Files
ai-platform/src/modules/jieya-ibp/lib/answers.ts
T
2026-10-18 15:27:02 +08:00

468 lines
23 KiB
TypeScript
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
import type { MetricLink, MetricNode } from '../data/sqlite';
import { queryProductLines, queryRegions, queryRisks, queryTrend } from '../data/sqlite';
import { achieveRate, fmt, fmtSigned, gapOf } from './metric';
import type { ParsedAsk } from './intent';
/**
* 结构化回答引擎:消费 intent.ts 的解析结果,查询 SQL 数据层,
* 产出对齐设计书 5.3 五段式(结论 + 迷你图 + 分点 + 行动 + 来源)的 AiReply。
* 全部计算离线完成:外推用最小二乘、传播用网络边强度、归因用偏差×权重贡献分解。
*/
export interface AiSpark {
/** 折线值:历史 12 期(+ 假设/外推段) */
values: number[];
/** values 末尾多少点属于外推(虚线渲染) */
forecastCount?: number;
/** 目标参考虚线 */
target?: number;
unit: string;
}
export interface AiPoint { label?: string; text: string; tone?: 'good' | 'warn' | 'bad' }
export interface AiAction {
label: string;
/** 跳转到指标(联动左侧树与中央看板) */
code?: string;
/** 以该问句继续追问 */
ask?: string;
}
export interface AiReply {
/** 核心结论:1-2 句,直接回答问句 */
conclusion: string;
spark?: AiSpark;
/** 分点详细分析 */
points: AiPoint[];
/** 建议行动(可点击) */
actions: AiAction[];
/** 数据来源标注:数据域 + 更新频率 + 权限 */
source: string;
/** 回答聚焦的指标,用于行动按钮联动 */
focusCode?: string;
}
export interface AnswerCtx {
flat: MetricNode[];
byCode: Map<string, MetricNode>;
links: MetricLink[];
/** 全局时间游标:历史序列截断到该期,结论文字与迷你图同口径(节点值已由 App 层投影) */
period: number;
}
const label = (n: MetricNode) => n.name_zh;
const unitOf = (n: MetricNode) => n.unit;
const SRC_METRIC = '指标中台(metric_node / metric_trend / metric_link)· 月度更新 · 内部可见';
const SRC_BIZ = '经营分析(product_line / region_data / risk_matrix)· 季度更新 · 内部可见';
/** 最小二乘线性回归:返回斜率与外推函数 */
function linreg(ys: number[]) {
const n = ys.length;
if (n < 2) return { slope: 0, at: (_i: number) => ys[0] ?? 0 };
const mx = (n - 1) / 2;
const my = ys.reduce((a, b) => a + b, 0) / n;
let num = 0, den = 0;
for (let i = 0; i < n; i++) { num += (i - mx) * (ys[i] - my); den += (i - mx) ** 2; }
const slope = den ? num / den : 0;
return { slope, at: (i: number) => my + slope * (i - mx) };
}
/** 游标前的历史序列(含游标当期):回答不会把“未来月份”当成已经历的实绩 */
const trendOf = (code: string, period: number): number[] =>
queryTrend(code).map(p => p.actual).slice(0, Math.max(2, period + 1));
/** 游标口径的红黄绿计数:ctx.flat 已由 App 投影,直接按定级字段统计即得该时点分布 */
const countAlerts = (ctx: AnswerCtx) => {
let red = 0, yellow = 0, green = 0;
ctx.flat.forEach(n => {
if (n.level < 2) return;
if (n.alert_level === 'red') red++;
else if (n.alert_level === 'yellow') yellow++;
else green++;
});
return { red, yellow, green, total: ctx.flat.length };
};
/** 偏差 / 变化量单位:% 类指标用百分点 pt(与目标值本身的 % 区分口径),其余按语言取单位 */
const gapUnit = (n: MetricNode) => (n.unit === '%' ? 'pt' : n.unit);
/* ── 各意图引擎 ─────────────────────────────────────────────────────────── */
/** 指标查询:当期值 / 目标 / 偏差 / 达成率 + 趋势 + 最大拖累 */
function answerQuery(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const n = ctx.byCode.get(parsed.metricCode!)!;
const gap = gapOf(n);
const rate = achieveRate(n);
const trend = trendOf(n.code, ctx.period);
const last3 = trend.slice(-3);
