feat(citation): measure citation coverage and mark unsupported statements

src/main/services/citationCandidates.ts

@@ -1,4 +1,5 @@
 import type { CitationCandidate } from '../../shared/types/citation'
+import { splitSentences, type TextRange } from '../../shared/utils/sentenceSegmentation'
 import type { EvidenceLocator } from './retrieval'
 
 /**
@@ -16,6 +17,10 @@ import type { EvidenceLocator } from './retrieval'
  * spans that already exist* removes the guess entirely.
  *
  * The chunk stays the retrieval unit. Candidates are the citation unit.
+ *
+ * Sentence boundaries come from `shared/utils/sentenceSegmentation`, the same
+ * splitter #156 uses to check an answer sentence by sentence — the two must agree
+ * or a candidate and a coverage statement would be about different spans.
  */
 
 /**
@@ -32,115 +37,12 @@ export interface CitationCandidateSource {
   content?: string
 }
 
-/** 相对传入字符串的字符区间。 */
-export interface TextRange {
-  start: number
-  end: number
-}
-
 /** 一个 candidate 大致的最小字符数：短句会与相邻句合并，而不是各自成条。 */
 const MIN_CANDIDATE_CHARS = 80
 
 /** 一个 candidate 最多合并多少句。 */
 const MAX_CANDIDATE_SENTENCES = 3
 
-/**
- * 常见的缩写。它们后面的句点不是句末 —— 误判成句末会把句子切碎，而碎 candidate
- * 比略大的 candidate 更糟：模型拿不到能支撑 claim 的完整片段。
- */
-const ABBREVIATIONS = new Set([
-  'al',
-  'approx',
-  'cf',
-  'dr',
-  'etc',
-  'fig',
-  'inc',
-  'jr',
-  'ltd',
-  'mr',
-  'mrs',
-  'ms',
-  'prof',
-  'ref',
-  'sec',
-  'sr',
-  'st',
-  'vol',
-  'vs'
-])
-
-/** 形如 `U.S` / `e.g` / `i.e` 的点分隔缩写。 */
-const DOTTED_INITIALS = /^(?:[a-z]\.)+[a-z]?$/
-
-/** 可以紧跟句末标点而不影响断句的收尾符号。 */
-const CLOSERS = /["')\]}»”’】」』]/
-
-const isCjk = (char: string): boolean => /[\u3000-\u9fff\uff00-\uffef]/.test(char)
-
-/**
- * `text[i]` 是不是一个句末标点。
- *
- * 刻意保守：`3.14`、`U.S.`、`Fig. 3`、`et al.` 里面的点都不断句。宁可把两句话合成
- * 一个 candidate，也不要把一句话切成两半。
- */
-function isSentenceBoundary(text: string, i: number): boolean {
-  const char = text[i]
-
-  if (char === '。' || char === '！' || char === '？') return true
-  if (char === '!' || char === '?') return true
-  if (char !== '.') return false
-
-  const prev = text[i - 1]
-  const next = text[i + 1]
-
-  // 3.14 —— 数字之间的小数点
-  if (prev !== undefined && next !== undefined && /\d/.test(prev) && /\d/.test(next)) return false
-
-  // 点前面的 token
-  let start = i
-  while (start > 0 && /[A-Za-z.\u00c0-\u024f]/.test(text[start - 1])) start--
-  const token = text.slice(start, i).toLowerCase()
-  if (token.length === 1) return false // 缩写首字母："J. Smith"
-  if (DOTTED_INITIALS.test(token)) return false
-  if (ABBREVIATIONS.has(token)) return false
-
-  // 点后面粘连着一个普通字母（"word.Foo"）不是句末；CJK 与收尾符号可以。
-  if (next !== undefined && !/\s/.test(next) && !isCjk(next) && !CLOSERS.test(next)) return false
-
-  return true
-}
-
-/** 把 `[start, end)` 去空白后加入区间列表。 */
-function pushRange(ranges: TextRange[], text: string, start: number, end: number): void {
-  let from = start
-  let to = end
-  while (from < to && /\s/.test(text[from])) from++
-  while (to > from && /\s/.test(text[to - 1])) to--
-  if (to > from) ranges.push({ start: from, end: to })
-}
-
-/**
- * 按句子/换行切分文本，返回相对 `text` 的区间。
- *
- * 纯函数：只有偏移，不产出字符串，调用方自己切片，因此 `quote` 与被引用的原文
- * 一定逐字一致（reader 的高亮就靠这个）。
- */
-export function splitSentences(text: string): TextRange[] {
-  const ranges: TextRange[] = []
-  let start = 0
-
-  for (let i = 0; i < text.length; i++) {
-    const boundary = text[i] === '\n' || isSentenceBoundary(text, i)
-    if (!boundary) continue
-    pushRange(ranges, text, start, i + 1)
-    start = i + 1
-  }
-
-  pushRange(ranges, text, start, text.length)
-  return ranges
-}
-
 /**
  * 句子区间 → candidate 区间。
  *

