}-\d{2}-\d{2}$/.test(row.reviewedAt) || Number.isNaN(Date.parse(row.reviewedAt)) ||
+        new Date(row.reviewedAt).toISOString().slice(0, 10) !== row.reviewedAt || row.reviewedAt > snapshot.asOf ||
+        !row.reason.trim() || !row.evidenceIds.length || row.evidenceIds.some((id) => !evidence.has(id)))
+      throw new Error(`Invalid directory recheck: ${row.ycId}`);
+  }
   if (coverage && (coverage.profilesRead + coverage.failedProfiles !== coverage.candidateRecords ||
       coverage.matchingProfiles !== coverage.existingProfiles + coverage.newProfiles ||
       coverage.matchingProfiles + coverage.otherOrMissingYear + coverage.unclearAI + coverage.dateConflicts !== coverage.profilesRead ||

apps/docs/tests/research.test.ts

@@ -23,9 +23,11 @@ test("research revisions share the authentication database and preserve replay l
     for (let i = 0; i < 20; i++) consumeCode(config, "invalid");
     assert.equal(consumeCode(config, "invalid"), "limited");
     const initial = readResearchSnapshot();
-    assert.equal(initial.companies.length, 1015);
+    assert.equal(initial.companies.length, 1021);
-    assert.equal(initial.evidence.length, 1091);
+    assert.equal(initial.evidence.length, 1201);
-    assert.equal(initial.observations.length, 183);
+    assert.equal(initial.observations.length, 187);
+    assert.deepEqual(initial.candidateBatches, seed.candidateBatches, "candidate gaps persist in the same database metadata");
+    assert.deepEqual(initial.directoryRechecks, seed.directoryRechecks, "follow-up classifications persist without rewriting the discovery audit");
     assert.deepEqual(readResearchSnapshot(), initial, "reopening does not duplicate records");
     const db = openSiteDatabase();
     const tables = db.prepare("SELECT name FROM sqlite_master WHERE type='table'").all() as { name: string }[];
@@ -62,9 +64,9 @@ test("research validation rejects broken evidence and unsafe source links", () =
 
 test("reviews link explicit companies without turning global or excluded cases into companies", () => {
   const snapshot = structuredClone(seed) as ResearchSnapshot;
-  assert.equal(snapshot.reviews.filter((r) => r.scope === "companies").length, 43);
+  assert.equal(snapshot.reviews.filter((r) => r.scope === "companies").length, 52);
-  assert.equal(snapshot.reviews.filter((r) => r.scope === "dataset").length, 3);
+  assert.equal(snapshot.reviews.filter((r) => r.scope === "dataset").length, 4);
-  assert.equal(snapshot.reviews.filter((r) => r.scope === "excluded-candidate").length, 8);
+  assert.equal(snapshot.reviews.filter((r) => r.scope === "excluded-candidate").length, 12);
   const review = snapshot.reviews.find((r) => r.id === "R4")!;
   assert.deepEqual(reviewCompanyIds(snapshot, review), ["AI007", "AI013", "AI021", "AI034"]);
   assert.deepEqual(reviewCompanyIds(snapshot, snapshot.reviews.find((r) => r.id === "CN-R-PSI")!), []);
@@ -83,11 +85,11 @@ test("reviews link explicit companies without turning global or excluded cases i
 test("disclosure counts exclude client income, biographies, targets and duplicate milestones", () => {
   const snapshot = seed as ResearchSnapshot;
   const summary = categorySummary(snapshot);
-  assert.equal(summary.reduce((n, c) => n + c.count, 0), 1015);
+  assert.equal(summary.reduce((n, c) => n + c.count, 0), 1021);
   assert.equal(summary.reduce((n, c) => n + c.self, 0), 26);
   assert.equal(summary.reduce((n, c) => n + c.estimate, 0), 1);
   assert.equal(summary.reduce((n, c) => n + c.platform, 0), 16);
-  assert.equal(summary.reduce((n, c) => n + c.unknown, 0), 972);
+  assert.equal(summary.reduce((n, c) => n + c.unknown, 0), 978);
   assert.equal(snapshot.companies.find((c) => c.name === "Rork")?.revenueStatus, "unknown");
   assert.equal(snapshot.companies.find((c) => c.name === "Result")?.revenueStatus, "unknown");
   assert.ok(snapshot.observations.filter((o) => o.selection === "signal").every((o) => o.scope === "公司整体"));
@@ -95,8 +97,8 @@ test("disclosure counts exclude client income, biographies, targets and duplicat
 