const good = gap >= 0;
const kids = ctx.flat
.filter(k => k.parent_code === n.code)
.sort((a, b) => Math.abs(gapOf(b)) - Math.abs(gapOf(a)));
const worst = kids.find(k => gapOf(k) < 0);
const mom = last3.length >= 2 ? last3[2] - last3[0] : 0;
const conclusion = `「${label(n)}」当期 ${fmt(n.value)}${unitOf(n)},目标 ${fmt(n.target)}${unitOf(n)}(偏差 ${fmtSigned(gap)}${gapUnit(n)}),达成率 ${fmt(rate, 0)}%,${good ? '处于达标线以上。' : '尚未达标。'}`;
return {
conclusion,
spark: { values: trend, target: n.target, unit: unitOf(n) },
focusCode: n.code,
points: [
{ label: '责任部门', text: `${n.owner} · ${n.frequency}更新` },
{ label: '近 3 期', text: `${last3.map(v => fmt(v)).join(' → ')}${unitOf(n)}(${mom >= 0 ? ('改善基调') : ('走弱')})`, tone: mom >= 0 ? 'good' : 'warn' },
worst && {
label: '最大拖累',
text: `下层「${label(worst)}」偏差最大(${fmtSigned(gapOf(worst))}${gapUnit(worst)})`,
tone: 'bad',
},
{ label: '指标说明', text: n.desc_zh},
].filter(Boolean) as AiPoint[],
actions: [
{ label: `查看「${label(n)}」归因`, code: n.code, ask: `${n.name_zh}为什么低于目标`},
{ label: '趋势预测', ask: `预测${n.name_zh}下月走势`},
],
source: SRC_METRIC,
};
}
/** 归因分析:子因子偏差×权重贡献分解 + 跨树驱动边排序 */
function answerAttribute(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const n = ctx.byCode.get(parsed.metricCode ?? 'R04')!;
const gap = gapOf(n);
const kids = ctx.flat.filter(k => k.parent_code === n.code);
const contribs = kids
.map(k => ({ node: k, contrib: +(gapOf(k) * k.weight).toFixed(2) }))
.sort((a, b) => a.contrib - b.contrib); // 负贡献(拖累)优先
const totalNeg = contribs.filter(c => c.contrib < 0).reduce((a, b) => a + Math.abs(b.contrib), 0) || 1;
const outsiders = ctx.links
.filter(l => l.to_code === n.code && !kids.some(k => k.code === l.from_code))
.map(l => ({ link: l, node: ctx.byCode.get(l.from_code)! }))
.filter(x => x.node)
.sort((a, b) => Math.abs(b.link.strength) - Math.abs(a.link.strength))
.slice(0, 2);
const top = contribs.slice(0, 3);
const topStr = top
.map(c => `${label(c.node)}(贡献 ${Math.round(Math.abs(c.contrib) / totalNeg * 100)}%)`)
.join('、');
const conclusion = `「${label(n)}」偏差 ${fmtSigned(gap)}${gapUnit(n)} 的主因:${topStr}。属于结构性偏离,建议优先调整产品与供应结构,而非单纯压费用。`;
return {
conclusion,
spark: { values: trendOf(n.code, ctx.period), target: n.target, unit: unitOf(n) },
focusCode: n.code,
points: [
...top.map(c => ({
label: label(c.node),
text: `当期 ${fmt(c.node.value)}${unitOf(c.node)} / 目标 ${fmt(c.node.target)}${unitOf(c.node)}(${fmtSigned(gapOf(c.node))}${gapUnit(c.node)})· ${c.node.owner}`,
tone: c.contrib < 0 ? 'bad' as const : 'good' as const,
})),
...outsiders.map(x => ({
label: '跨树驱动',
text: `「${label(x.node)}」→ 本指标(强度 ${x.link.strength},${x.link.relation === 'conflict' ? '负向制约' : '正向驱动'})`,
})),
],
actions: [
...(top[0] ? [{ label: `打开主因「${label(top[0].node)}」`, code: top[0].node.code }] : []),
...(top[0] ? [{ label: `模拟改善路径`, ask: `如果${top[0].node.name_zh}提升到目标会怎样`}] : []),
],
source: SRC_METRIC,
};
}
/** 趋势预测:最小二乘外推 3 期 + 达标判断(在手订单特例:收入支撑度) */
function answerForecast(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const n = ctx.byCode.get(parsed.metricCode ?? 'R01')!;
