@gracker/smartperfetto 1.0.15 → 1.0.17

package/dist/agent/core/conclusionGenerator.js

@@ -1710,7 +1710,7 @@ function parseClaimRowSelector(value) {
         }
         else {
             const parsed = {};
-            for (const part of text.split(/[,，]/)) {
+            for (const part of text.split(/\s+(?:AND|and)\s+|[,，]/)) {
                 const match = part.trim().match(/^([^=：:]+)\s*(?:=|:|：)\s*(.+)$/);
                 if (!match)
                     continue;
@@ -1735,7 +1735,64 @@ function parseClaimRowSelector(value) {
     }
     return Object.keys(selector).length > 0 ? selector : undefined;
 }
-function parseClaimReferenceFromRecord(record) {
+function splitClaimList(value) {
+    if (value === undefined || value === null)
+        return [];
+    return String(value)
+        .split(/[,，]/)
+        .map(item => item.trim())
+        .filter(Boolean);
+}
+function parseClaimRowIndices(value) {
+    if (value === undefined || value === null || value === '')
+        return [];
+    if (typeof value === 'number' && Number.isInteger(value) && value >= 0)
+        return [value];
+    const text = String(value).trim();
+    const range = text.match(/^(\d+)\s*(?:-|\.\.|~|至|到)\s*(\d+)$/);
+    if (range) {
+        const start = Number(range[1]);
+        const end = Number(range[2]);
+        if (Number.isInteger(start) && Number.isInteger(end) && start >= 0 && end >= start && end - start <= 50) {
+            return Array.from({ length: end - start + 1 }, (_, index) => start + index);
+        }
+    }
+    const parsed = parseNumberFromUnknown(value);
+    return parsed !== undefined && Number.isInteger(parsed) && parsed >= 0 ? [parsed] : [];
+}
+function parseClaimTupleValues(rawValue, rowCount, columnCount) {
+    if (rawValue === undefined)
+        return [];
+    if (rowCount > 1) {
+        const rowGroups = splitClaimList(rawValue);
+        if (rowGroups.length === rowCount) {
+            return rowGroups.map(group => {
+                const slashParts = String(group)
+                    .split('/')
+                    .map(item => item.trim())
+                    .filter(Boolean);
+                if (columnCount > 1 && slashParts.length === columnCount) {
+                    return slashParts.map(parseClaimScalar);
+                }
+                return [parseClaimScalar(group)];
+            });
+        }
+    }
+    if (columnCount > 1) {
+        const values = splitClaimList(rawValue);
+        if (values.length >= columnCount) {
+            const normalizedValues = values.length === columnCount
+                ? values
+                : [
+                    ...values.slice(0, columnCount - 1),
+                    values.slice(columnCount - 1).join(', '),
+                ];
+            return [normalizedValues.map(parseClaimScalar)];
+        }
+    }
+    return [[parseClaimScalar(rawValue)]];
+}
+function parseClaimReferencesFromRecord(record) {
     const evidenceRefId = String(readValueFromAliases(record, [
         'evidenceRefId', 'evidence_ref_id', 'evidenceId', 'evidence_id',
     ]) || '').trim();
@@ -1746,21 +1803,97 @@ function parseClaimReferenceFromRecord(record) {
         'sourceToolCallId', 'source_tool_call_id', 'toolCallId', 'tool_call_id',
     ]) || '').trim();
     const rowIndexRaw = readValueFromAliases(record, ['rowIndex', 'row_index']);
-    const rowIndex = parseNumberFromUnknown(rowIndexRaw);
+    const rowIndices = parseClaimRowIndices(rowIndexRaw);
     const rowSelector = parseClaimRowSelector(readValueFromAliases(record, ['rowSelector', 'row_selector']));
-    const column = String(readValueFromAliases(record, ['column', 'col']) || '').trim();
-    const value = parseClaimScalar(readValueFromAliases(record, ['value']));
-    if (!evidenceRefId && !sourceRef && !sourceToolCallId)
-        return null;
-    return {
-        evidenceRefId,
-        ...(rowIndex !== undefined ? { rowIndex } : {}),
-        ...(rowSelector ? { rowSelector } : {}),
-        ...(column ? { column } : {}),
-        ...(value !== undefined ? { value } : {}),
+    const columnRaw = readValueFromAliases(record, ['column', 'col']);
