oh-my-knowledge 0.0.1 → 0.19.0

package/dist/src/analysis/gap-analyzer.js

@@ -0,0 +1,439 @@
+/**
+ * Knowledge gap signal analyzer.
+ *
+ * Implements the v0.1 detection pipeline defined in
+ * docs/knowledge-gap-signal-spec.md:
+ *
+ *   gap_rate = (samples with ≥1 gap signal) / (successfully-executed samples)
+ *
+ * Four signal types are extracted per sample:
+ *   1. failed_search      — Grep/Read/Bash search that returned empty or failed
+ *   2. explicit_marker    — 【推断】【知识缺口】【未知】 markers in agent output
+ *   3. hedging            — hedging language in agent output (accepts FP in v0.1)
+ *   4. repeated_failure   — ≥3 consecutive failed searches of the same tool type
+ *
+ * Intentionally does NOT touch report rendering, trends, or CI — those are
+ * step-3 work. This module only computes the structured GapReport.
+ */
+import { classifyHedgingCandidates } from './hedging-classifier.js';
+// ---------- Pattern tables ----------
+const EXPLICIT_MARKER_PATTERNS = [
+    /【推断】/g,
+    /【知识缺口】/g,
+    /【未知】/g,
+    /\[inferred\]/gi,
+    /\[unknown\]/gi,
+    /\[knowledge\s*gap\]/gi,
+];
+// v0.1 hedging patterns. Known to have false positives (e.g. "可能是" in normal
+// speech). Spec Section 4.3 accepts this in v0.1 and plans LLM-assisted
+// classification for v0.2.
+const HEDGING_PATTERNS = [
+    /我不确定/g,
+    /没有足够信息/g,
+    /需要查证/g,
+    /无法确认/g,
+    /\bI['\s]?m not sure\b/gi,
+    /\binsufficient information\b/gi,
+    /\bneed to verify\b/gi,
+    /\bpresumably\b/gi,
+];
+/** Minimum consecutive failed-search count to trigger a repeated_failure signal. */
+const REPEATED_FAILURE_THRESHOLD = 3;
+/**
+ * v0.2 severity weights (spec §6).
+ *
+ * Strong (1.0): 硬证据类——agent 工具层真的撞墙了
+ *   - failed_search: Grep/Read 返回空或失败,确定性 miss
+ *   - repeated_failure: 同一类查询连续 ≥3 次失败,已不是偶然
+ *
+ * Weak (0.5): 自我陈述类——依赖模型配合,可能假阳
+ *   - explicit_marker: 依赖 agent 按约定打【推断】等标记,可能漏标
+ *   - hedging: 字符串匹配 "我不确定/可能是"等,假阳率已知较高(spec §2.8)
+ *
+ * Aggregation: 每个样本取其信号的最高权重作"样本严重度",平均到 weightedGapRate。
+ * weightedGapRate ≤ gapRate,差值反映"软信号占比"——读者据此判断结果可信度。
+ */
+export const SIGNAL_WEIGHTS = {
+    failed_search: 1.0,
+    repeated_failure: 1.0,
+    explicit_marker: 0.5,
+    hedging: 0.5,
+};
+// ---------- Signal 1: failed searches ----------
+/**
+ * Decide whether a tool call counts as a "failed search" signal.
+ *
+ * Covers:
+ *   - Read with success: false
+ *   - Grep with success: false
+ *   - Grep with success but empty/"No matches found" output (still a miss)
+ *   - Bash with grep/rg/find in the command and either failure or empty output
+ */
+export function isFailedSearchTool(tc) {
+    const output = typeof tc.output === 'string' ? tc.output : String(tc.output ?? '');
+    const emptyOutput = output.trim() === '' || /No matches found/i.test(output);
+    if (tc.tool === 'Read') {
+        return tc.success === false;
+    }
+    if (tc.tool === 'Grep') {
+        if (tc.success === false)
+            return true;
+        return emptyOutput;
+    }
+    if (tc.tool === 'Bash') {
+        const cmd = (tc.input?.command) || '';
+        if (!/\b(grep|rg|find)\b/.test(cmd))
+            return false;
+        if (tc.success === false)
+            return true;
+        return emptyOutput;
+    }
+    return false;
+}
+function formatToolEvidence(tc) {
+    const input = tc.input || {};
+    let pattern = '';
+    let path = '';
+    if (tc.tool === 'Grep') {
+        pattern = String(input.pattern ?? '');
+        path = String(input.path ?? '');
+    }
+    else if (tc.tool === 'Read') {
+        path = String(input.file_path ?? '');
+    }
+    else if (tc.tool === 'Bash') {
+        pattern = String(input.command ?? '').slice(0, 120);
+    }
+    const context = `${tc.tool}: ${pattern || path}`.slice(0, 160);
+    return { pattern, path, context };
+}
+/**
+ * Extract failed-search signals from a flat tool call sequence.
+ * Deduplicates consecutive duplicates (same tool + same pattern) so an agent
+ * that retries the exact same Grep 5 times counts as one signal, not five.
+ */
+export function extractFailedSearchSignals(toolCalls) {
+    const raw = [];
+    for (const tc of toolCalls) {
+        if (!isFailedSearchTool(tc))
+            continue;
+        const { pattern, path, context } = formatToolEvidence(tc);
+        raw.push({
+            sampleId: '',
+            type: 'failed_search',
+            context,
+            evidence: { tool: tc.tool, pattern, path, success: tc.success },
+            weight: SIGNAL_WEIGHTS.failed_search,
+        });
+    }
+    // Dedup consecutive identical signals
+    const deduped = [];
+    for (const s of raw) {
+        const prev = deduped[deduped.length - 1];
+        if (prev && prev.type === s.type && prev.context === s.context)
+            continue;
+        deduped.push(s);
+    }
+    return deduped;
+}
+// ---------- Signals 2 & 3: text-based (markers + hedging) ----------
+function collectAssistantText(vr) {
+    const parts = [];
+    if (vr.turns) {
+        for (const turn of vr.turns) {
+            if (turn.role === 'assistant' && turn.content)
+                parts.push(turn.content);
+        }
+    }
+    // Fallback: fullOutput if turns are missing (some executors only emit final text)
+    if (parts.length === 0 && vr.fullOutput)
+        parts.push(vr.fullOutput);
+    return parts.join('\n');
+}
+/**
+ * Find explicit "I'm inferring / I don't know" markers in agent output text.
+ * Each marker occurrence produces one signal; dedupe is at the sample level
+ * (handled by the aggregator, not here).