const trend = trendOf(n.code, ctx.period);
const { slope, at } = linreg(trend);
/* 从游标往后外推三期(不是从 12 月末),历史越短回归越不稳定,因而标注样本期数 */
const h = trend.length;
const ext = [at(h), at(h + 1), at(h + 2)].map(v => +v.toFixed(2));
const end = ext[2];
const endGap = n.polarity === 'down' ? -(end - n.target) : end - n.target;
const willHit = endGap >= 0;
let conclusion: string;
const tail = `${fmt(trend[h - 1])} → ${fmt(ext[0])} → ${fmt(ext[2])}${unitOf(n)}`;
if (n.code === 'R10') {
const r1 = ctx.byCode.get('R01')!;
const cover = +((n.value / (r1.value * 3) * 100)).toFixed(1);
conclusion = `在手订单 ${fmt(n.value)}${unitOf(n)}约可覆盖未来 3 个月收入节奏(月均 ${fmt(r1.value)}${unitOf(r1)})的 ${fmt(cover, 0)}%,缺口需靠新签订单补齐。在手订单趋势:${tail}。`;
} else {
conclusion = `按游标前 ${h} 期线性趋势外推:「${label(n)}」下期 ${fmt(ext[0])}${unitOf(n)},3 期后 ${fmt(end)}${unitOf(n)}。对比目标 ${fmt(n.target)}${unitOf(n)},${willHit ? '有望达成' : `仍差 ${fmt(Math.abs(endGap))}${gapUnit(n)}`}。`;
}
const mom = slope; // 每期平均变化
return {
conclusion,
spark: { values: [...trend, ...ext], forecastCount: 3, target: n.target, unit: unitOf(n) },
focusCode: n.code,
points: [
{ label: '趋势斜率', text: `每期 ${fmtSigned(mom, 2)}${gapUnit(n)}(${h} 期回归斜率)`, tone: mom >= 0 ? 'good' : 'warn' },
{ label: '近期动能', text: trend.slice(-3).map(v => fmt(v)).join(' → ') + unitOf(n) },
{ label: '达标判定', text: willHit ? (`3 期外推终点回到目标线以上`) : (`当前节奏下缺口将持续,需要额外举措拉动`), tone: willHit ? 'good' : 'bad' },
],
actions: [
{ label: `查看「${label(n)}」归因`, code: n.code, ask: `${n.name_zh}为什么低于目标`},
],
source: SRC_METRIC + (' ※线性回归外推'),
};
}
/** 假设模拟:假设值 → 相对变化率沿网络边强度传播(单位无关近似) */
function answerWhatIf(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const n = ctx.byCode.get(parsed.metricCode ?? 'R04')!;
const hasAbs = parsed.absValue !== undefined;
const hasDelta = parsed.deltaValue !== undefined;
if (!hasAbs && !hasDelta) {
return {
conclusion: `模拟需要给出假设数值。例如:「如果高毛利产品占比提升到 40% 会怎样」,指定指标与目标值即可推演。`,
points: [],
actions: [{ label: '试一个例子', ask: '如果高毛利产品占比提升到 40% 会怎样'}],
source: SRC_METRIC,
focusCode: n.code,
};
}
const newValue = hasAbs ? parsed.absValue! : +(n.value + parsed.deltaValue!).toFixed(2);
const delta = +(newValue - n.value).toFixed(2);
const relChange = n.value !== 0 ? delta / Math.abs(n.value) : 0;
const outs = ctx.links
.filter(l => l.from_code === n.code)
.map(l => ({ link: l, node: ctx.byCode.get(l.to_code)! }))
.filter(x => x.node)
.sort((a, b) => Math.abs(b.link.strength) - Math.abs(a.link.strength))
.slice(0, 3);
const selfGapNew = n.polarity === 'down' ? -(newValue - n.target) : newValue - n.target;
const conclusion = `若「${label(n)}」从 ${fmt(n.value)}${unitOf(n)} 变为 ${fmt(newValue)}${unitOf(n)}(${fmtSigned(delta)}${gapUnit(n)}):${selfGapNew >= 0 ? '该指标将回到目标线以上' : '该指标仍低于目标'}。${outs.length ? `影响最大的下游是「${label(outs[0].node)}」。` : ''}`;
return {
conclusion,
spark: { values: [...trendOf(n.code, ctx.period), newValue], forecastCount: 1, target: n.target, unit: unitOf(n) },
focusCode: n.code,
points: [
{
label: label(n),