+    const columns = splitClaimList(columnRaw);
+    const rawValue = readValueFromAliases(record, ['value']);
+    const artifactId = String(readValueFromAliases(record, ['artifactId', 'artifact_id']) || '').trim();
+    const sourceArtifactId = String(readValueFromAliases(record, ['sourceArtifactId', 'source_artifact_id']) || '').trim();
+    if (!evidenceRefId && !sourceRef && !sourceToolCallId && !artifactId && !sourceArtifactId)
+        return [];
+    const base = {
+        ...(evidenceRefId ? { evidenceRefId } : {}),
         ...(sourceRef ? { sourceRef } : {}),
         ...(sourceToolCallId ? { sourceToolCallId } : {}),
+        ...(artifactId ? { artifactId } : {}),
+        ...(sourceArtifactId ? { sourceArtifactId } : {}),
     };
+    const rowTargets = rowIndices.length > 0 ? rowIndices : [undefined];
+    const columnTargets = columns.length > 0 ? columns : [undefined];
+    const tupleValues = parseClaimTupleValues(rawValue, rowTargets.length, columnTargets.length);
+    const refs = [];
+    rowTargets.forEach((rowIndex, rowOffset) => {
+        columnTargets.forEach((column, columnOffset) => {
+            const value = tupleValues[rowOffset]?.[columnOffset]
+                ?? (rowTargets.length === 1 ? tupleValues[0]?.[columnOffset] : undefined);
+            refs.push({
+                ...base,
+                ...(rowIndex !== undefined ? { rowIndex } : {}),
+                ...(rowSelector ? { rowSelector } : {}),
+                ...(column ? { column } : {}),
+                ...(value !== undefined ? { value: value } : {}),
+            });
+        });
+    });
+    return refs;
+}
+function parseClaimKind(value) {
+    const normalized = String(value || '').trim();
+    const allowed = [
+        'numeric',
+        'categorical',
+        'time_range',
+        'identity',
+        'causal',
+        'comparison',
+        'inference',
+        'recommendation',
+    ];
+    return allowed.includes(normalized)
+        ? normalized
+        : undefined;
+}
+function parseClaimSupportLevel(value) {
+    const normalized = String(value || '').trim();
+    const allowed = ['verified', 'partial', 'inference', 'unsupported'];
+    return allowed.includes(normalized)
+        ? normalized
+        : undefined;
+}
+function parseStringArray(value) {
+    if (!Array.isArray(value))
+        return undefined;
+    const out = value
+        .map(item => String(item || '').trim())
+        .filter(Boolean);
+    return out.length > 0 ? Array.from(new Set(out)) : undefined;
+}
+function parseClaimArtifactRefs(value) {
+    if (!Array.isArray(value))
+        return undefined;
+    const refs = value
+        .map(item => toRecord(item))
+        .filter((item) => Boolean(item))
+        .map(item => {
+        const artifactId = String(readValueFromAliases(item, [
+            'artifactId',
+            'artifact_id',
+            'sourceArtifactId',
+            'source_artifact_id',
+        ]) || '').trim();
+        if (!artifactId)
+            return undefined;
+        const rowIndex = parseNumberFromUnknown(readValueFromAliases(item, ['rowIndex', 'row_index']));
+        const rowSelector = toRecord(readValueFromAliases(item, ['rowSelector', 'row_selector'])) || undefined;
+        return {
+            artifactId,
+            ...(rowIndex !== undefined ? { rowIndex } : {}),
+            ...(rowSelector ? { rowSelector } : {}),
+        };
+    })
+        .filter((item) => Boolean(item));
+    return refs.length > 0 ? refs : undefined;
 }
 function parseClaimItemsFromUnknown(value) {
     if (!Array.isArray(value))
@@ -1777,19 +1910,26 @@ function parseClaimItemsFromUnknown(value) {
             ? referencesSource
                 .map(ref => toRecord(ref))
                 .filter((ref) => Boolean(ref))
-                .map(ref => parseClaimReferenceFromRecord(ref))
-                .filter((ref) => Boolean(ref))
+                .flatMap(ref => parseClaimReferencesFromRecord(ref))
             : [];
-        if (references.length === 0)
+        const artifactRefs = parseClaimArtifactRefs(readValueFromAliases(record, ['artifactRefs', 'artifact_refs']));