+ */
+export function extractMarkerSignals(text) {
+    const signals = [];
+    if (!text)
+        return signals;
+    for (const pat of EXPLICIT_MARKER_PATTERNS) {
+        let match;
+        // Reset lastIndex for the global regex to be safe on reuse
+        pat.lastIndex = 0;
+        while ((match = pat.exec(text)) !== null) {
+            const idx = match.index;
+            const snippet = text.slice(Math.max(0, idx - 20), Math.min(text.length, idx + match[0].length + 60));
+            signals.push({
+                sampleId: '',
+                type: 'explicit_marker',
+                context: snippet.replace(/\s+/g, ' ').trim().slice(0, 160),
+                evidence: { marker: match[0] },
+                weight: SIGNAL_WEIGHTS.explicit_marker,
+            });
+        }
+    }
+    return signals;
+}
+/**
+ * Match hedging-language phrases in agent output text.
+ * v0.1 emits at most one hedging signal per sample — the aggregator upgrades
+ * this from "per pattern" to "per sample", since multiple hedges in the same
+ * turn are usually coming from the same epistemic state.
+ */
+export function extractHedgingSignals(text) {
+    if (!text)
+        return [];
+    for (const pat of HEDGING_PATTERNS) {
+        pat.lastIndex = 0;
+        const match = pat.exec(text);
+        if (!match)
+            continue;
+        const idx = match.index;
+        const snippet = text.slice(Math.max(0, idx - 15), Math.min(text.length, idx + match[0].length + 50));
+        return [{
+                sampleId: '',
+                type: 'hedging',
+                context: snippet.replace(/\s+/g, ' ').trim().slice(0, 160),
+                evidence: { matched: match[0] },
+                weight: SIGNAL_WEIGHTS.hedging,
+            }];
+    }
+    return [];
+}
+// ---------- Signal 4: repeated failure pattern ----------
+/**
+ * Detect ≥N consecutive failed searches of the same tool type within a turn
+ * sequence. Each qualifying run emits exactly one signal anchored at the
+ * first failure in the run.
+ */
+export function extractRepeatedFailureSignals(turns) {
+    if (!turns || turns.length === 0)
+        return [];
+    const signals = [];
+    let run = null;
+    let emittedForThisRun = false;
+    for (let ti = 0; ti < turns.length; ti++) {
+        const turn = turns[ti];
+        if (!turn.toolCalls || turn.toolCalls.length === 0)
+            continue;
+        for (const tc of turn.toolCalls) {
+            if (isFailedSearchTool(tc)) {
+                if (run && run.tool === tc.tool) {
+                    run.count += 1;
+                }
+                else {
+                    run = { tool: tc.tool, count: 1, startTurn: ti };
+                    emittedForThisRun = false;
+                }
+                if (!emittedForThisRun && run.count >= REPEATED_FAILURE_THRESHOLD) {
+                    signals.push({
+                        sampleId: '',
+                        type: 'repeated_failure',
+                        turn: run.startTurn,
+                        context: `${run.tool}: 连续失败 ≥${REPEATED_FAILURE_THRESHOLD} 次`,
+                        evidence: { tool: run.tool, count: run.count, startTurn: run.startTurn },
+                        weight: SIGNAL_WEIGHTS.repeated_failure,
+                    });
+                    emittedForThisRun = true;
+                }
+            }
+            else {
+                run = null;
+                emittedForThisRun = false;
+            }
+        }
+    }
+    return signals;
+}
+// ---------- Per-sample aggregation ----------
+/**
+ * Run all four signal extractors on a single variant's result for one sample.
+ * Output signals carry the provided sampleId.
+ */
+export function extractGapSignalsFromSample(variantResult, sampleId) {
+    // 1. Failed searches — flatten all tool calls across turns + top-level fallback
+    const toolCalls = [];
+    if (variantResult.turns) {
+        for (const turn of variantResult.turns) {
+            if (turn.toolCalls)
+                toolCalls.push(...turn.toolCalls);
+        }
+    }
+    if (toolCalls.length === 0 && variantResult.toolCalls) {
+        toolCalls.push(...variantResult.toolCalls);
+    }
+    const failedSearchSignals = extractFailedSearchSignals(toolCalls);
+    // 2 & 3. Text-based signals
+    const assistantText = collectAssistantText(variantResult);
+    const markerSignals = extractMarkerSignals(assistantText);
+    const hedgingSignals = extractHedgingSignals(assistantText);
+    // 4. Repeated failure (turn-ordered, so we need turns not flat toolCalls)
+    const repeatedFailureSignals = extractRepeatedFailureSignals(variantResult.turns);
+    const all = [...failedSearchSignals, ...markerSignals, ...hedgingSignals, ...repeatedFailureSignals];
+    return all.map((s) => ({ ...s, sampleId }));
+}
+// ---------- Report-level aggregation ----------
+/**
+ * Compute the gap report for one variant across all samples in a report.
+ *
+ * Denominator rule (spec §5): samples with `ok: false` (the variant completely
+ * failed to execute) are EXCLUDED from both numerator and denominator.
+ * gap_rate is computed only on samples that actually produced a trace.
+ */
+export function computeGapReport(results, variant) {
+    const signals = [];
+    const sampleIdsWithGap = new Set();
+    // 每样本取信号中的最强权重,用于 weightedGapRate(v0.2 §6)。
+    // 无信号样本 sampleWeight 隐含为 0,自然不计入 weighted 和。
+    const sampleMaxWeight = new Map();
+    let sampleCount = 0;
+    for (const entry of results) {
+        const vr = entry.variants?.[variant];
+        if (!vr || vr.ok === false)
+            continue;
+        sampleCount += 1;
+        const sampleSignals = extractGapSignalsFromSample(vr, entry.sample_id);
+        if (sampleSignals.length > 0) {
+            sampleIdsWithGap.add(entry.sample_id);
+            signals.push(...sampleSignals);
+            const maxW = sampleSignals.reduce((m, s) => (s.weight > m ? s.weight : m), 0);
+            sampleMaxWeight.set(entry.sample_id, maxW);
+        }
+    }
+    const samplesWithGap = sampleIdsWithGap.size;
+    const gapRate = sampleCount > 0 ? Number((samplesWithGap / sampleCount).toFixed(4)) : 0;
+    const weightSum = Array.from(sampleMaxWeight.values()).reduce((a, b) => a + b, 0);
+    const weightedGapRate = sampleCount > 0 ? Number((weightSum / sampleCount).toFixed(4)) : 0;
+    const byType = {
+        failed_search: 0,
+        explicit_marker: 0,
+        hedging: 0,
+        repeated_failure: 0,
+    };
+    for (const s of signals)
+        byType[s.type] += 1;
+    return {
+        variant,
+        sampleCount,
+        samplesWithGap,
+        gapRate,
+        weightedGapRate,
+        testSetPath: null,
+        testSetHash: null,
+        signals,
+        byType,
+    };
+}
+/**
+ * Compute gap reports for every variant in a Report. Convenience wrapper.