src/main/services/citations.ts

@@ -115,7 +115,10 @@ export function buildRAGContext(searchResults: SearchResult[]): RAGContext {
 
 ${promptParts.join('\n\n---\n\n')}
 
-请基于以上片段回答用户的问题。引用某一个片段时，请在该句子后使用 [n] 标注片段编号（n 为上面的编号，例如 [1]）。如果这些片段不足以回答问题，请说明并尽力提供有帮助的回答。`
+请基于以上片段回答用户的问题：
+- 每一句来自片段的陈述，都请在该句后使用 [n] 标注片段编号（n 为上面的编号，例如 [1]）。
+- 如果某句话是你自己的推断、并不来自这些片段，请在该句后标注 [inference]，不要给它加 [n]。
+- 如果这些片段不足以回答问题，请说明并尽力提供有帮助的回答。`
 
   return { context, sources, citations, citationContexts }
 }

src/renderer/src/components/notebook/chat/AnswerCoverage.tsx

@@ -0,0 +1,30 @@
+import type { ReactElement } from 'react'
+import { useTranslation } from 'react-i18next'
+import type { ClaimSupportReport } from '../../../../../shared/utils/claimSupport'
+
+interface AnswerCoverageProps {
+  coverage: ClaimSupportReport
+}
+
+/**
+ * 这条回答的 citation coverage（#156）。
+ *
+ * 只报告确定的事情：哪些句子没有引用、哪些引用指向了不存在的来源、哪些句子被模型
+ * 明确标注为自己的推断。它**不**声称某条引用在语义上证明了这个句子 —— 那是 NLI，
+ * 不是一个规则能回答的问题，假装能回答只会给出读者会相信的错误结论。
+ *
+ * 没有任何未支撑陈述时不渲染：正常的回答不需要一句「一切正常」。
+ */
+export default function AnswerCoverage({ coverage }: AnswerCoverageProps): ReactElement | null {
+  const { t } = useTranslation('chat')
+
+  const notices: string[] = []
+  if (coverage.invalid > 0) notices.push(t('answerCoverageInvalid', { count: coverage.invalid }))
+  if (coverage.uncited > 0) notices.push(t('answerCoverageUncited', { count: coverage.uncited }))
+  if (coverage.inference > 0)
+    notices.push(t('answerCoverageInference', { count: coverage.inference }))
+
+  if (notices.length === 0) return null
+
+  return <p className="px-2 text-xs text-subtle-foreground">{notices.join(' ')}</p>
+}

src/renderer/src/components/notebook/chat/MessageItem.tsx

@@ -9,6 +9,7 @@ import { useTranslation } from 'react-i18next'
 import type { ChatMessage } from '../../../types/notebook'
 import ReasoningContent from './ReasoningContent'
 import AnswerSources from './AnswerSources'
+import AnswerCoverage from './AnswerCoverage'
 import CitationChip from './CitationChip'
 import remarkCitationMarkers, { parseCitationHref } from './citationMarkers'
 import { useItemStore } from '../../../store/itemStore'
@@ -18,6 +19,7 @@ import { useSourceAnchorNavigation } from '../../../hooks/useSourceAnchorNavigat
 import { useUIStore } from '../../../store/uiStore'
 import { parseRetrievalStatus, sourcesForDisplay } from '../../../../../shared/utils/answerSources'
 import { parseCitations, sourceDocumentExists } from '../../../../../shared/utils/citations'
+import { classifyClaimSupport } from '../../../../../shared/utils/claimSupport'
 import {
   citationToSourceAnchor,
   sourceAnchorsEqual
@@ -68,6 +70,11 @@ export default function MessageItem({ message }: MessageItemProps): ReactElement
   const citations = parseCitations(message.metadata)
   const citationByIndex = new Map(citations.map((citation) => [citation.index, citation]))
 