+        if (references.length === 0 && (!artifactRefs || artifactRefs.length === 0))
             return;
         const claimId = String(readValueFromAliases(record, ['id', 'claimId', 'claim_id']) || '').trim();
         const conclusionId = normalizeConclusionId(String(readValueFromAliases(record, ['conclusionId', 'conclusion_id', 'conclusion']) || '').trim(), idx + 1);
         const text = String(readValueFromAliases(record, ['text', 'statement', 'claim']) || '').trim() || `claim ${idx + 1}`;
+        const kind = parseClaimKind(readValueFromAliases(record, ['kind', 'claimKind', 'claim_kind']));
+        const supportLevel = parseClaimSupportLevel(readValueFromAliases(record, ['supportLevel', 'support_level']));
+        const relationRefs = parseStringArray(readValueFromAliases(record, ['relationRefs', 'relation_refs']));
         claims.push({
             ...(claimId ? { id: claimId } : {}),
             conclusionId,
             text,
+            ...(kind ? { kind } : {}),
             references,
+            ...(artifactRefs ? { artifactRefs } : {}),
+            ...(relationRefs ? { relationRefs } : {}),
+            ...(supportLevel ? { supportLevel } : {}),
         });
     });
     return claims;
@@ -1802,6 +1942,10 @@ function formatClaimReferenceMarkdown(ref) {
         parts.push(`source_ref=${ref.sourceRef}`);
     if (ref.sourceToolCallId)
         parts.push(`source_tool_call_id=${ref.sourceToolCallId}`);
+    if (ref.artifactId)
+        parts.push(`artifact_id=${ref.artifactId}`);
+    if (ref.sourceArtifactId)
+        parts.push(`source_artifact_id=${ref.sourceArtifactId}`);
     if (typeof ref.rowIndex === 'number')
         parts.push(`row_index=${ref.rowIndex}`);
     if (ref.rowSelector)
@@ -1812,15 +1956,57 @@ function formatClaimReferenceMarkdown(ref) {
         parts.push(`value=${String(ref.value)}`);
     return parts.join('; ');
 }
-function parseClaimReferenceFromMarkdownLine(line) {
-    const record = {};
+function parseClaimReferencesFromMarkdownLine(line) {
+    const identifierKeys = new Set([
+        'evidence_ref_id',
+        'evidenceRefId',
+        'evidence_id',
+        'evidenceId',
+        'source_ref',
+        'sourceRef',
+        'ref',
+        'source_tool_call_id',
+        'sourceToolCallId',
+        'tool_call_id',
+        'toolCallId',
+        'artifact_id',
+        'artifactId',
+        'source_artifact_id',
+        'sourceArtifactId',
+    ]);
+    const localKeys = new Set(['row_index', 'rowIndex', 'row_selector', 'rowSelector', 'column', 'col', 'value']);
+    const base = {};
+    const groups = [];
+    let current = {};
+    const flush = () => {
+        if (Object.keys(current).length === 0)
+            return;
+        groups.push({ ...base, ...current });
+        current = {};
+    };
     for (const part of String(line || '').split(';')) {
         const match = part.trim().match(/^([a-zA-Z_]+)\s*=\s*(.*)$/);
         if (!match)
             continue;
-        record[match[1]] = match[2].trim();
+        const key = match[1];
+        const value = match[2].trim();
+        if (identifierKeys.has(key)) {
+            base[key] = value;
+            continue;
+        }
+        if (localKeys.has(key) &&
+            (Object.prototype.hasOwnProperty.call(current, key)
+                || ((key === 'row_index' || key === 'rowIndex' || key === 'row_selector' || key === 'rowSelector')
+                    && (current.column !== undefined || current.col !== undefined || current.value !== undefined))
+                || ((key === 'column' || key === 'col') && current.value !== undefined))) {
+            flush();
+        }
+        current[key] = value;
     }
-    return parseClaimReferenceFromRecord(record);
+    flush();
+    if (groups.length === 0 && Object.keys(base).length > 0)
+        groups.push({ ...base });
+    return groups.flatMap(group => parseClaimReferencesFromRecord(group));
 }
 function parseClaimItemsFromMarkdownSection(sectionBody) {
     const claims = [];
@@ -1848,8 +2034,8 @@ function parseClaimItemsFromMarkdownSection(sectionBody) {
             continue;
         }
         const refLine = line.replace(/^[-*]\s+/, '').trim();