 test("technical reviews keep accessible materials separate from tested capabilities", () => {
   const snapshot = structuredClone(seed) as ResearchSnapshot;
-  assert.equal(snapshot.companies.filter((c) => c.technicalReview).length, 7);
+  assert.equal(snapshot.companies.filter((c) => c.technicalReview).length, 13);
-  assert.equal(snapshot.companies.filter((c) => c.technicalReview?.level === "public-artifact").length, 3);
+  assert.equal(snapshot.companies.filter((c) => c.technicalReview?.level === "public-artifact").length, 6);
   assert.equal(snapshot.companies.filter((c) => c.technicalReview?.level === "independent-test").length, 1);
   assert.equal(snapshot.companies.find((c) => c.id === "AI051")?.revenueStatus, "unknown");
   assert.match(snapshot.companies.find((c) => c.id === "AI051")!.technicalReview!.testScope!, /Sep 2026/);
@@ -255,3 +257,88 @@ test("distribution statistics match independently calculated snapshot references
     }
   }
 });
+
+
+test("product launch candidates preserve unresolved years and many launches per company", () => {
+  const snapshot = structuredClone(seed) as ResearchSnapshot;
+  const batch = snapshot.candidateBatches![0];
+  assert.equal(batch.complete, false);
+  assert.equal(batch.records.length, 19);
+  const kilo = batch.records.filter((r) => r.companyId === "AI072");
+  assert.equal(kilo.length, 2, "IDE and hosted products resolve to one company");
+  assert.equal(snapshot.companies.filter((c) => c.id === "AI072").length, 1);
+  assert.ok(batch.records.some((r) => r.status === "pending-year" && !r.companyId));
+  assert.ok(batch.records.some((r) => r.status === "existing-outside-cohort" && r.companyId === "AI054"));
+  kilo[0].companyId = "missing-company";
+  assert.throws(() => validateResearch(snapshot), /research candidate/);
+});
+
+test("annual launch archives exclude older companies only with evidence and reuse existing records", () => {
+  const snapshot = structuredClone(seed) as ResearchSnapshot;
+  const batch = snapshot.candidateBatches!.find((b) => b.id === "PH-2025-visible-1")!;
+  assert.equal(batch.complete, false);
+  assert.equal(batch.records.length, 19);
+  const older = batch.records.filter((r) => r.status === "outside-cohort");
+  assert.equal(older.length, 7);
+  assert.ok(older.every((r) => !r.companyId && r.evidenceIds?.length));
+  assert.equal(batch.records.find((r) => r.name === "Trace")?.companyId, "YC30687");
+  const trace = snapshot.companies.find((c) => c.id === "YC30687")!;
+  assert.equal(trace.profileOnly, false);
+  assert.equal(trace.directoryOrigin, "yc-2025-2026");
+  assert.equal(trace.revenueStatus, "unknown");
+  assert.ok(!snapshot.observations.some((o) => o.evidence === "TRACE-HOME"), "demo metrics must not become reported results");
+  older[0].evidenceIds = [];
+  assert.throws(() => validateResearch(snapshot), /research candidate/);
+  older[0].evidenceIds = ["missing-proof"];
+  assert.throws(() => validateResearch(snapshot), /research candidate/);
+});
+
+test("unreconciled directory financials remain source reviews instead of income observations", () => {
+  const snapshot = seed as ResearchSnapshot;
+  const review = snapshot.reviews.find((r) => r.id === "PH-R-SOURCE-QUALITY")!;
+  assert.equal(review.scope, "dataset");