+ */
+export function computeReportGapRates(results, variants) {
+    const out = {};
+    for (const v of variants) {
+        out[v] = computeGapReport(results, v);
+    }
+    return out;
+}
+// ---------- v0.2 hedging classifier integration ----------
+/**
+ * 重新计算 GapReport 的衍生字段(samplesWithGap / gapRate / weightedGapRate /
+ * byType),用于 hedging signal 被 classifier 过滤后的二次聚合。
+ * sampleCount 不变(分母:成功执行的样本数,与是否有信号无关)。
+ */
+function recomputeAggregates(report) {
+    const sampleIdsWithGap = new Set();
+    const sampleMaxWeight = new Map();
+    const byType = {
+        failed_search: 0,
+        explicit_marker: 0,
+        hedging: 0,
+        repeated_failure: 0,
+    };
+    for (const s of report.signals) {
+        sampleIdsWithGap.add(s.sampleId);
+        const cur = sampleMaxWeight.get(s.sampleId) ?? 0;
+        if (s.weight > cur)
+            sampleMaxWeight.set(s.sampleId, s.weight);
+        byType[s.type] += 1;
+    }
+    const samplesWithGap = sampleIdsWithGap.size;
+    const gapRate = report.sampleCount > 0
+        ? Number((samplesWithGap / report.sampleCount).toFixed(4))
+        : 0;
+    const weightSum = Array.from(sampleMaxWeight.values()).reduce((a, b) => a + b, 0);
+    const weightedGapRate = report.sampleCount > 0
+        ? Number((weightSum / report.sampleCount).toFixed(4))
+        : 0;
+    return { ...report, samplesWithGap, gapRate, weightedGapRate, byType };
+}
+/**
+ * 对一份 GapReport 做 hedging classifier 二次过滤(spec §四.3, v0.2)。
+ * 流程:
+ *   1) 收集 signals 中所有 type='hedging' 的 candidate
+ *   2) 调 classifyHedgingCandidates 拿 verdict
+ *   3) verdict.isUncertainty=false 的 signal 丢弃
+ *   4) verdict.isUncertainty=true 的保留并挂 classifierVerdict
+ *   5) recompute samplesWithGap / gapRate / weightedGapRate / byType
+ *
+ * 失败降级在 classifier 层:单批失败 → 该 batch 全部 isUncertainty=true (保守保留)。
+ * 调用方拿到的 report 永远是合法的,不会因 classifier 失败而崩。
+ */
+export async function applyHedgingClassifier(report, executor, opts) {
+    const hedgingIndices = [];
+    const candidates = [];
+    for (let i = 0; i < report.signals.length; i++) {
+        const s = report.signals[i];
+        if (s.type !== 'hedging')
+            continue;
+        hedgingIndices.push(i);
+        const matched = s.evidence?.matched ?? s.context;
+        candidates.push({
+            sampleId: s.sampleId,
+            sentence: matched,
+            context: s.context,
+        });
+    }
+    if (candidates.length === 0) {
+        return { report, costUSD: 0, truncated: false };
+    }
+    const { verdicts, costUSD, truncated } = await classifyHedgingCandidates(candidates, executor, opts);
+    // 按 verdict 重组 signals:非 hedging 原样保留,hedging 按 verdict.isUncertainty 过滤 + 挂 verdict
+    const keptHedging = new Set();
+    hedgingIndices.forEach((idx, i) => {
+        if (verdicts[i].isUncertainty)
+            keptHedging.add(idx);
+    });
+    const newSignals = [];
+    for (let i = 0; i < report.signals.length; i++) {
+        const s = report.signals[i];
+        if (s.type !== 'hedging') {
+            newSignals.push(s);
+            continue;
+        }
+        if (!keptHedging.has(i))
+            continue; // dropped by classifier
+        const verdictIdx = hedgingIndices.indexOf(i);
+        newSignals.push({ ...s, classifierVerdict: verdicts[verdictIdx] });
+    }
+    const next = recomputeAggregates({ ...report, signals: newSignals });
+    return { report: next, costUSD, truncated };
+}
+/**
+ * 批量为一组 variant 的 gap reports 跑 classifier。
+ * 串行(不并行)避免 rate limit + 让 cache hit 在第一批后被后续 batch 复用。
+ */
+export async function applyHedgingClassifierToReports(reports, executor, opts) {
+    const out = {};
+    let totalCost = 0;
+    for (const [variant, report] of Object.entries(reports)) {
+        const result = await applyHedgingClassifier(report, executor, opts);
+        out[variant] = result.report;
+        totalCost += result.costUSD;
+    }
+    return { reports: out, costUSD: totalCost };
+}
+//# sourceMappingURL=gap-analyzer.js.map

package/dist/src/analysis/gap-analyzer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gap-analyzer.js","sourceRoot":"","sources":["../../../src/analysis/gap-analyzer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAGH,OAAO,EAAE,yBAAyB,EAA+C,MAAM,yBAAyB,CAAC;AAKjH,uCAAuC;AAEvC,MAAM,wBAAwB,GAAa;IACzC,OAAO;IACP,SAAS;IACT,OAAO;IACP,gBAAgB;IAChB,eAAe;IACf,uBAAuB;CACxB,CAAC;AAEF,6EAA6E;AAC7E,wEAAwE;AACxE,2BAA2B;AAC3B,MAAM,gBAAgB,GAAa;IACjC,OAAO;IACP,SAAS;IACT,OAAO;IACP,OAAO;IACP,yBAAyB;IACzB,gCAAgC;IAChC,sBAAsB;IACtB,kBAAkB;CACnB,CAAC;AAEF,oFAAoF;AACpF,MAAM,0BAA0B,GAAG,CAAC,CAAC;AAErC;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,MAAM,cAAc,GAAkC;IAC3D,aAAa,EAAE,GAAG;IAClB,gBAAgB,EAAE,GAAG;IACrB,eAAe,EAAE,GAAG;IACpB,OAAO,EAAE,GAAG;CACb,CAAC;AAEF,kDAAkD;AAElD;;;;;;;;GAQG;AACH,MAAM,UAAU,kBAAkB,CAAC,EAAgB;IACjD,MAAM,MAAM,GAAG,OAAO,EAAE