text: `对比目标 ${fmt(n.target)}${unitOf(n)}:${selfGapNew >= 0 ? '达标(偏差' : '未达标(缺口'}${fmtSigned(selfGapNew)}${gapUnit(n)})`,
tone: selfGapNew >= 0 ? 'good' : 'warn',
},
...outs.map(x => {
const impacted = +(x.node.value * (1 + x.link.strength * relChange)).toFixed(2);
const chg = +(impacted - x.node.value).toFixed(2);
return {
label: label(x.node),
text: `${fmt(x.node.value)} → ${fmt(impacted)}${unitOf(x.node)}(${fmtSigned(chg)}${gapUnit(x.node)},传播强度 ${x.link.strength})`,
tone: chg >= 0 ? ('good' as const) : ('warn' as const),
};
}),
],
actions: [
{ label: `打开「${label(n)}」`, code: n.code },
...(outs[0] ? [{ label: `查看受影响的「${label(outs[0].node)}」`, code: outs[0].node.code }] : []),
],
source: SRC_METRIC + (' ※按网络边强度线性传播近似'),
};
}
/** 对比分析:产品线 / 区域 */
function answerCompare(parsed: ParsedAsk, _ctx: AnswerCtx): AiReply {
if (parsed.compareScope === 'region') {
const regions = queryRegions();
const top = regions[0];
const weakMargin = [...regions].sort((a, b) => a.gross_margin - b.gross_margin)[0];
const bestPerCapita = [...regions].sort((a, b) => b.per_capita - a.per_capita)[0];
return {
conclusion: `${top.name_zh}收入贡献最大(${fmt(top.revenue)} 亿元,占 ${fmt((top.revenue / regions.reduce((a, r) => a + r.revenue, 0)) * 100, 0)}%);人均产值以${bestPerCapita.name_zh}最高(${fmt(bestPerCapita.per_capita, 0)} 万元);${weakMargin.name_zh}毛利率最低(${fmt(weakMargin.gross_margin)}%),是盈利结构的洼地。`,
points: regions.map(r => ({
label: r.name_zh,
text: `收入 ${fmt(r.revenue)} 亿元 · 毛利率 ${fmt(r.gross_margin)}% · 人均产值 ${fmt(r.per_capita, 0)} 万元 · ${fmt(r.headcount, 0)} 人`,
tone: r.gross_margin === weakMargin.gross_margin ? 'bad' as const : undefined,
})),
actions: [{ label: '切到指标明细查看区域对比', code: 'R01' }],
source: SRC_BIZ,
};
}
const lines = queryProductLines();
const byRevenue = [...lines].sort((a, b) => b.revenue - a.revenue);
const byMargin = [...lines].sort((a, b) => b.gross_margin - a.gross_margin);
const byGrowth = [...lines].sort((a, b) => b.yoy_growth - a.yoy_growth);
const total = lines.reduce((a, l) => a + l.revenue, 0);
return {
conclusion: `${byRevenue[0].name_zh}收入规模最大(${fmt(byRevenue[0].revenue)} 亿元,占 ${fmt(byRevenue[0].revenue / total * 100, 0)}%);${byMargin[0].name_zh}毛利率最高(${fmt(byMargin[0].gross_margin)}%);${byGrowth[0].name_zh}增速最快(+${fmt(byGrowth[0].yoy_growth)}%),是结构升级的主引擎。`,
points: lines.map(l => ({
label: l.name_zh,
text: `收入 ${fmt(l.revenue)} 亿元(占比 ${fmt(l.share)}%)· 毛利率 ${fmt(l.gross_margin)}% · 同比 +${fmt(l.yoy_growth)}%`,
tone: l.gross_margin === byMargin[byMargin.length - 1].gross_margin ? 'warn' as const : undefined,
})),
actions: [
{ label: '模拟高毛利占比提升', ask: '如果高毛利产品占比提升到 40% 会怎样'},
{ label: '毛利率归因分析', ask: '毛利率为什么低于目标'},
],
source: SRC_BIZ,
};
}
/** 异常检测:红色 + 黄色偏离清单 */
function answerAnomaly(_parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const counts = countAlerts(ctx);
const bad = ctx.flat
.filter(n => n.level >= 2 && (n.alert_level === 'red' || (n.alert_level === 'yellow' && gapOf(n) < 0)))
.sort((a, b) => (a.alert_level === 'red' ? -1 : 1) - (b.alert_level === 'red' ? -1 : 1) || achieveRate(a) - achieveRate(b))
.slice(0, 5);
