guanwei 1.1.1

package/server/src/services/hourInference.ts

@@ -0,0 +1,150 @@
+// 时辰反推引擎：给定出生年月日 + 关键人生事件，对候选时辰按「流年 × 时柱互动」打分排序
+// 原理：大运/起运不受时辰影响（已实测验证），区分候选时辰的关键在时柱与流年的冲合刑害、时干伏吟
+import { baziCalc, shishen } from '../../../shared/core/engine/bazi.js';
+import { GAN, ZHI, jiaziIndex } from '../../../shared/core/data/ganzhi.js';
+import type { GeoLocation } from '../../../shared/core/types';
+
+export interface HourInferEvent {
+  year: number;
+  text: string;
+  /** 事件类别（可选）：health 健康/手术/伤病；love 恋爱/分手/结婚；job 事业/辞职/入职/公司；
+   *  family 父母/家庭变故；money 财务/赔偿；study 学业/升学；move 搬家/远行。缺省时从文本关键词推断 */
+  type?: 'health' | 'love' | 'job' | 'family' | 'money' | 'study' | 'move' | 'breakup';
+}
+export interface HourInferHit { year: number; text: string; reason: string }
+export interface HourInferCandidate {
+  hourIndex: number;      // 0-11
+  hourGZ: string;         // 时柱干支
+  shichen: string;        // 时辰名（子丑…）
+  score: number;
+  hits: HourInferHit[];
+}
+export interface HourInferResult {
+  best: HourInferCandidate;
+  candidates: HourInferCandidate[];
+  chart: { yearGZ: string; monthGZ: string; dayGZ: string };
+  note: string;
+}
+
+const SHICHEN = ['子', '丑', '寅', '卯', '辰', '巳', '午', '未', '申', '酉', '戌', '亥'];
+
+// ─── 地支关系 ───
+const CHONG: Record<string, string> = { 子: '午', 午: '子', 丑: '未', 未: '丑', 寅: '申', 申: '寅', 卯: '酉', 酉: '卯', 辰: '戌', 戌: '辰', 巳: '亥', 亥: '巳' };
+const LIUHE: Record<string, string> = { 子: '丑', 丑: '子', 寅: '亥', 亥: '寅', 卯: '戌', 戌: '卯', 辰: '酉', 酉: '辰', 巳: '申', 申: '巳', 午: '未', 未: '午' };
+const SANHE_GROUP: Record<string, string> = { 申: '水', 子: '水', 辰: '水', 寅: '火', 午: '火', 戌: '火', 巳: '金', 酉: '金', 丑: '金', 亥: '木', 卯: '木', 未: '木' };
+const PO: Record<string, string> = { 子: '酉', 酉: '子', 丑: '辰', 辰: '丑', 寅: '亥', 亥: '寅', 卯: '午', 午: '卯', 巳: '申', 申: '巳', 未: '戌', 戌: '未' };
+const XING: Record<string, string> = { 寅: '巳', 巳: '申', 申: '寅', 丑: '戌', 戌: '未', 未: '丑', 子: '卯', 卯: '子' };
+
+function liuGZ(year: number): string {
+  const idx = jiaziIndex(year);
+  return GAN[idx % 10] + ZHI[idx % 12];
+}
+
+// 对单个事件打分：返回 (score, reason?)——只记「涉及时柱/日支/月支的应象」理由，供用户理解
+function scoreEvent(ev: HourInferEvent, dayGan: string, hourGZ: string, dayGZ: string, monthGZ: string): { score: number; reason: string } {
+  const gz = liuGZ(ev.year);
+  const lg = gz[0], lz = gz[1];
+  const sz = hourGZ[1], dz = dayGZ[1], mz = monthGZ[1];
+  const t = ev.text || '';
+  const has = (...ks: string[]) => ks.some(k => t.includes(k));
+  // 类别判断：优先用显式 type，其次文本关键词
+  const is = (cat: NonNullable<HourInferEvent['type']>) =>
+    ev.type === cat || (ev.type === undefined && (
+      (cat === 'health' && has('病', '癌', '手术', '住院', '伤', '灾', '去世', '死', '切除', '健康')) ||
+      (cat === 'love' && has('恋', '对象', '男友', '女友', '结婚', '相亲', '遇到', '姻缘')) ||
+      (cat === 'breakup' && has('分手', '破裂', '退婚', '分开')) ||
+      (cat === 'job' && has('辞职', '入职', '工作', '公司', '事业', '创业', '裁员', '解散', '失业', '跳槽', '离职', '外贸', '职场', '晋升', '职位')) ||
+      (cat === 'study' && has('上学', '高考', '大学', '读书', '复学', '学业', '高中', '初中', '升学')) ||