,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,EAAE,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC;IACnF,MAAM,WAAW,GAAG,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,IAAI,mBAAmB,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAE7E,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACvB,OAAO,EAAE,CAAC,OAAO,KAAK,KAAK,CAAC;IAC9B,CAAC;IAED,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACvB,IAAI,EAAE,CAAC,OAAO,KAAK,KAAK;YAAE,OAAO,IAAI,CAAC;QACtC,OAAO,WAAW,CAAC;IACrB,CAAC;IAED,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACvB,MAAM,GAAG,GAAG,CAAE,EAAE,CAAC,KAAqC,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC;QACvE,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,GAAG,CAAC;YAAE,OAAO,KAAK,CAAC;QAClD,IAAI,EAAE,CAAC,OAAO,KAAK,KAAK;YAAE,OAAO,IAAI,CAAC;QACtC,OAAO,WAAW,CAAC;IACrB,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,kBAAkB,CAAC,EAAgB;IAC1C,MAAM,KAAK,GAAI,EAAE,CAAC,KAAwC,IAAI,EAAE,CAAC;IACjE,IAAI,OAAO,GAAG,EAAE,CAAC;IACjB,IAAI,IAAI,GAAG,EAAE,CAAC;IACd,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACvB,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;QACtC,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC;IAClC,CAAC;SAAM,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QAC9B,IAAI,GAAG,MAAM,CAAC,KAAK,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC;IACvC,CAAC;SAAM,IAAI,EAAE,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QAC9B,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IACtD,CAAC;IACD,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC,IAAI,KAAK,OAAO,IAAI,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC/D,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;AACpC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,0BAA0B,CAAC,SAAyB;IAClE,MAAM,GAAG,GAAgB,EAAE,CAAC;IAC5B,KAAK,MAAM,EAAE,IAAI,SAAS,EAAE,CAAC;QAC3B,IAAI,CAAC,kBAAkB,CAAC,EAAE,CAAC;YAAE,SAAS;QACtC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,GAAG,kBAAkB,CAAC,EAAE,CAAC,CAAC;QAC1D,GAAG,CAAC,IAAI,CAAC;YACP,QAAQ,EAAE,EAAE;YACZ,IAAI,EAAE,eAAe;YACrB,OAAO;YACP,QAAQ,EAAE,EAAE,IAAI,EAAE,EAAE,CAAC,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,EAAE,CAAC,OAAO,EAAE;YAC/D,MAAM,EAAE,cAAc,CAAC,aAAa;SACrC,CAAC,CAAC;IACL,CAAC;IACD,sCAAsC;IACtC,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,KAAK,MAAM,CAAC,IAAI,GAAG,EAAE,CAAC;QACpB,MAAM,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;QACzC,IAAI,IAAI,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,CAAC,IAAI,IAAI,IAAI,CAAC,OAAO,KAAK,CAAC,CAAC,OAAO;YAAE,SAAS;QACzE,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,sEAAsE;AAEtE,SAAS,oBAAoB,CAAC,EAAiB;IAC7C,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,EAAE,CAAC,KAAK,EAAE,CAAC;QACb,KAAK,MAAM,IAAI,IAAI,EAAE,CAAC,KAAK,EAAE,CAAC;YAC5B,IAAI,IAAI,CAAC,IAAI,KAAK,WAAW,IAAI,IAAI,CAAC,OAAO;gBAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC1E,CAAC;IACH,CAAC;IACD,kFAAkF;IAClF,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,EAAE,CAAC,UAAU;QAAE,KAAK,CAAC,IAAI,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC;IACnE,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,oBAAoB,CAAC,IAAY;IAC/C,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,IAAI,CAAC,IAAI;QAAE,OAAO,OAAO,CAAC;IAC1B,KAAK,MAAM,GAAG,IAAI,wBAAwB,EAAE,CAAC;QAC3C,IAAI,KAA6B,CAAC;QAClC,2DAA2D;QAC3D,GAAG,CAAC,SAAS,GAAG,CAAC,CAAC;QAClB,OAAO,CAAC,KAAK,GAAG,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,KAAK,IAAI,EAAE,CAAC;YACzC,MAAM,GAAG,GAAG,KAAK,CAAC,KAAK,CAAC;YACxB,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,EAAE,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC;YACrG,OAAO,CAAC,IAAI,CAAC;gBACX,QAAQ,EAAE,EAAE;gBACZ,IAAI,EAAE,iBAAiB;gBACvB,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBAC1D,QAAQ,EAAE,EAAE,MAAM,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE;gBAC9B,MAAM,EAAE,cAAc,CAAC,eAAe;aACvC,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,qBAAqB,CAAC,IAAY;IAChD,IAAI,CAAC,IAAI;QAAE,OAAO,EAAE,CAAC;IACrB,KAAK,MAAM,GAAG,IAAI,gBAAgB,EAAE,CAAC;QACnC,GAAG,CAAC,SAAS,GAAG,CAAC,CAAC;QAClB,MAAM,KAAK,GAAG,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC7B,IAAI,CAAC,KAAK;YAAE,SAAS;QACrB,MAAM,GAAG,GAAG,KAAK,CAAC,KAAK,CAAC;QACxB,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,EAAE,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,EAAE,CAAC,CAAC,CAAC;QACrG,OAAO,CAAC;gBACN,QAAQ,EAAE,EAAE;gBACZ,IAAI,EAAE,SAAS;gBACf,OAAO,EAAE,OAAO,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC;gBAC1D,QAAQ,EAAE,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,EAAE;gBAC/B,MAAM,EAAE,cAAc,CAAC,OAAO;aAC/B,CAAC,CAAC;IACL,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,2DAA2D;AAE3D;;;;GAIG;AACH,MAAM,UAAU,6BAA6B,CAAC,KAA6B;IACzE,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAC5C,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,IAAI,GAAG,GAA8D,IAAI,CAAC;IAC1E,IAAI,iBAAiB,GAAG,KAAK,CAAC;