+  assert.ok(review.evidenceIds.includes("PH-TANA-DR"));
+  assert.equal(reviewCompanyIds(snapshot, review).length, 0);
+  assert.ok(!snapshot.observations.some((o) => o.evidence === "PH-TANA-DR"));
+  assert.equal(snapshot.candidateBatches!.flatMap((b) => b.records).filter((r) => r.status === "pending-year").length, 21);
+});
+
+test("product pivots and founder histories do not become new company years or current revenue", () => {
+  const snapshot = seed as ResearchSnapshot;
+  const pivot = snapshot.companies.find((c) => c.id === "AI073")!;
+  assert.equal(pivot.year, null, "CRM launch is not a verified company founding year");
+  assert.equal(pivot.revenueStatus, "unknown");
+  const candidate = snapshot.candidateBatches!.flatMap((b) => b.records).find((r) => r.id === "PH26-11")!;
+  assert.equal(candidate.status, "existing-outside-cohort");
+  assert.equal(candidate.companyId, pivot.id);
+  assert.ok(!platformIncomeSample(snapshot, { windowDays: null }).points.some((point) => point.company.id === pivot.id));
+  assert.ok(!snapshot.observations.some((o) => ["PH-FUNDRAISLY-ABOUT", "LF-PIVOT", "LF-HISTORY"].includes(o.evidence)), "other ventures' ARR, users and funding must not transfer");
+});
+
+test("follow-up evidence can admit an uncertain directory candidate without rewriting its original status", () => {
+  const snapshot = structuredClone(seed) as ResearchSnapshot;
+  const followup = snapshot.directoryRechecks!.find((row) => row.ycId === 36088)!;
+  const original = coverageAudit.records.find((row) => row.ycId === followup.ycId)!;
+  assert.equal(original.status, "ai-unclear");
+  assert.equal(followup.originalStatus, original.status);
+  assert.equal(snapshot.companies.find((c) => c.id === followup.companyId)?.year, 2026);
+  assert.equal(snapshot.candidateBatches!.flatMap((b) => b.records).find((r) => r.name === "OpenSEO")?.companyId, followup.companyId);
+  assert.ok(!snapshot.observations.some((o) => ["SEO-PRICE", "UP-YC", "PAZI-PRICE"].includes(o.evidence)), "pricing, other ventures' ARR and prepaid balances stay outside income observations");
+  followup.companyId = "missing-company";
+  assert.throws(() => validateResearch(snapshot), /directory recheck/);
+});
+
+test("Chinese discovery entries preserve unread articles and use separate primary evidence", () => {
+  const snapshot = structuredClone(seed) as ResearchSnapshot;
+  const batch = snapshot.candidateBatches!.find((b) => b.id === "JQ-2026-targeted-1")!;
+  assert.equal(batch.complete, false);
+  assert.equal(batch.records.length, 4);
+  assert.ok(batch.records.every((row) => row.kind === "company" && row.profileStatus === "unavailable"));
+  assert.equal(batch.records.filter((row) => row.status === "linked").length, 2);
+  assert.equal(batch.records.find((row) => row.name === "Utopai")?.status, "pending-year");
+  assert.ok(!snapshot.observations.some((o) => ["PS-CODE", "SG-START", "SG-APR", "SSI-HISTORY"].includes(o.evidence)), "task costs, investment, capacity and financing are not revenue");
+  const prism = snapshot.companies.find((c) => c.id === "AI075")!;
+  assert.equal(prism.team, null, "a team range or linked employees must not become exact headcount");
+  batch.records[0].url = null;
+  assert.throws(() => validateResearch(snapshot), /research candidate/);
+});