+      (cat === 'move' && has('搬家', '到北京', '外地', '出国', '远行', '北漂', '去了', '变动')) ||
+      (cat === 'money' && has('赔', '破财', '损失', '亏', '没干起来', '财务', '收入', '奖金', '赔偿')) ||
+      (cat === 'family' && has('父母', '妈妈', '爸爸', '家里', '家庭', '离婚', '变故'))
+    ));
+  let s = 0;
+  const rs: string[] = [];
+  const push = (r: string) => { if (!rs.includes(r)) rs.push(r); };
+  const ss = shishen(dayGan, lg);
+
+  // ① 疾病/手术/灾伤/丧亡 → 冲刑破害日支（身体）、时支（子女/下焦）；半三合引动时支（如辰申合水主肾/生殖/下焦）
+  if (is('health')) {
+    if (lz === CHONG[dz]) { s += 2.5; push(lz + '冲日支' + dz + '（身体宫受冲）'); }
+    if (lz === CHONG[sz]) { s += 2.5; push(lz + '冲时支' + sz + '（子女/下焦宫受冲）'); }
+    if (lz === PO[dz]) { s += 2; push(lz + '破日支' + dz); }
+    if (lz === PO[sz]) { s += 2; push(lz + '破时支' + sz); }
+    if (lz === XING[dz]) { s += 1.5; push(lz + '刑日支' + dz); }
+    if (lz === XING[sz]) { s += 1.5; push(lz + '刑时支' + sz); }
+    // 半合引动时支与冲同权：如辰申合水主肾/骨/生殖（肾主骨，恰应骨骼与下焦之疾）
+    if (SANHE_GROUP[lz] && SANHE_GROUP[lz] === SANHE_GROUP[sz]) { s += 3; push(lz + sz + '半三合（' + SANHE_GROUP[lz] + '局，引动时支/对应脏腑）'); }
+  }
+  // ② 恋爱/对象/结婚/相亲（姻缘始）→ 时干伏吟 / 时支六合三合 / 日支（夫妻宫）六合 / 夫星（女命官杀）
+  if (is('love')) {
+    if (lg === hourGZ[0]) { s += 2; push('流年' + lg + '与时干伏吟（引动时柱/姻缘宫）'); }
+    if (lz === LIUHE[sz]) { s += 1.5; push(lz + '与' + sz + '六合（时支被合动）'); }
+    if (lz === LIUHE[dz]) { s += 1.5; push(lz + '与' + dz + '六合（日支/夫妻宫被合动）'); }
+    if (SANHE_GROUP[lz] && SANHE_GROUP[lz] === SANHE_GROUP[sz]) { s += 1.5; push(lz + sz + '半三合（' + SANHE_GROUP[lz] + '局）'); }
+    if (ss === '正官' || ss === '七杀') { s += 1.5; push('流年' + lg + '为' + ss + '（官星现，主姻缘/担当）'); }
+  }
+  // ③ 分手/感情破裂 → 伤官克官 / 冲日支（夫妻宫）
+  if (is('breakup') || (is('love') && has('分手'))) {
+    if (ss === '伤官' || ss === '七杀') { s += 2; push('流年' + lg + '为' + ss + '（克官伤情）'); }
+    if (lz === CHONG[dz]) { s += 2; push(lz + '冲日支' + dz + '（夫妻宫动）'); }
+    if (lz === CHONG[sz]) { s += 1; push(lz + '冲时支' + sz); }
+  }
+  // ④ 事业/职场变动（辞职/入职/换工作/公司/裁员/失业）
+  if (is('job')) {
+    if (lz === CHONG[mz]) { s += 1.5; push(lz + '冲月支' + mz + '（父母/根基宫动）'); }
+    if (lz === CHONG[sz]) { s += 1; push(lz + '冲时支' + sz); }
+    if (lz === sz) { s += 1.5; push('流年支' + lz + '伏吟时支（变动之象）'); }
+    if (ss === '伤官') { s += 1; push('流年' + lg + '为伤官（伤官见官，主与体制/上司冲突）'); }
+    if (ss === '正官' || ss === '七杀') { s += 1; push('流年' + lg + '为' + ss + '（官星动，主职场体制变动）'); }
+  }
+  // ⑤ 学业/考试
+  if (is('study')) {
+    if (ss === '正印' || ss === '偏印') { s += 1.5; push('流年' + lg + '为' + ss + '（印星主学业）'); }
+    if (ss === '食神' || ss === '伤官') { s += 1; push('流年' + lg + '为' + ss + '（食伤泄秀利学习输出）'); }
+  }
+  // ⑥ 搬家/远行/外地
+  if (is('move')) {
+    if (lz === CHONG[sz]) { s += 1.5; push(lz + '冲时支' + sz + '（远行之象）'); }
+    if (lz === CHONG[mz]) { s += 1; push(lz + '冲月支' + mz); }
+    if (lz === sz) { s += 1.5; push('流年支' + lz + '伏吟时支（变动之象）'); }
+  }
+  // ⑦ 破财/损失
+  if (is('money')) {
+    if (ss === '劫财') { s += 2; push('流年' + lg + '为劫财（夺财之象）'); }
+  }
+  // ⑧ 父母/家庭变故