IAE9B,KAAK,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,MAAM,EAAE,EAAE,EAAE,EAAE,CAAC;QACzC,MAAM,IAAI,GAAG,KAAK,CAAC,EAAE,CAAC,CAAC;QACvB,IAAI,CAAC,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC;YAAE,SAAS;QAE7D,KAAK,MAAM,EAAE,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YAChC,IAAI,kBAAkB,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC3B,IAAI,GAAG,IAAI,GAAG,CAAC,IAAI,KAAK,EAAE,CAAC,IAAI,EAAE,CAAC;oBAChC,GAAG,CAAC,KAAK,IAAI,CAAC,CAAC;gBACjB,CAAC;qBAAM,CAAC;oBACN,GAAG,GAAG,EAAE,IAAI,EAAE,EAAE,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,SAAS,EAAE,EAAE,EAAE,CAAC;oBACjD,iBAAiB,GAAG,KAAK,CAAC;gBAC5B,CAAC;gBACD,IAAI,CAAC,iBAAiB,IAAI,GAAG,CAAC,KAAK,IAAI,0BAA0B,EAAE,CAAC;oBAClE,OAAO,CAAC,IAAI,CAAC;wBACX,QAAQ,EAAE,EAAE;wBACZ,IAAI,EAAE,kBAAkB;wBACxB,IAAI,EAAE,GAAG,CAAC,SAAS;wBACnB,OAAO,EAAE,GAAG,GAAG,CAAC,IAAI,WAAW,0BAA0B,IAAI;wBAC7D,QAAQ,EAAE,EAAE,IAAI,EAAE,GAAG,CAAC,IAAI,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,EAAE,SAAS,EAAE,GAAG,CAAC,SAAS,EAAE;wBACxE,MAAM,EAAE,cAAc,CAAC,gBAAgB;qBACxC,CAAC,CAAC;oBACH,iBAAiB,GAAG,IAAI,CAAC;gBAC3B,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,GAAG,GAAG,IAAI,CAAC;gBACX,iBAAiB,GAAG,KAAK,CAAC;YAC5B,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,+CAA+C;AAE/C;;;GAGG;AACH,MAAM,UAAU,2BAA2B,CAAC,aAA4B,EAAE,QAAgB;IACxF,gFAAgF;IAChF,MAAM,SAAS,GAAmB,EAAE,CAAC;IACrC,IAAI,aAAa,CAAC,KAAK,EAAE,CAAC;QACxB,KAAK,MAAM,IAAI,IAAI,aAAa,CAAC,KAAK,EAAE,CAAC;YACvC,IAAI,IAAI,CAAC,SAAS;gBAAE,SAAS,CAAC,IAAI,CAAC,GAAG,IAAI,CAAC,SAAS,CAAC,CAAC;QACxD,CAAC;IACH,CAAC;IACD,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,aAAa,CAAC,SAAS,EAAE,CAAC;QACtD,SAAS,CAAC,IAAI,CAAC,GAAG,aAAa,CAAC,SAAS,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,mBAAmB,GAAG,0BAA0B,CAAC,SAAS,CAAC,CAAC;IAElE,4BAA4B;IAC5B,MAAM,aAAa,GAAG,oBAAoB,CAAC,aAAa,CAAC,CAAC;IAC1D,MAAM,aAAa,GAAG,oBAAoB,CAAC,aAAa,CAAC,CAAC;IAC1D,MAAM,cAAc,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE5D,0EAA0E;IAC1E,MAAM,sBAAsB,GAAG,6BAA6B,CAAC,aAAa,CAAC,KAAK,CAAC,CAAC;IAElF,MAAM,GAAG,GAAG,CAAC,GAAG,mBAAmB,EAAE,GAAG,aAAa,EAAE,GAAG,cAAc,EAAE,GAAG,sBAAsB,CAAC,CAAC;IACrG,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,QAAQ,EAAE,CAAC,CAAC,CAAC;AAC9C,CAAC;AAED,iDAAiD;AAEjD;;;;;;GAMG;AACH,MAAM,UAAU,gBAAgB,CAAC,OAAsB,EAAE,OAAe;IACtE,MAAM,OAAO,GAAgB,EAAE,CAAC;IAChC,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAU,CAAC;IAC3C,4CAA4C;IAC5C,6CAA6C;IAC7C,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkB,CAAC;IAClD,IAAI,WAAW,GAAG,CAAC,CAAC;IAEpB,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC5B,MAAM,EAAE,GAAG,KAAK,CAAC,QAAQ,EAAE,CAAC,OAAO,CAAC,CAAC;QACrC,IAAI,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,KAAK;YAAE,SAAS;QACrC,WAAW,IAAI,CAAC,CAAC;QAEjB,MAAM,aAAa,GAAG,2BAA2B,CAAC,EAAE,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC;QACvE,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC7B,gBAAgB,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;YACtC,OAAO,CAAC,IAAI,CAAC,GAAG,aAAa,CAAC,CAAC;YAC/B,MAAM,IAAI,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;YAC9E,eAAe,CAAC,GAAG,CAAC,KAAK,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,gBAAgB,CAAC,IAAI,CAAC;IAC7C,MAAM,OAAO,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,cAAc,GAAG,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAExF,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAClF,MAAM,eAAe,GAAG,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,SAAS,GAAG,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAE3F,MAAM,MAAM,GAAkC;QAC5C,aAAa,EAAE,CAAC;QAChB,eAAe,EAAE,CAAC;QAClB,OAAO,EAAE,CAAC;QACV,gBAAgB,EAAE,CAAC;KACpB,CAAC;IACF,KAAK,MAAM,CAAC,IAAI,OAAO;QAAE,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAE7C,OAAO;QACL,OAAO;QACP,WAAW;QACX,cAAc;QACd,OAAO;QACP,eAAe;QACf,WAAW,EAAE,IAAI;QACjB,WAAW,EAAE,IAAI;QACjB,OAAO;QACP,MAAM;KACP,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CAAC,OAAsB,EAAE,QAAkB;IAC9E,MAAM,GAAG,GAA8B,EAAE,CAAC;IAC1C,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACzB,GAAG,CAAC,CAAC,CAAC,GAAG,gBAAgB,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,4DAA4D;AAE5D;;;;GAIG;AACH,SAAS,mBAAmB,CAAC,MAAiB;IAC5C,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAU,CAAC;IAC3C,MAAM,eAAe,GAAG,IAAI,GAAG,EAAkB,CAAC;IAClD,MAAM,MAAM,GAAkC;QAC5C,aAAa,EAAE,CAAC;QAChB,eAAe,EAAE,CAAC;QAClB,OAAO,EAAE,CAAC;QACV,gBAAgB,EAAE,CAAC;KACpB,CAAC;IACF,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QAC/B,gBAAgB,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;QACjC,MAAM,GAAG,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACjD,IAAI,CAAC,CAAC,MAAM,GAAG,GAAG;YAAE,eAAe,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC;QAC9D,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACtB,CAAC;IACD,MAAM,cAAc,GAAG,gBAAgB,CAAC,IAAI,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,WAAW,GAAG,CAAC;QACpC,CAAC,CAAC,MAAM,CAAC,CAAC,cAAc,GAAG,