-        const ref = parseClaimReferenceFromMarkdownLine(refLine);
-        if (ref) {
+        const refs = parseClaimReferencesFromMarkdownLine(refLine);
+        if (refs.length > 0) {
             if (!current) {
                 current = {
                     id: `Q${claims.length + 1}`,
@@ -1857,7 +2043,7 @@ function parseClaimItemsFromMarkdownSection(sectionBody) {
                     references: [],
                 };
             }
-            current.references.push(ref);
+            current.references.push(...refs);
         }
     }
     flush();
@@ -2148,24 +2334,34 @@ function sanitizeConclusionContract(contract, options) {
         .map((item, idx) => {
         const references = (item.references || [])
             .map(ref => ({
-            evidenceRefId: sanitizeText(ref.evidenceRefId),
+            ...(ref.evidenceRefId ? { evidenceRefId: sanitizeText(ref.evidenceRefId) } : {}),
             ...(typeof ref.rowIndex === 'number' && Number.isFinite(ref.rowIndex) ? { rowIndex: ref.rowIndex } : {}),
             ...(ref.rowSelector ? { rowSelector: parseClaimRowSelector(ref.rowSelector) } : {}),
             ...(ref.column ? { column: sanitizeText(ref.column) } : {}),
             ...(ref.value !== undefined ? { value: parseClaimScalar(ref.value) } : {}),
             ...(ref.sourceRef ? { sourceRef: sanitizeText(ref.sourceRef) } : {}),
             ...(ref.sourceToolCallId ? { sourceToolCallId: sanitizeText(ref.sourceToolCallId) } : {}),
+            ...(ref.artifactId ? { artifactId: sanitizeText(ref.artifactId) } : {}),
+            ...(ref.sourceArtifactId ? { sourceArtifactId: sanitizeText(ref.sourceArtifactId) } : {}),
         }))
-            .filter(ref => ((ref.evidenceRefId || ref.sourceRef || ref.sourceToolCallId) &&
+            .filter(ref => ((ref.evidenceRefId || ref.sourceRef || ref.sourceToolCallId || ref.artifactId || ref.sourceArtifactId) &&
             (ref.value === undefined || typeof ref.value === 'string' || typeof ref.value === 'number' || typeof ref.value === 'boolean')));
+        const kind = parseClaimKind(item.kind);
+        const artifactRefs = parseClaimArtifactRefs(item.artifactRefs);
+        const relationRefs = parseStringArray(item.relationRefs);
+        const supportLevel = parseClaimSupportLevel(item.supportLevel);
         return {
             ...(item.id ? { id: sanitizeText(item.id) } : {}),
             ...(item.conclusionId ? { conclusionId: normalizeConclusionId(item.conclusionId, idx + 1) } : {}),
             text: sanitizeText(item.text) || `claim ${idx + 1}`,
+            ...(kind ? { kind } : {}),
             references,
+            ...(artifactRefs ? { artifactRefs } : {}),
+            ...(relationRefs ? { relationRefs } : {}),
+            ...(supportLevel ? { supportLevel } : {}),
         };
     })
-        .filter(item => item.text && item.references.length > 0)
+        .filter(item => item.text && (item.references.length > 0 || Boolean(item.artifactRefs?.length)))
         .slice(0, 50);
     const uncertainties = dedupe(contract.uncertainties).slice(0, 6);
     const nextSteps = dedupe(contract.nextSteps).slice(0, 6);
@@ -3045,7 +3241,7 @@ function buildInsightFirstPrompt(params) {
     }
     parts.push('- evidence_chain 必须按 C1/C2/C3 对齐（C1=结论1）：每项包含 conclusion_id 和 evidence 文本，且包含至少 1 个 evidence id（ev_xxxxxxxxxxxx）。');
     parts.push('- claims 必须列出结论中出现的关键数值/实体判断，用于前端逐句核验；没有可核验数据时传空数组。');
-    parts.push('- claims 每项包含：id(Q1...), conclusion_id(C1...), text, references。');
+    parts.push('- claims 每项包含：id(Q1...), conclusion_id(C1...), text, kind, references；kind 从 numeric/categorical/time_range/identity/causal/comparison/inference/recommendation 中选择。');
     parts.push('- claims[].references 每项包含：evidence_ref_id（优先使用 data:* 证据 ID）, source_ref（如 表 1/摘要 1）, source_tool_call_id（如可见）, row_index（0-based，不是第几行的 1-based 编号）, row_selector（行号不稳定时使用）, column, value。');
     parts.push('- metadata 可包含 confidence（0-100 或 0-1）与 rounds（正整数）。');
     parts.push('- metadata 可包含 cluster_policy 对象：{ output_mode: required|optional|none, frame_list_mode: none|top|full, max_frames_per_cluster?: number }。');