+  if (is('family')) {
+    if (lz === CHONG[mz]) { s += 2; push(lz + '冲月支' + mz + '（父母宫动）'); }
+  }
+
+  return { score: s, reason: rs.join('；') };
+}
+
+export function inferHour(input: { y: number; m: number; d: number; gender: '男' | '女'; location?: GeoLocation | null; candidates?: number[]; events: HourInferEvent[] }): HourInferResult {
+  const { y, m, d, gender, location, events } = input;
+  // 默认全 12 时辰候选（无真太阳时校正：候选按「时辰序」理解，前端可提示用户按钟表时辰选择）
+  const candidates = input.candidates && input.candidates.length ? input.candidates : [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11];
+  // 基准盘（时辰未知模式排年月日三柱，用于取 dayGan 与三柱）
+  const base = baziCalc({ y, m, d, hourIndex: -1, gender, location: location || undefined });
+
+  const rows: HourInferCandidate[] = candidates.map(hourIndex => {
+    const r = baziCalc({ y, m, d, hourIndex, gender, location: location || undefined });
+    let score = 0;
+    const hits: HourInferHit[] = [];
+    for (const ev of events) {
+      const { score: es, reason } = scoreEvent(ev, base.dayGan, r.hourGZ, r.dayGZ, r.monthGZ);
+      score += es;
+      if (reason) hits.push({ year: ev.year, text: ev.text, reason });
+    }
+    return { hourIndex, hourGZ: r.hourGZ, shichen: SHICHEN[hourIndex], score: Math.round(score * 10) / 10, hits };
+  });
+
+  rows.sort((a, b) => b.score - a.score);
+  const best = rows[0];
+  return {
+    best,
+    candidates: rows,
+    chart: { yearGZ: base.yearGZ, monthGZ: base.monthGZ, dayGZ: base.dayGZ },
+    note: '按「流年 × 时柱」冲合刑害与时干伏吟应象打分：疾病手术看冲刑日/时支，姻缘看时干伏吟与六合三合，事业变动看冲月支，父母变故看冲月支。大运序列不受时辰影响，本推演只区分时柱细节。',
+  };
+}

package/server/src/services/llmProvider.ts

@@ -0,0 +1,181 @@
+// LLM Provider 适配层：Gemini / Groq / DeepSeek / Qwen / 自定义 OpenAI 兼容端点
+// Key 只存服务端环境变量，绝不进入前端
+
+export interface ChatMessage {
+  role: 'system' | 'user' | 'assistant';
+  content: string;
+}
+
+export type ProviderId = 'gemini' | 'groq' | 'deepseek' | 'qwen' | 'custom';
+
+interface ProviderConfig {
+  id: ProviderId;
+  label: string;
+  keyEnv: string;
+  endpoint?: string;
+  modelEnv?: string;
+  defaultModel: string;
+  // OpenAI 兼容（chat/completions）还是 Google 格式
+  kind: 'openai' | 'google';
+}
+
+const PROVIDERS: ProviderConfig[] = [
+  { id: 'gemini', label: 'Gemini', keyEnv: 'LLM_GEMINI_KEY', modelEnv: 'LLM_GEMINI_MODEL', defaultModel: 'gemini-2.0-flash', kind: 'google' },
+  { id: 'groq', label: 'Groq', keyEnv: 'LLM_GROQ_KEY', modelEnv: 'LLM_GROQ_MODEL', defaultModel: 'llama-3.3-70b-versatile', kind: 'openai', endpoint: 'https://api.groq.com/openai/v1/chat/completions' },
+  { id: 'deepseek', label: 'DeepSeek', keyEnv: 'LLM_DEEPSEEK_KEY', modelEnv: 'LLM_DEEPSEEK_MODEL', defaultModel: 'deepseek-chat', kind: 'openai', endpoint: 'https://api.deepseek.com/v1/chat/completions' },
+  { id: 'qwen', label: '通义千问', keyEnv: 'LLM_QWEN_KEY', modelEnv: 'LLM_QWEN_MODEL', defaultModel: 'qwen-turbo', kind: 'openai', endpoint: 'https://dashscope.aliyuncs.com/compatible-mode/v1/chat/completions' },
