dsh-dock 0.11.2 → 0.11.3

package/features/tokenlog/host.js

@@ -1,883 +1,882 @@
-// dsh-dock · 功能模块【用量记录】· 宿主半部（v0.4.0，移植自 @wycto/dsh-token-usage lib/index.js）
-//
-// 记录 DSH 会话日志中的每次 LLM 调用（assistant/message 的 usage），
-// 支持历史全量扫描 + 实时 session/event 增量；RPC 经 webServer HTTP 路由
-//   POST /dsh-dock/tokenlog/query|export|scan|hello|pricing|setpricing
-// 向浏览器半部提供查询/统计/导出，以及单价的读取与保存。
-//
-// 数据模型（标量，无 Host 对象引用）：
-// {
-//   id, time, sessionId, provider, model, apiKey(掩码),
-//   inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens, reasoningTokens,
-//   billedInput, totalTokens, cacheHitPercent, cost, pricingSource, effort, status, llmMs, turn, step
-// }
-//
-// 单价配置保存在 settings 的 dsh-dock 命名空间 tokenlog 段（schema 见 src/host-core.js）：
-//   dsh-dock:
-//     tokenlog:
-//       usdCnyRate: 7.2
-//       fetchOfficial: true
-//       pricing:
-//         - { match: deepseek-v4-flash, input: 1.5, output: 4.5, cacheRead: 0.05, cacheWrite: 0,
-//             peaks: [{ start: 9, end: 14, input: 3, output: 9, cacheRead: 0.1, cacheWrite: 0 }] }
-// 定价优先级：用户自定义单价 > 官网抓取刊例价 > 内置默认表 > 兜底单价。
-// 分时段价支持多段（start>end 跨零点、start=end 全天、小数小时表示半点）；旧单段 peak 自动归一。
-// 费用在查询/导出时按当前单价即时重算，因此改价无需重扫历史。
-import { DOCK_NS, sendJson, readBody } from '../../src/host-core.js'
-
-export const feature = {
-  id: 'tokenlog',
-  name: '用量记录',
-  description: '记录全部 LLM API 调用并统计 Token 用量与花费',
-  defaultEnabled: false,
-  setup(ctx) {
-    const disposers = []
-    const dispose = () => {
-      while (disposers.length > 0) {
-        const fn = disposers.pop()
-        try { if (typeof fn === 'function') fn() } catch { /* 停用清理失败不阻断 */ }
-      }
-    }
-
-    // 用 Map<id, record> 存记录: id = sessionId:seq 天然唯一。
-    // 实时监听与历史扫描存在重叠窗口(活跃会话既被实时 push 又被扫描全量读),
-    // 用 set 幂等去重, 避免同一条 LLM 调用被统计两次(此前用数组 push 导致"两倍")。
-    const records = new Map()
-    let scannedSessions = new Set()
-    const headerCache = new Map()
-    const turnError = new Map()
-    const turnStatus = new Map()
-    const stepStart = new Map()
-
-    function num(v) { return typeof v === 'number' && Number.isFinite(v) ? v : 0 }
-
-    // ===== 定价与金额估算 =====
-    // 定价表以「人民币元 / 百万 tokens」为基准(DeepSeek/智谱官方刊例即人民币), 金额先算 CNY,
-    // 对外记录的 cost 保持 USD，供旧版客户端兼容；当前客户端和 CSV 统一展示人民币。
-    // 单价来源与优先级(resolvePricing):
-    //   1) 用户在面板「单价设置」保存的自定义单价(settings dsh-dock.tokenlog.pricing)——最高优先;
-    //   2) 官网抓取刊例价(fetchOfficial 关闭时不参与);
-    //   3) 内置默认表 DEFAULT_PRICING;
-    //   4) 兜底单价(settings dsh-dock.tokenlog.fallback)。
-    // 费用在查询/导出时即时重算(见 withCost), 所以改价立即生效、无需重扫历史。
-    const DEFAULT_PRICING_URL = 'https://api-docs.deepseek.com/zh-cn/quick_start/pricing'
-    const DEFAULT_FETCH_INTERVAL_HOURS = 24
-    const DEFAULT_FALLBACK = { input: 2.16, output: 6.48, cacheRead: 0.43, cacheWrite: 4.32 }
-    /** 计价来源标签(随记录下发, 供界面标注)。 */
-    const PRICING_SOURCE_LABELS = { custom: '自定义', official: '官网', builtin: '内置', fallback: '兜底' }
-    let usdCnyRate = 7.2
-    let fallbackPricing = { match: '*', ...DEFAULT_FALLBACK }
-    // 内置定价表(人民币元 / 百万 tokens), 按 model 子串匹配; 未匹配走 fallback。
-    // 支持分时段定价: 条目可带 peaks=[{start,end,input,output,cacheRead,cacheWrite}, ...],
-    // 命中某段(插件运行机器本地时间)用该段价, 其余时段用基准价; 支持多段与跨零点(start>end)。
-    // 官方来源: DeepSeek-V4 系列 2026-08-17 生效的分时段刊例(高峰每日 9:00-14:00, 空闲为高峰一半)。
-    const DEFAULT_PRICING = [
-      { match: 'deepseek-v4-flash', input: 1.5, output: 4.5, cacheRead: 0.05, cacheWrite: 0,
