fix(yuanbao): track new hunyuan_t1 SSE protocol and align inline citations with DeepSeek

- findInteractiveByText now also accepts elements carrying business attributes [dt-button-id]/[data-button-id]. The new yuanbao toolbar renders the deep-think toggle as <div dt-button-id="deep_think"> rather than a <button>, so the previous text-only locator missed it and probes silently ran in the fast model — without inline citations. - Lower <div data-idx-list="1,2,10"> placeholders to bare "[1][2][10]" (previously "[citation:1] [citation:2] [citation:10]") and accept both forms in YUANBAO_CITATION_MARKER_PATTERN. Final answer text runs through normalizeCitationMarkers so any residual SSE [citation:N] is rewritten to [N], matching DeepSeek's chip rendering on the SaaS side. - searchGuid frames in the new protocol carry every reference inside docs[] (each with a 1-based index that mirrors the answer markers) and citations is now always null. Promote docs into the citations bucket so citations[index-1] aligns with the inline marker, eliminating the spurious yuanbao_incomplete_citations failures. - Update the SaaS detail view: createYuanbaoCitationPattern accepts both "[N]" and "[citation:N]" so yuanbao answers render the same clickable superscript chips as DeepSeek. Tests cover the new searchGuid.docs alignment, fast-cache answers (no markers, no sources -> still succeed), legacy [citation:N] -> [N] rewriting, and mixed-format marker extraction. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-08 21:51:24 +08:00
parent af35301d9b
commit 1e6ed77453
3 changed files with 173 additions and 6 deletions
@@ -10,6 +10,7 @@ const {
  extractQuestionText,
  findUnresolvedCitationIndexes,
  mergeText,
+  normalizeCitationMarkers,
  normalizeSourceUrl,
  parseYuanbaoCaptures,
  resolveCitationsFromMarkers,
@@ -129,6 +130,124 @@ describe('yuanbao adapter helpers', () => {
    expect(findUnresolvedCitationIndexes('正文没有引用小图标', [])).toEqual([])
  })

+  it("recognises both '[N]' and '[citation:N]' inline markers", () => {
+    // The new yuanbao model lowers <div data-idx-list="1,2,10"> placeholders to "[1][2][10]";
+    // older SSE frames may still emit "[citation:N]". Both must extract the same indexes.
+    expect(extractCitationMarkerIndexes('合肥老牌选 [1][2] 性价比 [10]。')).toEqual([1, 2, 10])
+    expect(
+      extractCitationMarkerIndexes('混合两种格式 [citation:3] 还有 [4]'),
+    ).toEqual([3, 4])
+
+    const sources = [
+      { url: 'https://example.com/1', normalized_url: 'https://example.com/1' },
+      { url: 'https://example.com/2', normalized_url: 'https://example.com/2' },
+    ]
+    expect(
+      resolveCitationsFromMarkers('正文 [1] 然后 [2]', [[], sources]).map((item) => item.url),
+    ).toEqual(['https://example.com/1', 'https://example.com/2'])
+  })
+
+  it('rewrites legacy [citation:N] markers to bare [N] for SaaS rendering', () => {
+    expect(normalizeCitationMarkers('正文 [citation:1] 和 [citation:10]')).toBe(
+      '正文 [1] 和 [10]',
+    )
+    // Already-normalised text passes through.
+    expect(normalizeCitationMarkers('已经是 [1][2] 格式')).toBe('已经是 [1][2] 格式')
+    // Null/empty pass through.
+    expect(normalizeCitationMarkers(null)).toBeNull()
+    expect(normalizeCitationMarkers('')).toBe('')
+  })
+
+  it("does not invent citations for fast-cache answers (no searchGuid frame, no [N] markers)", () => {
+    // hunyuan_t1 sometimes returns a cached / non-search answer ("已深度思考(用时1秒)"
+    // with no "找到了 N 篇相关资料" panel). The SSE then carries only `text`
+    // frames and no `searchGuid`. The adapter must still surface that answer to
+    // the SaaS pipeline — see the top-level query path: as long as `answer` is
+    // non-empty and there are no [N] markers, neither the empty-response nor
+    // the incomplete-citations branches fire, so we fall through to status=succeeded.
+    const sseFrames = [
+      JSON.stringify({ type: 'text', msg: '在合肥，' }),
+      JSON.stringify({ type: 'text', msg: '全屋定制市场非常成熟，' }),
+      JSON.stringify({ type: 'text', msg: '没有绝对"最好"的商家。' }),
+    ]
+      .map((line) => `data: ${line}`)
+      .join('\n\n')
+
+    const summary = parseYuanbaoCaptures([
+      {
+        url: 'https://yuanbao.tencent.com/api/chat/cache-hit',
+        status: 200,
+        contentType: 'text/event-stream',
+        body: sseFrames,
+      },
+    ])
+
+    expect(summary.answer).toBe('在合肥，全屋定制市场非常成熟，没有绝对"最好"的商家。')
+    expect(summary.citations).toEqual([])
+    expect(summary.searchResults).toEqual([])
+    // No marker → no unresolved indexes, so the adapter does not gate on citations.
+    expect(extractCitationMarkerIndexes(summary.answer)).toEqual([])
+    expect(findUnresolvedCitationIndexes(summary.answer, summary.citations)).toEqual([])
+  })
+
+  it("aligns citations[index-1] with answer markers when SSE places refs in searchGuid.docs (new hunyuan_t1 protocol)", () => {
+    // Captured against the live yuanbao endpoint: deep-search now ships every
+    // reference inside `searchGuid.docs[]` with a 1-based `index`, and the
+    // legacy `citations` field is always null in this frame. Whatever the
+    // adapter normalises to must preserve idx alignment so that the
+    // unresolved-citations check passes.
+    const sseFrames = [
+      JSON.stringify({
+        type: 'searchGuid',
+        title: '引用 3 篇资料作为参考',
+        docs: [
+          {
+            index: 1,
+            title: '志邦家居官网',
+            url: 'https://www.zbom.com/about',
+            web_url: 'https://www.zbom.com/about',
+            web_site_name: '志邦家居',
+          },
+          {
+            index: 2,
+            title: '合肥本地装修评测',
+            url: 'https://example.com/hefei-review',
+            web_site_name: '今日头条',
+          },
+          {
+            index: 3,
+            title: '客来福官方介绍',
+            url: 'https://example.com/kelaifu',
+            web_site_name: '客来福',
+          },
+        ],
+        citations: null,
+      }),
+      JSON.stringify({
+        type: 'text',
+        msg: '志邦家居 [1] 与客来福 [3] 都是合肥本土头部 [2]。',
+      }),
+    ]
+      .map((line) => `data: ${line}`)
+      .join('\n\n')
+
+    const summary = parseYuanbaoCaptures([
+      {
+        url: 'https://yuanbao.tencent.com/api/chat/abc',
+        status: 200,
+        contentType: 'text/event-stream',
+        body: sseFrames,
+      },
+    ])
+
+    expect(summary.citations.map((item) => item.url)).toEqual([
+      'https://www.zbom.com/about',
+      'https://example.com/hefei-review',
+      'https://example.com/kelaifu',
+    ])
+    expect(findUnresolvedCitationIndexes(summary.answer, summary.citations)).toEqual([])
+  })
+
  it('collects source candidates from deep search thinking fields', () => {
    const summary = parseYuanbaoCaptures([
      {