+  { id: 'custom', label: '自定义端点', keyEnv: 'LLM_CUSTOM_KEY', modelEnv: 'LLM_CUSTOM_MODEL', defaultModel: 'gpt-4o-mini', kind: 'openai' },
+];
+
+export function activeProvider(): ProviderConfig | null {
+  const env = process.env.LLM_PROVIDER as ProviderId | undefined;
+  if (env) {
+    const p = PROVIDERS.find(x => x.id === env);
+    if (p && process.env[p.keyEnv]) return p;
+  }
+  // 自动探测：第一个有 Key 的 Provider
+  return PROVIDERS.find(p => process.env[p.keyEnv]) || null;
+}
+
+export function providerStatus(): { id: string; label: string; configured: boolean; model: string }[] {
+  return PROVIDERS.map(p => ({
+    id: p.id,
+    label: p.label,
+    configured: !!process.env[p.keyEnv],
+    model: process.env[p.modelEnv || ''] || p.defaultModel,
+  }));
+}
+
+async function callOpenAI(cfg: ProviderConfig, messages: ChatMessage[], stream: boolean): Promise<Response> {
+  const endpoint = cfg.endpoint || process.env.LLM_CUSTOM_ENDPOINT || 'https://api.openai.com/v1/chat/completions';
+  return fetch(endpoint, {
+    method: 'POST',
+    headers: {
+      'Content-Type': 'application/json',
+      'Authorization': 'Bearer ' + process.env[cfg.keyEnv],
+    },
+    body: JSON.stringify({
+      model: process.env[cfg.modelEnv || ''] || cfg.defaultModel,
+      messages,
+      stream,
+      // 温度：偏低以保证解读稳定自洽（同一盘面多次解读不互相矛盾、不编造盘面事实）；留有余地避免呆板
+      temperature: Number(process.env.LLM_TEMPERATURE || 0.35),
+      // 输出上限：长报告（原始解读+性格+阶段+三域+建议）约需 3-6K tokens
+      max_tokens: Number(process.env.LLM_MAX_TOKENS || 8192),
+      // 关闭思考模式：pro 模型默认思考，reasoning_content 会占满输出预算导致正文截断（finish_reason: length）
+      ...(cfg.id === 'deepseek' ? { thinking: { type: 'disabled' } } : {}),
+      // 强制 JSON 输出（DeepSeek/Groq/Qwen 及 OpenAI 兼容端点均支持）
+      response_format: { type: 'json_object' },
+    }),
+  });
+}
+
+async function callGoogle(cfg: ProviderConfig, messages: ChatMessage[], stream: boolean): Promise<Response> {
+  const model = process.env[cfg.modelEnv || ''] || cfg.defaultModel;
+  const endpoint = `https://generativelanguage.googleapis.com/v1beta/models/${model}:generateContent?key=${process.env[cfg.keyEnv]}`;
+  // 合并 messages 为 Google 格式：system → systemInstruction，其余 → contents
+  const system = messages.filter(m => m.role === 'system').map(m => m.content).join('\n\n');
+  const contents = messages.filter(m => m.role !== 'system').map(m => ({
+    role: m.role === 'assistant' ? 'model' : 'user',
+    parts: [{ text: m.content }],
+  }));
+  return fetch(endpoint, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      systemInstruction: system ? { parts: [{ text: system }] } : undefined,
+      contents,
+      ...(stream ? { stream: true } : {}),
+      generationConfig: { temperature: 0.8, responseMimeType: 'application/json' },
+    }),
+  });
+}
+
+// 瞬时错误重试（429/5xx/网络），最多 2 次
+async function withRetry<T>(fn: () => Promise<T>, label: string): Promise<T> {
+  let lastErr: any;
+  for (let attempt = 0; attempt < 3; attempt++) {
+    try {
+      return await fn();