content/research/ai-companies.sources.json

@@ -0,0 +1,242 @@
+{
+  "version": "2026-10-30-9",
+  "scopeConfirmedAt": "2026-10-30",
+  "acceptance": "以确认的公开来源清单为边界，逐条核对并标明缺项。",
+  "provenance": "S01—S12恢复自已提交用户的《AI公司研究_来源与分类模板》来源清单；S13—S16单列后续已使用的补充证据渠道，不将使用次数当作全量覆盖。",
+  "sources": [
+    {
+      "id": "S01",
+      "name": "Y Combinator",
+      "url": "https://www.ycombinator.com/companies",
+      "role": "discovery",
+      "status": "candidate-audit",
+      "hosts": [
+        "www.ycombinator.com"
+      ],
+      "initialAccess": "官网目录可访问",
+      "boundary": "已读取八个2025—2026批次镜像候选池1428条，再核YC官方页面；不声称覆盖全部YC历年或未公开公司。",
+      "nextCheck": "继续复核规则初分和缺项；候选明细见逐条台账。",
+      "limitation": "批次不等于成立年；不代表全部公司"
+    },
+    {
+      "id": "S02",
+      "name": "TechCrunch AI",
+      "url": "https://techcrunch.com/category/artificial-intelligence/",
+      "role": "reporting",
+      "status": "targeted",
+      "hosts": [
+        "techcrunch.com"
+      ],
+      "initialAccess": "AI栏目可读取",
+      "boundary": "栏目首页选择的3条公司线索已追溯原始资料：Wabi据公开名录2025年份关联公司；DreamHQ与Thrive Holdings成立年待核，后者2025发布公告另存。不是首页全部条目或两年档案；公开分类接口与RSS普通HTTP请求返回403。",
+      "nextCheck": "补可公开取得的2025、2026档案及剩余分页；核DreamHQ、Thrive Holdings成立依据及Wabi精确注册、团队和付费口径。",
+      "limitation": "需追溯原始披露，部分文章受访问限制",
+      "accessRecheckedAt": "2026-10-30"
+    },
+    {
+      "id": "S03",
+      "name": "Dealroom",
+      "url": "https://dealroom.co/",
+      "role": "discovery",
+      "status": "targeted",
+      "hosts": [
+        "dealroom.co",
+        "app.dealroom.co"
+      ],
+      "initialAccess": "官网可读取",
+      "boundary": "已读Tana与Wabi公开公司档案：前者财务字段不一致留缺，后者2025成立字段与官方业务关联入表；尚未取得两年成立筛选结果集，两个个案不等于目录全量。",
+      "nextCheck": "确认公开可取得的两年成立筛选清单；逐项追溯财务原始凭证及字段口径，受权限限制保留缺项。",
+      "limitation": "完整档案需权限；概览及财务表不能未经勾稽直接导入ARR",
+      "accessRecheckedAt": "2026-10-30"
+    },
+    {
+      "id": "S04",
+      "name": "Product Hunt",
+      "url": "https://www.producthunt.com/",
+      "role": "discovery",
+      "status": "targeted",
+      "hosts": [
+        "www.producthunt.com",
+        "producthunt.com"
+      ],
+      "initialAccess": "官网可读取",
+      "boundary": "2026与2025年度档案各登记首个公开响应19条，共38条候选/推广；2025另有一个仅图片位置未解析。已逐条保留关联、范围外与年份待核状态，未穷尽两年档案。",
+      "nextCheck": "继续补两年其余分页/日期、仅图片位置及18条年份待核候选；保留Pazi日期冲突，核清品牌与主体，避免转移历史业绩。",
+      "limitation": "上线时间不等于成立时间",
+      "accessRecheckedAt": "2026-10-30"
+    },
+    {
+      "id": "S05",
+      "name": "Sacra",
+      "url": "https://sacra.com/",
+      "role": "estimation",
+      "status": "targeted",
+      "hosts": [
+        "sacra.com"
+      ],
+      "initialAccess": "官网可读取",
+      "boundary": "已用于定向估算及日期口径交叉核查；无全站公司名录读取台账。",
+      "nextCheck": "保留估算与公司披露差异，核实历史截至日。",
+      "limitation": "部分付费；不能把估算当审计数"
+    },
+    {
+      "id": "S06",
+      "name": "Bessemer",
+      "url": "https://www.bvp.com/atlas/the-state-of-ai-2025",
+      "role": "context",
+      "status": "context",
+      "hosts": [
+        "www.bvp.com",
+        "bvp.com"
+      ],
+      "initialAccess": "报告页面可读取",
+      "boundary": "用于行业分类和增长研究背景；不是新成立公司普查。",
+      "nextCheck": "核报告样本、方法与年份，不能将组合公司清单当总体。",
+      "limitation": "投资机构选样偏差"
+    },
+    {
+      "id": "S07",
+      "name": "Menlo Ventures",
+      "url": "https://menlovc.com/perspective/2025-the-state-of-generative-ai-in-the-enterprise/",
+      "role": "context",
+      "status": "context",
+      "hosts": [
+        "menlovc.com",
+        "www.menlovc.com"
+      ],
+      "initialAccess": "报告页面可读取",
+      "boundary": "用于企业采购背景；不把调研比例转换为公司ARR。",
+      "nextCheck": "保留调研方法、采购样本及与新公司样本的差异。",
+      "limitation": "口径与样本需查看方法说明"
+    },
+    {
+      "id": "S08",
+      "name": "Stripe",