-        peaks: [{ start: 9, end: 14, input: 3.0, output: 9.0, cacheRead: 0.10, cacheWrite: 0 }] },
-      { match: 'deepseek-v4-pro', input: 4.5, output: 13.5, cacheRead: 0.15, cacheWrite: 0,
-        peaks: [{ start: 9, end: 14, input: 9.0, output: 27.0, cacheRead: 0.30, cacheWrite: 0 }] },
-      { match: 'deepseek-v3', input: 3.6, output: 7.2, cacheRead: 0.72, cacheWrite: 7.2 },
-      { match: 'deepseek-chat', input: 1.94, output: 7.92, cacheRead: 0.5, cacheWrite: 7.92 },
-      { match: 'deepseek-reasoner', input: 3.96, output: 15.77, cacheRead: 1.01, cacheWrite: 7.92 },
-      { match: 'nemotron', input: 1.08, output: 2.88, cacheRead: 0.22, cacheWrite: 2.16 },
-      { match: 'glm', input: 2.88, output: 8.64, cacheRead: 0.58, cacheWrite: 7.2 },
-    ]
-    let pricingTable = DEFAULT_PRICING
-    // 用户自定义单价(最高优先)与官网抓取开关/地址——均由 loadPricingConfig() 从 settings 刷新。
-    let userPricing = []
-    let fetchOfficialEnabled = true
-    let pricingUrl = DEFAULT_PRICING_URL
-    let fetchIntervalMs = DEFAULT_FETCH_INTERVAL_HOURS * 3600000
-
-    function normalizePricing(match, e) {
-      const n = (x) => (typeof x === 'number' && Number.isFinite(x) ? x : 0)
-      const row = { match, input: n(e.input), output: n(e.output), cacheRead: n(e.cacheRead), cacheWrite: n(e.cacheWrite) }
-      // 分时段价支持多段: peaks=[{start,end,input,output,cacheRead,cacheWrite}, ...]。
-      // 兼容旧的单段写法 peak={...}(以及旧版本存盘数据), 统一归一到 peaks。
-      const src = Array.isArray(e.peaks) && e.peaks.length ? e.peaks : (e.peak && typeof e.peak === 'object' ? [e.peak] : [])
-      const peaks = []
-      for (const s of src) {
-        if (!s || typeof s !== 'object') continue
-        peaks.push({
-          start: n(s.start), end: n(s.end),
-          input: n(s.input), output: n(s.output),
-          cacheRead: n(s.cacheRead), cacheWrite: n(s.cacheWrite),
-        })
-      }
-      if (peaks.length) row.peaks = peaks
-      return row
-    }
-
-    /** 浅拷贝分时段数组（RPC 下发与落盘都用它，避免调用方改到内存态）。 */
-    function clonePeaks(peaks) {
-      return Array.isArray(peaks) ? peaks.map((s) => Object.assign({}, s)) : []
-    }
-
-    // 读 settings 的 dsh-dock.tokenlog 段并刷新模块级定价状态。幂等, 可重复调用。
-    // 取值做防御性归一(schema 已给默认, 但 settings 未挂载/异常时仍需兜底)。
-    function loadPricingConfig() {
-      try {
-        const settings = ctx.get('settings')
-        const root = settings && typeof settings.get === 'function' ? settings.get(DOCK_NS) : null
-        const t = root && typeof root === 'object' && root.tokenlog && typeof root.tokenlog === 'object' ? root.tokenlog : null
-        if (!t) return
-        if (typeof t.usdCnyRate === 'number' && t.usdCnyRate > 0) usdCnyRate = t.usdCnyRate
-        if (typeof t.fetchOfficial === 'boolean') fetchOfficialEnabled = t.fetchOfficial
-        if (typeof t.pricingUrl === 'string' && t.pricingUrl) pricingUrl = t.pricingUrl
-        if (typeof t.pricingFetchIntervalHours === 'number' && t.pricingFetchIntervalHours > 0) {
-          fetchIntervalMs = t.pricingFetchIntervalHours * 3600000
-        }
-        userPricing = Array.isArray(t.pricing)
-          ? t.pricing
-            .filter((e) => e && typeof e === 'object' && typeof e.match === 'string' && e.match)
-            .map((e) => normalizePricing(String(e.match).toLowerCase(), e))
-          : []
-        if (t.fallback && typeof t.fallback === 'object') fallbackPricing = normalizePricing('*', t.fallback)
-      } catch (e) {