+    } catch (e: any) {
+      lastErr = e;
+      if (!/AI_HTTP_(429|5\d\d|400)/.test(e.message || '')) throw e;
+      if (attempt < 2) {
+        console.warn('[llm] ' + label + ' 第 ' + (attempt + 1) + ' 次失败(' + e.message + ')，重试…');
+        await new Promise(r => setTimeout(r, 1500 * (attempt + 1)));
+      }
+    }
+  }
+  throw lastErr;
+}
+
+// 非流式调用：返回完整文本
+export async function chatOnce(messages: ChatMessage[]): Promise<string> {
+  const cfg = activeProvider();
+  if (!cfg) throw new Error('AI_UNCONFIGURED');
+  const res = await withRetry(async () => {
+    const r0 = cfg.kind === 'google' ? await callGoogle(cfg, messages, false) : await callOpenAI(cfg, messages, false);
+    if (!r0.ok) {
+      const body = await r0.text().catch(() => '');
+      console.error('[llm] HTTP ' + r0.status + ' 响应体:', body.slice(0, 500));
+      throw new Error('AI_HTTP_' + r0.status);
+    }
+    return r0;
+  }, 'chatOnce');
+  const data = await res.json();
+  if (cfg.kind === 'google') {
+    const text = data?.candidates?.[0]?.content?.parts?.map((p: any) => p.text).join('') ?? '';
+    return text;
+  }
+  return data?.choices?.[0]?.message?.content ?? '';
+}
+
+// 最近一次流式调用的结束原因（'stop' | 'length' 截断 | null）
+export let lastFinishReason: string | null = null;
+
+// 流式调用：逐块输出文本
+export async function* chatStream(messages: ChatMessage[]): AsyncGenerator<string> {
+  const cfg = activeProvider();
+  if (!cfg) throw new Error('AI_UNCONFIGURED');
+  const res = await withRetry(async () => {
+    const r0 = cfg.kind === 'google' ? await callGoogle(cfg, messages, true) : await callOpenAI(cfg, messages, true);
+    if (!r0.ok) {
+      const body = await r0.text().catch(() => '');
+      console.error('[llm] stream HTTP ' + r0.status + ' 响应体:', body.slice(0, 500));
+      throw new Error('AI_HTTP_' + r0.status);
+    }
+    return r0;
+  }, 'chatStream');
+  lastFinishReason = null;
+  const reader = res.body?.getReader();
+  if (!reader) return;
+  const decoder = new TextDecoder();
+  let buffer = '';
+  while (true) {
+    const { done, value } = await reader.read();
+    if (done) break;
+    buffer += decoder.decode(value, { stream: true });
+    // 按行解析 SSE
+    const lines = buffer.split('\n');
+    buffer = lines.pop() || '';
+    for (const line of lines) {
+      const t = line.trim();
+      if (t.startsWith('data:')) {
+        const payload = t.slice(5).trim();
+        if (payload === '[DONE]') continue;
+        try {
+          const json = JSON.parse(payload);
+          if (cfg.kind === 'google') {
+            const text = json?.candidates?.[0]?.content?.parts?.map((p: any) => p.text).join('') ?? '';
+            if (text) yield text;
+          } else {
+            const choice = json?.choices?.[0];
+            if (choice?.finish_reason) lastFinishReason = choice.finish_reason;
+            const delta = choice?.delta?.content ?? '';
+            if (delta) yield delta;
+          }
+        } catch { /* 忽略非 JSON 行 */ }
+      }
+    }
+  }
+}