+      "url": "https://stripe.com/annual-updates/2025",
+      "role": "context",
+      "status": "context",
+      "hosts": [
+        "stripe.com",
+        "www.stripe.com"
+      ],
+      "initialAccess": "年度信可读取",
+      "boundary": "年度平台观察，不提供可公开穷尽的新AI商户名录。",
+      "nextCheck": "与TrustMRR支付重建数据分开；不据平台总量推算单家公司。",
+      "limitation": "平台样本不能代表全部AI公司"
+    },
+    {
+      "id": "S09",
+      "name": "SaaS Capital",
+      "url": "https://www.saas-capital.com/research/",
+      "role": "context",
+      "status": "context",
+      "hosts": [
+        "www.saas-capital.com",
+        "saas-capital.com"
+      ],
+      "initialAccess": "研究页可读取",
+      "boundary": "B2B订阅软件基准，不能直接代表新AI公司、硬件或服务企业。",
+      "nextCheck": "按报告样本及营收规模使用，不强套为行业ARR分布。",
+      "limitation": "不直接套用硬件与项目制业务"
+    },
+    {
+      "id": "S10",
+      "name": "机器之心",
+      "url": "https://www.jiqizhixin.com/",
+      "role": "reporting",
+      "status": "targeted",
+      "hosts": [
+        "www.jiqizhixin.com",
+        "jiqizhixin.com",
+        "pro.jiqizhixin.com"
+      ],
+      "initialAccess": "首页可读取",
+      "boundary": "已登记4篇定向检索线索；三篇直达仅登录外壳，一篇读取失败。2项经公开原始资料补证关联公司，1项范围外，1项仍待核；未阅读全文或穷尽两年档案。",
+      "nextCheck": "补公开可读文章、2025档案及Utopai原始成立材料；受限正文留缺，不以搜索摘录代全文。",
+      "limitation": "全文访问受限；仅作发现时明确标识，技术和财务另追原始资料",
+      "accessRecheckedAt": "2026-10-30"
+    },
+    {
+      "id": "S11",
+      "name": "36氪",
+      "url": "https://www.36kr.com/",
+      "role": "reporting",
+      "status": "access-limited",
+      "hosts": [
+        "www.36kr.com",
+        "36kr.com"
+      ],
+      "initialAccess": "安全检测，内容未核验",
+      "boundary": "2026-10-30仍返回安全检测页面，未把首页作为已读文章。",
+      "nextCheck": "使用可公开读取的具体文章或其他原始来源；不绕过访问保护。",
+      "limitation": "暂不作为已读取内容证据",
+      "accessRecheckedAt": "2026-10-30"
+    },
+    {
+      "id": "S12",
+      "name": "公司官网与官方博客",
+      "url": null,
+      "role": "primary",
+      "status": "per-company",
+      "hosts": [],
+      "initialAccess": "按公司逐一核验",
+      "boundary": "公司官网和官方材料逐条绑定证据；不是可预先列尽的全站候选池。",
+      "nextCheck": "核定价、主体、交付与收入定义，自述与独立验证分开。",
+      "limitation": "自述不等于独立审计"
+    },
+    {
+      "id": "S13",
+      "name": "TrustMRR",
+      "url": "https://trustmrr.com/",
+      "role": "platform",
+      "status": "targeted",
+      "hosts": [
+        "trustmrr.com"
+      ],
+      "initialAccess": "见具体证据访问记录",
+      "boundary": "已采集产品与支付快照；公开者自选，不是全部商户或独立财务审计。",
+      "nextCheck": "核同步日、连接失效、支付平台公式及产品/法人归属。",
+      "limitation": "来源可读不等于声明已独立验证。"
+    },
+    {
+      "id": "S14",
+      "name": "a16z Speedrun",
+      "url": "https://speedrun.a16z.com/",
+      "role": "investor",
+      "status": "targeted",
+      "hosts": [
+        "speedrun.a16z.com"
+      ],
+      "initialAccess": "见具体证据访问记录",
+      "boundary": "投资机构公司介绍定向样本，收入数字仍多来自公司。",
+      "nextCheck": "核币种、相对时间及管线/履历；未读取的页面保持缺项。",
+      "limitation": "来源可读不等于声明已独立验证。"
+    },
+    {
+      "id": "S15",
+      "name": "Artificial Analysis",
+      "url": "https://artificialanalysis.ai/",
+      "role": "technical",
+      "status": "targeted",
+      "hosts": [
+        "artificialanalysis.ai"
+      ],
+      "initialAccess": "见具体证据访问记录",
+      "boundary": "第三方已发表技术评测，仅覆盖所列模型版本和任务。",
+      "nextCheck": "核榜单版本、测评方法及置信区间，不将排名变成企业整体能力。",
+      "limitation": "来源可读不等于声明已独立验证。"
+    },
+    {
+      "id": "S16",
+      "name": "其他公开补充资料",
+      "url": null,
+      "role": "supplement",
+      "status": "targeted",
+      "hosts": [],
+      "initialAccess": "见具体证据访问记录",
+      "boundary": "未归入上述来源的公开材料按原始证据链接逐条保留；不视为新增已批准的完整候选池。",
+      "nextCheck": "逐项核来源归属、可靠性及可读正文，不能将搜索摘要当全文。",
+      "limitation": "来源可读不等于声明已独立验证。"
+    }
+  ]
+}