-        console.error('[dsh-dock] tokenlog read pricing config failed', e && e.message)
-      }
-    }
-
-    // 分时段命中: start<end 为普通区间 [start,end); start>end 表示跨零点(如 23~7 点);
-    // start===end 视为全天。多段命中时取第一段(按配置顺序)。
-    function pickPeakSegment(peaks, hour) {
-      if (!Array.isArray(peaks) || hour < 0) return null
-      for (const s of peaks) {
-        const a = Number(s.start), b = Number(s.end)
-        if (!Number.isFinite(a) || !Number.isFinite(b)) continue
-        const hit = a === b ? true : (a < b ? (hour >= a && hour < b) : (hour >= a || hour < b))
-        if (hit) return s
-      }
-      return null
-    }
-
-    // 命中定价行: 用户自定义 → 官网抓取(可关闭) → 内置表 → 兜底; 再按调用时刻取分时价。
-    // 官网获取的数据若为 USD 则自动折成 CNY(内置表已是 CNY)。返回单价(元/百万 tokens)与来源标签。
-    function resolvePricing(model, timeMs) {
-      const m = String(model || '').toLowerCase()
-      let row = null
-      let source = 'fallback'
-      // 1) 用户自定义单价: 最高优先(官网抓取价常与实际计费不符, 用户填了就以其为准)
-      for (const r of userPricing) {
-        if (m.includes(r.match)) { row = r; source = 'custom'; break }
-      }
-      // 2) 官网自动获取的定价(48h 内有效, 超时回退内置; 可在设置里关闭抓取)
-      if (!row && fetchOfficialEnabled) {
-        const fetched = fetchedPricing
-        if (fetched && fetched.length > 0 && (Date.now() - lastFetchTime < fetchIntervalMs * 2)) {
-          for (const r of fetched) {
-            if (m.includes(r.match)) { row = r; source = 'official'; break }
-          }
-        }
-      }
-      const fromFetch = source === 'official'
-      // 3) 内置表
-      if (!row) {
-        for (const r of pricingTable) {
-          if (m.includes(r.match)) { row = r; source = 'builtin'; break }
-        }
-      }
-      // 4) 兜底
-      if (!row) { row = fallbackPricing; source = 'fallback' }
-      // 分时段判断(本地时间, 支持多段与跨零点; 用小数小时以便 08:30 这类半点边界); 未命中用基准价
-      let h = -1
-      if (typeof timeMs === 'number' && Number.isFinite(timeMs)) {
-        const d = new Date(timeMs)
-        h = d.getHours() + d.getMinutes() / 60
-      }
-      const seg = pickPeakSegment(row.peaks, h)
-      const prices = seg
-        ? { input: seg.input, output: seg.output, cacheRead: seg.cacheRead, cacheWrite: seg.cacheWrite }
-        : { input: row.input, output: row.output, cacheRead: row.cacheRead, cacheWrite: row.cacheWrite }
-      // 币种转换: 官网抓取的数据若是 USD 则折成 CNY(内置表已是 CNY, 无需转换)
-      if (fromFetch && row.currency === 'usd') {
-        prices.input *= usdCnyRate
-        prices.output *= usdCnyRate
-        prices.cacheRead *= usdCnyRate
-        prices.cacheWrite *= usdCnyRate
-      }
-      return { ...prices, source }
-    }
-
-    // 金额估算: 表为人民币价 → 先算 CNY, 再折成 USD(= CNY / usdCnyRate) 供客户端/CSV 使用。
-    // 返回 { cost(USD), source(计价来源) }。
-    function costOf(usage, model, timeMs) {
-      const p = resolvePricing(model, timeMs)
-      const cny = (
-        num(usage.inputTokens) * p.input +
-        num(usage.cacheReadTokens) * p.cacheRead +
-        num(usage.cacheWriteTokens) * p.cacheWrite +
-        num(usage.outputTokens) * p.output
-      ) / 1e6
-      return { cost: cny / usdCnyRate, source: p.source }
-    }
-
-    // 按当前单价重算一条记录的费用(浅拷贝)：面板改价后无需重扫历史即可即时生效。
-    function withCost(r) {
-      const c = costOf(r, r.model, r.time)
-      return Object.assign({}, r, { cost: c.cost, pricingSource: c.source })
-    }
-
-    // 面板「单价设置」读取：返回当前生效的单价表 + 兜底/汇率/抓取开关 + 已出现过的模型
-    // (含已配置 Provider 的模型, 供用户按模型逐条配置单价; 未配置的模型可手输匹配串)。
-    function readPricingConfig() {
