const RELEASE_UUID = '[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}'; const RELEASE_ID = 'release[_ -]?id'; const CANDIDATE = '候选[ \\t]+\\d{8}-\\d{6}(?:[ \\t]*[,,][ \\t]*ACTIVE)?'; const STORY_SLUG = 'auto-quality-\\d{4}-\\d{2}-[a-z0-9-]+'; type TextRange = { start: number; end: number }; /** Only match data-release locking language; product/process versions are business facts. */ const INTERNAL_DATA_LOCK_PATTERNS = [ new RegExp( `(?:本轮|本次)(?:质量数据|查询数据|数据)(?:口径)?\\s*(?:将\\s*)?(?:固定使用|锁定使用|固定|锁定)(?:\\s*同一)?\\s*(?:发布|数据)?版本\\s*(?:[::=]\\s*)?(?:\\x60?${RELEASE_ID}\\s*[:=:]?\\s*)?(?:\\x60?${RELEASE_UUID}\\x60?)?(?:\\s*[,,;;]\\s*该版本(?:下)?[^。!?\\n]*)?`, 'gi', ), /(?:本轮|本次|当前)?(?:质量数据|数据|发布)版本\s*(?:已\s*)?(?:固定|锁定)(?:\s*[,,;;]\s*[^。!?\n]{0,30}?(?:不能|不可|不得)\s*(?:切换|更换)(?:发布|数据)?版本)?/gi, /(?:数据|发布|质量数据)(?:版本|口径)[^。!?\n]{0,30}(?:不能|不可|不得)\s*(?:切换|更换)(?:数据|发布)?版本/gi, ]; /** Metadata-only deletions. UUIDs and version numbers are never removed without these labels. */ const DISPLAY_METADATA_PATTERNS = [ // A release label and its UUID can be split across and text nodes in old HTML. new RegExp( `(?:^|[ \\t |,,;;])(?:\\*{0,2})(?:发布版本|数据版本|固定版本)(?:\\*{0,2})[ \\t]*[::=][ \\t]*(?:${RELEASE_ID}[ \\t]*[:=:]?[ \\t]*)?${RELEASE_UUID}(?:[ \\t]*[((][ \\t]*${CANDIDATE}[ \\t]*[))])?`, 'gim', ), new RegExp( `(?:^|[ \\t ,,;;])同一[ \\t]*(?:发布版本|数据版本)[ \\t]*[((][ \\t]*${RELEASE_ID}[ \\t]*[:=:]?[ \\t]*${RELEASE_UUID}(?:[ \\t]*[,,][ \\t]*${CANDIDATE})?[ \\t]*[))]`, 'gi', ), new RegExp(`(?:^|[ \\t ,,;;])${RELEASE_ID}[ \\t]*[:=:]?[ \\t]*${RELEASE_UUID}`, 'gi'), new RegExp(`[((][ \\t]*${CANDIDATE}[ \\t]*[))]|(?:^|[ \\t ,,;;])${CANDIDATE}`, 'gi'), new RegExp(`[((][ \\t]*\\x60?${STORY_SLUG}\\x60?(?:[ \\t]*[,,;;/][ \\t]*version[ \\t]*(?:id|no\\.?|number)?[ \\t]*[:=:]?[ \\t]*\\d+(?:\\.\\d+)*)?[ \\t]*[))]`, 'gi'), new RegExp(`(?:^|[ \\t ,,;;])Story[ \\t]*(?:ID|slug|version)(?:[ \\t]*[/,][ \\t]*(?:ID|slug|version))*[ \\t]*[:=:]?[ \\t]*[A-Za-z0-9][A-Za-z0-9._:/-]*`, 'gi'), new RegExp(`(?:^|[ \\t ,,;;])${STORY_SLUG}`, 'gi'), /数据目录与发布版本/gi, /[((]\s*(?:\*{2}\s*)?v\d+(?:\.\d+)*\s*复核版(?:\s*\*{2})?\s*[))]|(?:\*{2}\s*)?v\d+(?:\.\d+)*\s*复核版(?:\s*\*{2})?/gi, /^[ \t]*>[ \t]*v\d+(?:\.\d+)*\s*复核说明\s*[::]?[^\n]*$/gim, ]; function rangesForPattern(value: string, pattern: RegExp): TextRange[] { const ranges: TextRange[] = []; pattern.lastIndex = 0; let match: RegExpExecArray | null; while ((match = pattern.exec(value))) { if (match[0].length) ranges.push({ start: match.index, end: match.index + match[0].length }); else pattern.lastIndex += 1; } return ranges; } /** Ranges shared by the text and HTML renderers, so split inline tags cannot leak a label's value. */ export function analysisDisplayTextRemovalRanges(value: string): TextRange[] { const ranges = [ ...INTERNAL_DATA_LOCK_PATTERNS.flatMap(pattern => rangesForPattern(value, pattern)), ...DISPLAY_METADATA_PATTERNS.flatMap(pattern => rangesForPattern(value, pattern)), ].sort((left, right) => left.start - right.start || right.end - left.end); const merged: TextRange[] = []; for (const range of ranges) { const previous = merged.at(-1); if (previous && range.start <= previous.end) previous.end = Math.max(previous.end, range.end); else merged.push({ ...range }); } return merged; } function deleteRanges(value: string, ranges: TextRange[]): string { let cleaned = value; for (const range of [...ranges].reverse()) cleaned = cleaned.slice(0, range.start) + cleaned.slice(range.end); return cleaned; } /** Remove internal release/story metadata while preserving business versions, values, and evidence IDs. */ export function sanitizeAnalysisDisplayText(value: string): string { // Keep the useful status phrase while hiding the report's internal revision tag. const withoutRevisionPhrase = value.replace(/]*>[\s\S]*?(?:<\/tool_call>|$)/gi, '').replace( /报告已更新为\s*(?:\*{1,2})?v\d+(?:\.\d+)*\s*复核版(?:\*{1,2})?/gi, '报告已更新', ); let cleaned = deleteRanges(withoutRevisionPhrase, analysisDisplayTextRemovalRanges(withoutRevisionPhrase)); // Clean only punctuation made redundant by a removed metadata field. Do not trim leading colons: // HTML commonly stores a bold label and its business value in separate text nodes. return cleaned .replace(/数学/g, '') .replace(/[((]\s*[`*_~,,;;::/|\-–—\s]*[))]/g, '') .replace(/[,,;;::]{2,}/g, marks => marks.includes(':') || marks.includes(':') ? ':' : ';') .replace(/([。!?!?])\1+/g, '$1') .replace(/[ \t]{2,}/g, ' ') .replace(/[,,;;::][ \t]*[。!?!?]/g, match => match.trim().at(-1) || '') .replace(/\n{3,}/g, '\n\n') .replace(/^\n+|\n+$/g, ''); }