content/research/ai-income-analysis.json

@@ -1,5 +1,5 @@
 {
-  "sourceVersion": "2026-10-30-depth-4",
+  "sourceVersion": "2026-10-30-tc-2",
   "asOf": "2026-10-30",
   "metric": "platform-mrr-usd",
   "sampleUnit": "product-or-business",

scripts/audit-ai-research.py

@@ -9,11 +9,122 @@ from urllib.parse import urlsplit
 ROOT = Path(__file__).resolve().parents[1]
 DATA = json.loads((ROOT / 'content/research/ai-companies.seed.json').read_text())
 AUDIT = json.loads((ROOT / 'content/research/ai-companies.coverage.json').read_text())
+SOURCES = json.loads((ROOT / 'content/research/ai-companies.sources.json').read_text())
 TARGET = ROOT / '.artifacts/reports/02-coverage-and-gaps.md'
 SOURCE_SLUG = 'ai-company-research/02-coverage-and-gaps'
+
 BASE = '/docs/other-research/ai-company-research/overview'
 
 
+
+def source_register():
+    sources = SOURCES['sources']
+    ids = {s['id'] for s in sources}
+    assert len(ids) == len(sources) and ids == {f'S{i:02}' for i in range(1, 17)}
+    assert SOURCES['scopeConfirmedAt'] <= DATA['asOf']
+    host_map = {}
+    for source in sources:
+        assert source['status'] in {'candidate-audit', 'targeted', 'pending', 'context', 'access-limited', 'per-company'}
+        assert source['boundary'] and source['nextCheck']
+        if source['url']:
+            assert urlsplit(source['url']).scheme in {'http', 'https'} and urlsplit(source['url']).netloc
+        for host in source['hosts']:
+            assert host not in host_map, f'Duplicate source host: {host}'
+            host_map[host] = source['id']
+    assigned = {key: [] for key in ids}
+    for evidence in DATA['evidence']:
+        owner = host_map.get(urlsplit(evidence['url']).netloc)
+        if owner is None:
+            owner = 'S12' if any(word in evidence['kind'] for word in ('官方', '公司自述', '公司披露')) else 'S16'
+        assigned[owner].append(evidence)
+    assert sum(len(v) for v in assigned.values()) == len(DATA['evidence'])
+    statuses = {'candidate-audit': '候选池逐条已读', 'targeted': '定向使用，未穷尽', 'pending': '候选台账待建',
+                'context': '背景研究', 'access-limited': '访问受限', 'per-company': '随公司逐条核对'}
+    lines = ['', '## 来源清单与验收边界', '', SOURCES['provenance'], '',
+        '每条证据在下表只计入一个渠道：先按所列域名归属，再将类型明确标为官方/公司自述/公司披露的剩余项计入S12，其余计入S16。公司可跨渠道出现，所以公司数不可跨行相加。零条证据表示尚未进入研究表，不表示该来源没有相关公司。背景报告不承担公司发现全集的角色。', '',
+        '| 来源 | 当前状态 | 证据 / 公司 / 观测数 | 已核对范围 | 剩余核查 |', '| --- | --- | --- | --- | --- |']
+    for source in sources:
+        evidence = assigned[source['id']]
+        evidence_ids = {e['id'] for e in evidence}
+        company_ids = {e['companyId'] for e in evidence if e['companyId']}
+        observations = sum(o['evidence'] in evidence_ids for o in DATA['observations'])
+        name = f"[{source['name']}]({source['url']})" if source['url'] else source['name']
+        lines.append(f"| {source['id']} {name} | {statuses[source['status']]} | {len(evidence)} / {len(company_ids)} / {observations} | {source['boundary']} | {source['nextCheck']} |")
+    lines += ['', 'Dealroom、Product Hunt、机器之心与36氪入口于2026-10-30重新访问。官网首页可读不证明已取得完整筛选结果；安全检测页面不作为文章内容。候选池之外的来源边界尚需建立可回查条目，完成前不宣布来源清单已全部覆盖。', '']
+    return lines
+
+
+
+def candidate_register():
+    batches = DATA.get('candidateBatches', [])
+    companies = {c['id']: c for c in DATA['companies']}
+    categories = {c['id']: c['name'] for c in DATA['categories']}
+    known_sources = {s['id'] for s in SOURCES['sources']}
+    evidence = {e['id']: e for e in DATA['evidence']}
+    assert len({b['id'] for b in batches}) == len(batches)
+    all_ids = [r['id'] for b in batches for r in b['records']]
+    assert len(set(all_ids)) == len(all_ids)
+    lines = []
+    statuses = {'pending-year': '成立年待核', 'linked': '已关联公司',
+                'existing-outside-cohort': '已有关联，年份范围外',
+                'outside-cohort': '成立年范围外，不新增公司', 'promotion': '推广活动，非公司新增'}
+    for batch in batches:
+        assert batch['sourceId'] in known_sources
+        assert batch['accessedAt'] <= DATA['asOf']
+        assert any(e['url'] == batch['url'] for e in DATA['evidence'])
+        rows = batch['records']
+        counts = Counter(r['status'] for r in rows)