MAAM,CAAC,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC1D,CAAC,CAAC,CAAC,CAAC;IACN,MAAM,SAAS,GAAG,KAAK,CAAC,IAAI,CAAC,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;IAClF,MAAM,eAAe,GAAG,MAAM,CAAC,WAAW,GAAG,CAAC;QAC5C,CAAC,CAAC,MAAM,CAAC,CAAC,SAAS,GAAG,MAAM,CAAC,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QACrD,CAAC,CAAC,CAAC,CAAC;IACN,OAAO,EAAE,GAAG,MAAM,EAAE,cAAc,EAAE,OAAO,EAAE,eAAe,EAAE,MAAM,EAAE,CAAC;AACzE,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,sBAAsB,CAC1C,MAAiB,EACjB,QAAoB,EACpB,IAAsB;IAEtB,MAAM,cAAc,GAAa,EAAE,CAAC;IACpC,MAAM,UAAU,GAAuB,EAAE,CAAC;IAC1C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC/C,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5B,IAAI,CAAC,CAAC,IAAI,KAAK,SAAS;YAAE,SAAS;QACnC,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACvB,MAAM,OAAO,GAAI,CAAC,CAAC,QAAQ,EAAE,OAA8B,IAAI,CAAC,CAAC,OAAO,CAAC;QACzE,UAAU,CAAC,IAAI,CAAC;YACd,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,QAAQ,EAAE,OAAO;YACjB,OAAO,EAAE,CAAC,CAAC,OAAO;SACnB,CAAC,CAAC;IACL,CAAC;IACD,IAAI,UAAU,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,CAAC,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;IAClD,CAAC;IAED,MAAM,EAAE,QAAQ,EAAE,OAAO,EAAE,SAAS,EAAE,GAAG,MAAM,yBAAyB,CAAC,UAAU,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;IAErG,qFAAqF;IACrF,MAAM,WAAW,GAAG,IAAI,GAAG,EAAU,CAAC;IACtC,cAAc,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE;QAChC,IAAI,QAAQ,CAAC,CAAC,CAAC,CAAC,aAAa;YAAE,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IACtD,CAAC,CAAC,CAAC;IACH,MAAM,UAAU,GAAmB,EAAE,CAAC;IACtC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAC/C,MAAM,CAAC,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC5B,IAAI,CAAC,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;YACzB,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACnB,SAAS;QACX,CAAC;QACD,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;YAAE,SAAS,CAAC,wBAAwB;QAC3D,MAAM,UAAU,GAAG,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;QAC7C,UAAU,CAAC,IAAI,CAAC,EAAE,GAAG,CAAC,EAAE,iBAAiB,EAAE,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC;IACrE,CAAC;IAED,MAAM,IAAI,GAAG,mBAAmB,CAAC,EAAE,GAAG,MAAM,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC,CAAC;IACrE,OAAO,EAAE,MAAM,EAAE,IAAI,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;AAC9C,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,+BAA+B,CACnD,OAAkC,EAClC,QAAoB,EACpB,IAAsB;IAEtB,MAAM,GAAG,GAA8B,EAAE,CAAC;IAC1C,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,KAAK,MAAM,CAAC,OAAO,EAAE,MAAM,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QACxD,MAAM,MAAM,GAAG,MAAM,sBAAsB,CAAC,MAAM,EAAE,QAAQ,EAAE,IAAI,CAAC,CAAC;QACpE,GAAG,CAAC,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;QAC7B,SAAS,IAAI,MAAM,CAAC,OAAO,CAAC;IAC9B,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;AAC9C,CAAC"}

package/dist/src/analysis/hedging-classifier.d.ts

@@ -0,0 +1,34 @@
+/**
+ * Hedging classifier (spec §四.3, v0.2).
+ *
+ * regex 召回 → LLM 二次判定。把 extractHedgingSignals 的 candidate 句子送给小模型,
+ * 让模型判断"这一句是知识层面的不确定,还是业务推理 / 假设 / 礼貌表达"。
+ *
+ * 关键约束:
+ *   - cost 上限: 默认 maxCandidates=50, 超出截断 + warn
+ *   - cache: in-memory Map<sha256(sentence), HedgingVerdict>, 同句子不重复调用
+ *   - 失败降级: 调用 / 解析失败 → 该 candidate 默认 isUncertainty=true (保守保留)
+ */
+import type { ExecutorFn, HedgingVerdict } from '../types.js';
+export interface HedgingCandidate {
+    sampleId: string;
+    sentence: string;
+    context: string;
+}
+export interface ClassifyOptions {
+    maxCandidates?: number;
+    model?: string;
+    batchSize?: number;
+}
+export interface ClassifyResult {
+    verdicts: HedgingVerdict[];
+    costUSD: number;
+    truncated: boolean;
+}
+export declare function clearHedgingCache(): void;
+/**
+ * 对一批 candidate 做分类判定。返回与 input 等长的 verdicts。
+ * 失败降级:单批 LLM 调用 / 解析失败 → 该 batch 全部 isUncertainty=true (保守保留)。
+ */
+export declare function classifyHedgingCandidates(candidates: HedgingCandidate[], executor: ExecutorFn, opts?: ClassifyOptions): Promise<ClassifyResult>;
+//# sourceMappingURL=hedging-classifier.d.ts.map

package/dist/src/analysis/hedging-classifier.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"hedging-classifier.d.ts","sourceRoot":"","sources":["../../../src/analysis/hedging-classifier.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAIH,OAAO,KAAK,EAAE,UAAU,EAAE,cAAc,EAAE,MAAM,aAAa,CAAC;AAE9D,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAe;IAC9B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,cAAc,EAAE,CAAC;IAC3B,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;CACpB;AASD,wBAAgB,iBAAiB,IAAI,IAAI,CAExC;AA+DD;;;GAGG;AACH,wBAAsB,yBAAyB,CAC7C,UAAU,EAAE,gBAAgB,EAAE,EAC9B,QAAQ,EAAE,UAAU,EACpB,IAAI,CAAC,EAAE,eAAe,GACrB,OAAO,CAAC,cAAc,CAAC,CAmEzB"}

package/dist/src/analysis/hedging-classifier.js

@@ -0,0 +1,144 @@
+/**
+ * Hedging classifier (spec §四.3, v0.2).