+  // Citation coverage (#156): which sentences carry a resolvable citation. Computed
+  // from the answer and the persisted citations with the same pure function the
+  // tests exercise, so "grounded" means one thing in the UI and in a test.
+  const coverage = message.content ? classifyClaimSupport(message.content, citations) : null
+
   // A chip is disabled only once the library has been read and the document is
   // provably absent. Before that (or after a failed load) the list is "unknown",
   // not "deleted", so the chip stays clickable and the reader falls back.
@@ -271,6 +278,9 @@ export default function MessageItem({ message }: MessageItemProps): ReactElement
             }
           />
         )}
+        {/* Citation coverage (#156). Stated only when something is not backed by a
+            source; a normal answer does not need an "all good" line. */}
+        {coverage && !isLive && <AnswerCoverage coverage={coverage} />}
         {/* Why the answer stops where it does, when the model did not decide to stop
             there itself. Beside the answer, never instead of it: the part that
             arrived is what the reader was left with. */}

src/renderer/src/components/notebook/chat/citationMarkers.ts

@@ -18,6 +18,12 @@ export const CITATION_HREF_PREFIX = '#citation-'
 /** 只认 1-3 位数字，避免把 `[2024]` 这种正文误判成引用。 */
 const MARKER_PATTERN = /\[(\d{1,3})\]/g
 
+/**
+ * 模型标注「这是自己的推断，不是来自来源」的标记（#156）。它不会变成链接，也不该
+ * 出现在正文里 —— 逐句的判定由 `classifyClaimSupport` 完成，正文只显示干净的句子。
+ */
+const INFERENCE_MARKER = /\[inference\]/gi
+
 /** 从 link 的 `href` 里取回标记编号；不是 citation 链接时返回 null。 */
 export function parseCitationHref(href: string | undefined): number | null {
   if (!href || !href.startsWith(CITATION_HREF_PREFIX)) return null
@@ -37,6 +43,11 @@ interface MarkdownNode {
 
 /** 文本节点按标记切开，标记本身变成 link。 */
 function splitText(value: string): MarkdownNode[] {
+  // The inference marker is a machine annotation, not prose: drop it before the
+  // answer is rendered. The sentence's status still reaches the reader through the
+  // coverage notice.
+  value = value.replace(INFERENCE_MARKER, '')
+
   const parts: MarkdownNode[] = []
   let lastIndex = 0
   MARKER_PATTERN.lastIndex = 0

src/renderer/src/locales/en-US/chat.json

@@ -16,6 +16,9 @@
   "citationUnavailable": "Source no longer available",
   "answerNotGrounded": "No passages from your sources were used in this answer.",
   "answerRetrievalFailed": "Searching your sources failed, so this answer is not based on them.",
+  "answerCoverageUncited": "{{count}} statement(s) here are not backed by a source.",
+  "answerCoverageInvalid": "{{count}} citation(s) point to a source that does not exist.",
+  "answerCoverageInference": "{{count}} statement(s) are the model's own inference.",
   "answerTruncated": "The answer hit the model's output length limit, so it is cut off here.",
   "answerBlocked": "The provider refused this answer on content policy grounds.",
   "answerStopped": "The answer was stopped; this is the part that had been written.",

src/renderer/src/locales/zh-CN/chat.json

@@ -16,6 +16,9 @@
   "citationUnavailable": "来源已删除",
   "answerNotGrounded": "这条回答没有使用你来源中的任何段落。",
   "answerRetrievalFailed": "检索来源失败，这条回答没有依据你的来源。",
+  "answerCoverageUncited": "有 {{count}} 句陈述没有标注来源。",
+  "answerCoverageInvalid": "有 {{count}} 处引用指向不存在的来源。",
+  "answerCoverageInference": "有 {{count}} 句是模型自己的推断。",
   "answerTruncated": "回答达到模型输出长度上限，因此在这里被截断。",
   "answerBlocked": "提供方以内容策略为由拒绝了这次回答。",
   "answerStopped": "回答已停止，上面是已生成的部分。",