+        assert sum(counts.values()) == len(rows)
+        lines += ['', f"## {batch['sourceId']} 候选核对（{batch['id']}）", '',
+            f"[发现页面]({batch['url']})，读取于{batch['accessedAt']}。{batch['scope']}", '',
+            f"本批{len(rows)}条：{counts['linked']}条关联公司、{counts['existing-outside-cohort']}条关联范围外对照、{counts['outside-cohort']}条范围外不新增公司、{counts['pending-year']}条成立年待核、{counts['promotion']}条推广活动。完整来源覆盖标记：{'已完成' if batch['complete'] else '未完成'}。主营或产品任务的初分类不计入主公司表数量，只有已关联的公司记录进入对应公司明细。候选与缺项随研究版本保存在文档站同一SQLite的研究元数据中。", '',
+            '| 发现条目 | 主营或产品任务初分 | 核对状态 | 关联公司 | 核对证据 | 限制 |', '| --- | --- | --- | --- | --- | --- |']
+        for row in rows:
+            assert row['categoryId'] in categories
+            assert row['status'] in statuses
+            assert all(eid in evidence for eid in row.get('evidenceIds', []))
+            if row['status'] == 'outside-cohort':
+                assert row.get('evidenceIds') and not row.get('companyId')
+            name = f"[{row['name']}]({row['url']})" if row['url'] else row['name']
+            company = '—'
+            if row.get('companyId'):
+                assert row['companyId'] in companies
+                c = companies[row['companyId']]
+                company = f"[{c['name']}]({BASE}?company={c['id']}#research-company-detail)"
+            proof = '、'.join(f"[{eid}]({evidence[eid]['url']})" for eid in row.get('evidenceIds', [])) or '成立证据待补，产品页读取状态见候选记录'
+            category = f"[{categories[row['categoryId']]}]({BASE}?category={row['categoryId']})"
+            lines.append(f"| {name} | {category} | {statuses[row['status']]} | {company} | {proof} | {row['note']} |")
+        if batch['sourceId'] == 'S04':
+            lines += ['', '产品页可读只表示有可访问的资料；本批未将所有产品页完成公司、技术或财务深查。既有直接HTTP采集返回403，当前采用可读的公开页面响应，不声称取得底层全量列表；后续公开档案与未读页面继续保留缺项。', '']
+        else:
+            lines += ['', '发现条目和可读正文分开记录；登录外壳、读取失败或搜索摘录不充当已阅读全文。关联公司以列出的原始材料补证，仍保留技术、主体与收入未核事项。', '']
+    return lines
+
+
+def directory_rechecks():
+    rows = DATA.get('directoryRechecks', [])
+    if not rows:
+        return []
+    original = {row['ycId']: row for row in AUDIT['records']}
+    companies = {c['id']: c for c in DATA['companies']}
+    evidence = {e['id']: e for e in DATA['evidence']}
+    assert len({row['ycId'] for row in rows}) == len(rows)
+    lines = ['', '## 原候选池的后续复核', '',
+        '原始1,428条读取快照与初筛计数保持不变；下表记录当时未入表、后续补足证据后进入公司表的项目。初筛未通过不是永久排除。后续入表不回写为当时已匹配，也不重复计入928条原始新增目录项。', '',
+        '| 原候选 | 原始状态 | 后续公司记录 | 复核证据 | 处理与限制 |', '| --- | --- | --- | --- | --- |']
+    for row in rows:
+        baseline = original[row['ycId']]
+        assert baseline['status'] == row['originalStatus']
+        assert row['reviewedAt'] <= DATA['asOf'] and row['reason']
+        assert row['originalStatus'] in {'ai-unclear', 'year-outside-or-missing', 'date-conflict'}
+        company = companies[row['companyId']]
+        assert row['evidenceIds'] and all(eid in evidence for eid in row['evidenceIds'])
+        proof = '、'.join(f"[{eid}]({evidence[eid]['url']})" for eid in row['evidenceIds'])
+        lines.append(f"| [{baseline['name']} · YC{row['ycId']}]({baseline['source']}) | {row['originalStatus']} | [{company['name']}]({BASE}?company={company['id']}#research-company-detail) | {proof} | {row['reviewedAt']}：{row['reason']} |")
+    lines += ['', f"已记录{len(rows)}条后续复核入表；其余未入表候选仍按原状态保留缺项，不能把这一批复核视为完成全部模糊项。", '']
+    return lines
+
+
 def render():
     cov = DATA['coverage']
     companies = DATA['companies']
@@ -86,6 +197,9 @@ def render():
         lines.append(f"| {name} | {len(entries)} | {len({e['url'] for e in entries})} | {note} |")
     others = [e for e in DATA['evidence'] if urlsplit(e['url']).netloc not in included]
     lines.append(f"| 其他官网、评测及公开资料 | {len(others)} | {len({e['url'] for e in others})} | 按每条来源限制理解，不能统一当作独立证据 |")
+    lines += directory_rechecks()
+    lines += source_register()
+    lines += candidate_register()
     lines += ['', '## 尚未满足的研究条件', '',
         '行业ARR比较仍缺统一业务截止日、币种、经常性定义和业务边界。现有平台MRR可描述已采集产品，不足以代表全部新公司。技术评测只覆盖所列版本与任务，不宜跨领域汇总为统一强弱分数。', '',
         '下一轮优先补收入线索的日期及合同定义、非YC和非英文来源、规则初分的人工复核，以及停止服务或被收购案例。这样能减少“只见赢家”和“只见融资公司”的偏差。公开来源池之外的未披露公司仍无法证明已被穷尽。', '',