+ *
+ * regex 召回 → LLM 二次判定。把 extractHedgingSignals 的 candidate 句子送给小模型,
+ * 让模型判断"这一句是知识层面的不确定,还是业务推理 / 假设 / 礼貌表达"。
+ *
+ * 关键约束:
+ *   - cost 上限: 默认 maxCandidates=50, 超出截断 + warn
+ *   - cache: in-memory Map<sha256(sentence), HedgingVerdict>, 同句子不重复调用
+ *   - 失败降级: 调用 / 解析失败 → 该 candidate 默认 isUncertainty=true (保守保留)
+ */
+import { createHash } from 'node:crypto';
+const DEFAULT_MAX_CANDIDATES = 50;
+const DEFAULT_BATCH_SIZE = 10;
+const DEFAULT_MODEL = 'claude-haiku-4-5-20251001';
+// in-memory cache,进程内复用。key = sha256(sentence)。
+const verdictCache = new Map();
+export function clearHedgingCache() {
+    verdictCache.clear();
+}
+function hashSentence(sentence) {
+    return createHash('sha256').update(sentence).digest('hex').slice(0, 16);
+}
+function buildBatchPrompt(batch) {
+    const items = batch
+        .map((c, i) => `### ${i + 1}\n命中句: ${c.sentence}\n上下文: ${c.context}`)
+        .join('\n\n');
+    return [
+        '判断以下每段话是否表达了"知识 / 事实层面的不确定"。',
+        '',
+        '满足以下任一条算"不确定"(isUncertainty=true):',
+        '  - 表示自己不知道答案 / 缺少信息 / 需要查证',
+        '  - 给出答案但显著降级措辞("可能是 X 但不确定")',
+        '',
+        '以下情况 **不算** "不确定" (isUncertainty=false):',
+        '  - 业务可能性分析(讨论多种业务场景的可能性)',
+        '  - 礼貌 / 假设性措辞("如果你需要,可能可以...")',
+        '  - 对未来不确定("未来可能会..."),不属于知识不确定',
+        '',
+        items,
+        '',
+        '返回 JSON 数组(不要 markdown 代码块标记),按上面编号顺序,每项格式:',
+        '{"id": <编号>, "isUncertainty": <bool>, "confidence": <0-1 的小数>, "reason": "<10 字以内简短理由>"}',
+    ].join('\n');
+}
+function parseBatchResponse(text, batchSize) {
+    const trimmed = text.trim();
+    const jsonMatch = trimmed.match(/\[[\s\S]*\]/);
+    if (!jsonMatch)
+        throw new Error(`classifier returned non-JSON: ${trimmed.slice(0, 100)}`);
+    const parsed = JSON.parse(jsonMatch[0]);
+    if (!Array.isArray(parsed))
+        throw new Error('classifier response is not an array');
+    // 按 id 排序后填充,缺失的 id 走降级
+    const verdicts = [];
+    for (let i = 1; i <= batchSize; i++) {
+        const item = parsed.find((p) => Number(p.id) === i);
+        if (!item) {
+            verdicts.push({ isUncertainty: true, confidence: 0, reason: `classifier missing id ${i}` });
+            continue;
+        }
+        verdicts.push({
+            isUncertainty: Boolean(item.isUncertainty),
+            confidence: Math.max(0, Math.min(1, Number(item.confidence) || 0)),
+            reason: String(item.reason || '').slice(0, 60),
+        });
+    }
+    return verdicts;
+}
+function fallbackVerdict(reason) {
+    return { isUncertainty: true, confidence: 0, reason: `classifier failed: ${reason}` };
+}
+/**
+ * 对一批 candidate 做分类判定。返回与 input 等长的 verdicts。
+ * 失败降级:单批 LLM 调用 / 解析失败 → 该 batch 全部 isUncertainty=true (保守保留)。
+ */
+export async function classifyHedgingCandidates(candidates, executor, opts) {
+    const max = opts?.maxCandidates ?? DEFAULT_MAX_CANDIDATES;
+    const batchSize = opts?.batchSize ?? DEFAULT_BATCH_SIZE;
+    const model = opts?.model ?? DEFAULT_MODEL;
+    const truncated = candidates.length > max;
+    const work = candidates.slice(0, max);
+    const verdicts = new Array(candidates.length);
+    let costUSD = 0;
+    if (truncated) {
+        process.stderr.write(`[omk] hedging classifier: ${candidates.length} candidates exceeds maxCandidates=${max}, truncated\n`);
+        // 被截断的部分走降级保守保留
+        for (let i = max; i < candidates.length; i++) {
+            verdicts[i] = fallbackVerdict('truncated by maxCandidates');
+        }
+    }
+    // 先查 cache
+    const uncached = [];
+    for (let i = 0; i < work.length; i++) {
+        const key = hashSentence(work[i].sentence);
+        const hit = verdictCache.get(key);
+        if (hit) {
+            verdicts[i] = hit;
+        }
+        else {
+            uncached.push({ idx: i, cand: work[i] });
+        }
+    }
+    // 按 batch 调用
+    for (let b = 0; b < uncached.length; b += batchSize) {
+        const slice = uncached.slice(b, b + batchSize);
+        const batchCands = slice.map((s) => s.cand);
+        const prompt = buildBatchPrompt(batchCands);
+        let result;
+        try {
+            result = await executor({
+                model,
+                system: '你是一个判定 LLM 输出是否表达"知识层面不确定"的分类器。只返回 JSON 数组。',
+                prompt,
+            });
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            slice.forEach((s) => { verdicts[s.idx] = fallbackVerdict(`exec ${msg}`); });
+            continue;
+        }
+        if (!result.ok || !result.output) {
+            slice.forEach((s) => { verdicts[s.idx] = fallbackVerdict(`exec ${result.error ?? 'no output'}`); });
+            costUSD += result.costUSD ?? 0;
+            continue;
+        }
+        costUSD += result.costUSD ?? 0;
+        let parsed;
+        try {
+            parsed = parseBatchResponse(result.output, slice.length);
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            slice.forEach((s) => { verdicts[s.idx] = fallbackVerdict(`parse ${msg}`); });
+            continue;
+        }
+        slice.forEach((s, i) => {
+            verdicts[s.idx] = parsed[i];
+            verdictCache.set(hashSentence(s.cand.sentence), parsed[i]);
+        });
+    }
+    return { verdicts, costUSD, truncated };
+}
+//# sourceMappingURL=hedging-classifier.js.map