const worst = bad[0];
return {
conclusion: `当前红色告警 ${counts.red} 项、黄色偏离 ${counts.yellow} 项。最紧急:${worst ? `「${label(worst)}」达成率仅 ${fmt(achieveRate(worst), 0)}%` : '无'}。`,
points: bad.map(n => ({
label: `${label(n)}${n.alert_level === 'red' ? ' 🔴' : ' 🟡'}`,
text: `当期 ${fmt(n.value)}${unitOf(n)} / 目标 ${fmt(n.target)}${unitOf(n)}(偏差 ${fmtSigned(gapOf(n))}${gapUnit(n)})· ${n.owner}`,
tone: n.alert_level === 'red' ? 'bad' as const : 'warn' as const,
})),
actions: [
...(worst ? [{ label: `打开「${label(worst)}」`, code: worst.code }] : []),
{ label: '查看升级建议', ask: '哪些指标需要升级'},
],
source: SRC_METRIC,
};
}
/** 升级处置:红色 + 连续 3 期低于目标 */
function answerEscalate(_parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const streakOf = (n: MetricNode) => {
const ts = queryTrend(n.code).slice(0, Math.max(1, ctx.period + 1));
const gaps = ts.map(p => (n.polarity === 'down' ? -(p.actual - p.target) : p.actual - p.target));
let s = 0;
for (let i = gaps.length - 1; i >= 0 && gaps[i] < 0; i--) s++;
return s;
};
const targets = ctx.flat
.filter(n => n.level >= 2 && (n.alert_level === 'red' || streakOf(n) >= 3))
.map(n => ({ n, streak: streakOf(n) }))
.sort((a, b) => (a.n.alert_level === 'red' ? -1 : 1) - (b.n.alert_level === 'red' ? -1 : 1) || b.streak - a.streak)
.slice(0, 4);
return {
conclusion: `建议升级 ${targets.length} 项:${targets.map(x => label(x.n)).join('、') || '—'}。红色或连续 3 期未达标已触发上报阈值,建议纳入经营例会跟踪。`,
points: targets.map(x => ({
label: label(x.n),
text: `${x.streak >= 3 ? `连续 ${x.streak} 期低于目标` : '触发预警阈值'} · 责任部门 ${x.n.owner} · 达成率 ${fmt(achieveRate(x.n), 0)}%`,
tone: 'bad' as const,
})),
actions: targets.slice(0, 2).flatMap(x => [
{ label: `打开「${label(x.n)}」`, code: x.n.code },
]),
source: SRC_METRIC,
};
}
/** 行动建议:偏差拉回路径(子因子 top2) */
function answerAdvice(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const n = ctx.byCode.get(parsed.metricCode ?? 'R04')!;
const gap = gapOf(n);
const kids = ctx.flat
.filter(k => k.parent_code === n.code && gapOf(k) < 0)
.sort((a, b) => Math.abs(gapOf(b)) - Math.abs(gapOf(a)))
.slice(0, 2);
const paths = kids.length ? kids : ctx.flat.filter(k => k.parent_code === n.code).slice(0, 2);
return {
conclusion: `把「${label(n)}」拉回目标 ${fmt(n.target)}${unitOf(n)} 的两条路径:${paths.map(k => `将「${label(k)}」从 ${fmt(k.value)}${unitOf(k)} 提到 ${fmt(k.target)}${unitOf(k)}`).join(';')}。`,
spark: { values: trendOf(n.code, ctx.period), target: n.target, unit: unitOf(n) },
focusCode: n.code,
points: [
...paths.map((k, i) => ({
label: `路径 ${i + 1}`,
text: `把「${label(k)}」提到 ${fmt(k.target)}${unitOf(k)}(当前 ${fmt(k.value)}${unitOf(k)},偏差 ${fmtSigned(gapOf(k))}${gapUnit(k)})· 责任部门 ${k.owner}`,
tone: 'warn' as const,
})),
{
label: '现状',
text: `本指标达成率 ${fmt(achieveRate(n), 0)}%(偏差 ${fmtSigned(gap)}${gapUnit(n)}),${gap >= 0 ? '当前保持达标' : '需要纠偏动作'}`,
},
],
actions: [
...(paths[0] ? [
{ label: `模拟路径 1 影响`, ask: `如果${paths[0].name_zh}提升到 ${fmt(paths[0].target)}会怎样`},
{ label: `打开「${label(paths[0])}」`, code: paths[0].code },
] : []),
],
source: SRC_METRIC,
};
}
/** 经营总览:OV + 八维度 + 红黄绿 */