scripts/export-legacy-documents.py

@@ -7,14 +7,15 @@ import re
 import subprocess
 
 ROOT = Path(__file__).resolve().parents[1]
-BASELINE = '25c5e83d08eae4277fe429a692955bd3bb2e717f'  # Last main revision before database-backed documents.
+BASELINE = 'cffca6942b1bc0005321e1685f38593fb0b853d5'  # Integration snapshot before database migration.
+RESEARCH_REVISION = '74af205b2959e1bb6a1fe9f567bc87069ac6d12c'  # Final PR 14 update received during integration.
 
 
 def export(ref):
     commit = subprocess.check_output(['git', 'rev-parse', '--verify', ref + '^{commit}'], cwd=ROOT, text=True).strip()
 
-    def read(name):
+    def read(name, revision=commit):
-        return subprocess.check_output(['git', 'show', f'{commit}:{name}'], cwd=ROOT).decode()
+        return subprocess.check_output(['git', 'show', f'{revision}:{name}'], cwd=ROOT).decode()
 
     catalog = json.loads(read('content/catalog.json'))
     names = subprocess.check_output(['git', 'ls-tree', '-r', '--name-only', commit, '--', 'content/docs'], cwd=ROOT, text=True).splitlines()
@@ -23,7 +24,8 @@ def export(ref):
         source = name.removeprefix('content/docs/').removesuffix('.md')
         metadata = catalog['documents'][source]
         collection = next(c for c in catalog['collections'] if not c['path'] or source.startswith(c['path'] + '/'))
-        content = read(name)
+        revision = RESEARCH_REVISION if commit == BASELINE and source.startswith("ai-company-research/") else commit
+        content = read(name, revision)
         title = re.search(r'^# (.+)$', content, re.M)
         documents.append(dict(slug=metadata['slug'], sourceSlug=source, category=metadata['category'],
             title=title[1].strip() if title else source, content=content,