package/dist/src/analysis/hedging-classifier.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"hedging-classifier.js","sourceRoot":"","sources":["../../../src/analysis/hedging-classifier.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAsBzC,MAAM,sBAAsB,GAAG,EAAE,CAAC;AAClC,MAAM,kBAAkB,GAAG,EAAE,CAAC;AAC9B,MAAM,aAAa,GAAG,2BAA2B,CAAC;AAElD,gDAAgD;AAChD,MAAM,YAAY,GAAG,IAAI,GAAG,EAA0B,CAAC;AAEvD,MAAM,UAAU,iBAAiB;IAC/B,YAAY,CAAC,KAAK,EAAE,CAAC;AACvB,CAAC;AAED,SAAS,YAAY,CAAC,QAAgB;IACpC,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AAC1E,CAAC;AAED,SAAS,gBAAgB,CAAC,KAAyB;IACjD,MAAM,KAAK,GAAG,KAAK;SAChB,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,QAAQ,UAAU,CAAC,CAAC,OAAO,EAAE,CAAC;SACpE,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO;QACL,8BAA8B;QAC9B,EAAE;QACF,oCAAoC;QACpC,6BAA6B;QAC7B,+BAA+B;QAC/B,EAAE;QACF,0CAA0C;QAC1C,2BAA2B;QAC3B,iCAAiC;QACjC,iCAAiC;QACjC,EAAE;QACF,KAAK;QACL,EAAE;QACF,6CAA6C;QAC7C,0FAA0F;KAC3F,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AASD,SAAS,kBAAkB,CAAC,IAAY,EAAE,SAAiB;IACzD,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,SAAS,GAAG,OAAO,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC;IAC/C,IAAI,CAAC,SAAS;QAAE,MAAM,IAAI,KAAK,CAAC,iCAAiC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IAC1F,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAiB,CAAC;IACxD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,qCAAqC,CAAC,CAAC;IACnF,wBAAwB;IACxB,MAAM,QAAQ,GAAqB,EAAE,CAAC;IACtC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;QACpC,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC;QACpD,IAAI,CAAC,IAAI,EAAE,CAAC;YACV,QAAQ,CAAC,IAAI,CAAC,EAAE,aAAa,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,yBAAyB,CAAC,EAAE,EAAE,CAAC,CAAC;YAC5F,SAAS;QACX,CAAC;QACD,QAAQ,CAAC,IAAI,CAAC;YACZ,aAAa,EAAE,OAAO,CAAC,IAAI,CAAC,aAAa,CAAC;YAC1C,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC;YAClE,MAAM,EAAE,MAAM,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;SAC/C,CAAC,CAAC;IACL,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,SAAS,eAAe,CAAC,MAAc;IACrC,OAAO,EAAE,aAAa,EAAE,IAAI,EAAE,UAAU,EAAE,CAAC,EAAE,MAAM,EAAE,sBAAsB,MAAM,EAAE,EAAE,CAAC;AACxF,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,yBAAyB,CAC7C,UAA8B,EAC9B,QAAoB,EACpB,IAAsB;IAEtB,MAAM,GAAG,GAAG,IAAI,EAAE,aAAa,IAAI,sBAAsB,CAAC;IAC1D,MAAM,SAAS,GAAG,IAAI,EAAE,SAAS,IAAI,kBAAkB,CAAC;IACxD,MAAM,KAAK,GAAG,IAAI,EAAE,KAAK,IAAI,aAAa,CAAC;IAC3C,MAAM,SAAS,GAAG,UAAU,CAAC,MAAM,GAAG,GAAG,CAAC;IAC1C,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IACtC,MAAM,QAAQ,GAAqB,IAAI,KAAK,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC;IAChE,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,IAAI,SAAS,EAAE,CAAC;QACd,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,6BAA6B,UAAU,CAAC,MAAM,qCAAqC,GAAG,eAAe,CAAC,CAAC;QAC5H,gBAAgB;QAChB,KAAK,IAAI,CAAC,GAAG,GAAG,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,QAAQ,CAAC,CAAC,CAAC,GAAG,eAAe,CAAC,4BAA4B,CAAC,CAAC;QAC9D,CAAC;IACH,CAAC;IAED,WAAW;IACX,MAAM,QAAQ,GAA8C,EAAE,CAAC;IAC/D,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACrC,MAAM,GAAG,GAAG,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC;QAC3C,MAAM,GAAG,GAAG,YAAY,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,GAAG,EAAE,CAAC;YACR,QAAQ,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC;QACpB,CAAC;aAAM,CAAC;YACN,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;QAC3C,CAAC;IACH,CAAC;IAED,aAAa;IACb,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,IAAI,SAAS,EAAE,CAAC;QACpD,MAAM,KAAK,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,SAAS,CAAC,CAAC;QAC/C,MAAM,UAAU,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QAC5C,MAAM,MAAM,GAAG,gBAAgB,CAAC,UAAU,CAAC,CAAC;QAC5C,IAAI,MAAM,CAAC;QACX,IAAI,CAAC;YACH,MAAM,GAAG,MAAM,QAAQ,CAAC;gBACtB,KAAK;gBACL,MAAM,EAAE,6CAA6C;gBACrD,MAAM;aACP,CAAC,CAAC;QACL,CAAC;QAAC,OAAO,GAAY,EAAE,CAAC;YACtB,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,eAAe,CAAC,QAAQ,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC5E,SAAS;QACX,CAAC;QACD,IAAI,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;YACjC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,eAAe,CAAC,QAAQ,MAAM,CAAC,KAAK,IAAI,WAAW,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACpG,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC;YAC/B,SAAS;QACX,CAAC;QACD,OAAO,IAAI,MAAM,CAAC,OAAO,IAAI,CAAC,CAAC;QAC/B,IAAI,MAAwB,CAAC;QAC7B,IAAI,CAAC;YACH,MAAM,GAAG,kBAAkB,CAAC,MAAM,CAAC,MAAM,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC;QAC3D,CAAC;QAAC,OAAO,GAAY,EAAE,CAAC;YACtB,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,GAAG,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,eAAe,CAAC,SAAS,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC7E,SAAS;QACX,CAAC;QACD,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;YACrB,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;YAC5B,YAAY,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAC7D,CAAC,CAAC,CAAC;IACL,CAAC;IAED,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC;AAC1C,CAAC"}

package/dist/src/analysis/report-diagnostics.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Auto-analysis: detect patterns and generate insights from evaluation results.
+ */
+import type { Report, AnalysisResult } from '../types.js';
+/**
+ * Analyze an evaluation report and produce insights + suggestions.
+ */
+export declare function analyzeResults(report: Report): AnalysisResult;
+//# sourceMappingURL=report-diagnostics.d.ts.map

package/dist/src/analysis/report-diagnostics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"report-diagnostics.d.ts","sourceRoot":"","sources":["../../../src/analysis/report-diagnostics.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,EAAE,MAAM,EAAwB,cAAc,EAAE,MAAM,aAAa,CAAC;AAEhF;;GAEG;AACH,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,GAAG,cAAc,CA2C7D"}