function answerOverview(_parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
const ov = ctx.byCode.get('OV')!;
const dims = ctx.flat.filter(n => n.level === 1).sort((a, b) => gapOf(b) - gapOf(a));
const counts = countAlerts(ctx);
const weakest = dims[dims.length - 1];
const strongest = dims[0];
return {
conclusion: `整体经营健康度 ${fmt(ov.value)} 分(目标 ${fmt(ov.target)} 分,达成率 ${fmt(achieveRate(ov), 0)}%)。八大维度 ${dims.filter(d => d.alert_level === 'green').length} 项绿灯,最弱的是「${label(weakest)}」(${fmt(weakest.value)} 分)。`,
spark: { values: trendOf('OV', ctx.period), target: ov.target, unit: '分'},
focusCode: 'OV',
points: [
{ label: '最强维度', text: `「${label(strongest)}」${fmt(strongest.value)} 分(偏差 ${fmtSigned(gapOf(strongest))} 分)`, tone: 'good' },
{ label: '最弱维度', text: `「${label(weakest)}」${fmt(weakest.value)} 分(偏差 ${fmtSigned(gapOf(weakest))} 分)· 瓶颈在 AI 渗透与数据资产在线率`, tone: 'bad' },
{ label: '指标状态', text: `L2 及以下:红色 ${counts.red} 项 / 黄色 ${counts.yellow} 项 / 绿色 ${counts.green} 项(共 ${counts.total - 9} 项)`},
],
actions: [
{ label: `打开「${label(weakest)}」`, code: weakest.code },
{ label: '查看异常指标', ask: '哪些指标异常'},
],
source: SRC_METRIC,
};
}
/** 风险矩阵 */
function answerRisk(_parsed: ParsedAsk, _ctx: AnswerCtx): AiReply {
const risks = queryRisks();
const top = risks[0];
const escalated = risks.filter(r => r.status_zh.includes('升级'));
return {
conclusion: `当前登记风险 ${risks.length} 项。评分最高:${top.name_zh}(概率 ${fmt(top.probability, 1)} × 影响 ${fmt(top.impact, 1)})。${escalated.length ? `已升级 ${escalated.length} 项(${escalated.map(r => r.name_zh).join('、')}),需管理层跟踪。` : ''}`,
points: risks.slice(0, 4).map(r => ({
label: r.name_zh,
text: `概率 ${fmt(r.probability, 1)} × 影响 ${fmt(r.impact, 1)} · ${r.status_zh}`,
tone: (r.probability * r.impact) >= 0.4 ? 'bad' as const : 'warn' as const,
})),
actions: [{ label: '打开风险敞口指数', code: 'RK01' }],
source: SRC_BIZ,
};
}
/** 兜底:能力引导 */
function answerFallback(_parsed: ParsedAsk, _ctx: AnswerCtx): AiReply {
const samples: [string, string][] = ([
['指标查询', '毛利率现在多少?'],
['归因分析', '毛利率为什么低于目标?'],
['趋势预测', '预测下季度营业收入'],
['假设模拟', '如果高毛利产品占比提升到 40% 会怎样'],
['对比分析', '各产品线毛利率对比'],
]);
return {
conclusion: '这个问题我暂时没有理解。围绕经营指标,我可以帮你做这些:',
points: samples.map(([cap, q]) => ({ label: cap, text: q })),
actions: samples.map(([cap, q]) => ({ label: cap, ask: q })),
source: SRC_METRIC,
};
}
/* ── 主入口 ─────────────────────────────────────────────────────────────── */
export function buildAnswer(parsed: ParsedAsk, ctx: AnswerCtx): AiReply {
switch (parsed.intent) {
case 'query': return answerQuery(parsed, ctx);
case 'attribute': return answerAttribute(parsed, ctx);
case 'forecast': return answerForecast(parsed, ctx);
case 'whatif': return answerWhatIf(parsed, ctx);
case 'compare': return answerCompare(parsed, ctx);
case 'anomaly': return answerAnomaly(parsed, ctx);
case 'escalate': return answerEscalate(parsed, ctx);
case 'advice': return answerAdvice(parsed, ctx);
case 'overview': return answerOverview(parsed, ctx);
case 'risk': return answerRisk(parsed, ctx);
default: return answerFallback(parsed, ctx);
}
}