-      const cfg = readConfiguredProviders()
-      const seen = new Set()
-      for (const r of records.values()) if (r.model) seen.add(r.model)
-      for (const ids of Object.values(cfg)) for (const id of ids) if (id) seen.add(id)
-      // 每个模型的当前计价来源(供面板提示哪些模型走了兜底价, 提示用户去配单价)
-      const sourcesByModel = {}
-      for (const m of seen) sourcesByModel[m] = PRICING_SOURCE_LABELS[resolvePricing(m, Date.now()).source] || ''
-      return {
-        usdCnyRate,
-        fetchOfficial: fetchOfficialEnabled,
-        pricingUrl,
-        pricingFetchIntervalHours: Math.round(fetchIntervalMs / 3600000),
-        pricing: userPricing.map((r) => Object.assign({}, r, { peaks: clonePeaks(r.peaks) })),
-        fallback: Object.assign({}, fallbackPricing),
-        builtin: pricingTable.map((r) => ({ match: r.match, input: r.input, output: r.output, cacheRead: r.cacheRead, cacheWrite: r.cacheWrite, peaks: clonePeaks(r.peaks) })),
-        models: [...seen].sort(),
-        sourcesByModel,
-      }
-    }
-
-    // 保存面板提交的单价配置: 校验 → 合并进 dsh-dock.tokenlog(保留 pricingUrl 等未在面板暴露
-    // 的字段) → 落盘。保存后由调用方 loadPricingConfig() 刷新内存, 使改价立即影响费用估算。
-    async function savePricingConfig(input) {
-      const settings = ctx.get('settings')
-      if (!settings || typeof settings.get !== 'function' || typeof settings.mutate !== 'function') {
-        throw new Error('settings 服务不可用，单价无法持久化')
-      }
-      const bad = (msg) => { const err = new Error(msg); err.statusCode = 400; return err }
-      const nonNeg = (v, label) => {
-        const n = Number(v)
-        if (!Number.isFinite(n) || n < 0) throw bad(`「${label}」需为不小于 0 的数字`)
-        return n
-      }
-      const root = settings.get(DOCK_NS)
-      const current = root && typeof root === 'object' && root.tokenlog && typeof root.tokenlog === 'object' ? root.tokenlog : {}
-      const next = Object.assign({}, current)
-      if (input.usdCnyRate !== undefined) {
-        const r = Number(input.usdCnyRate)
-        if (!Number.isFinite(r) || r <= 0 || r > 1000) throw bad('「USD→CNY 汇率」需为 0~1000 的正数')
-        next.usdCnyRate = r
-      }
-      if (input.fetchOfficial !== undefined) next.fetchOfficial = !!input.fetchOfficial
-      if (input.pricing !== undefined) {
-        if (!Array.isArray(input.pricing)) throw bad('「模型单价」需为数组')
-        const rows = []
-        for (const raw of input.pricing) {
-          if (!raw || typeof raw !== 'object' || Array.isArray(raw)) throw bad('模型单价条目格式不符')
-          const match = String(raw.match || '').trim().toLowerCase()
-          if (!match) throw bad('模型单价条目的「模型匹配」不能为空')
-          if (match === '*') throw bad('模型匹配不能是 *；未匹配模型请填「兜底单价」')
-          const row = {
-            match,
-            input: nonNeg(raw.input, `${match} 输入单价`),
-            output: nonNeg(raw.output, `${match} 输出单价`),
-            cacheRead: nonNeg(raw.cacheRead, `${match} 缓存命中单价`),
-            cacheWrite: nonNeg(raw.cacheWrite, `${match} 缓存写入单价`),
-          }
-          // 分时段价(可多段): 每段 start/end 为 0~24 的本地小时(支持 8.5 表示 08:30);
-          // start>end 表示跨零点, start===end 表示全天。
-          const rawPeaks = Array.isArray(raw.peaks) ? raw.peaks : []
-          const peaks = []
-          for (const seg of rawPeaks) {
-            if (!seg || typeof seg !== 'object' || Array.isArray(seg)) throw bad(`「${match}」分时段条目格式不符`)
-            const start = Number(seg.start), end = Number(seg.end)
-            if (!Number.isFinite(start) || !Number.isFinite(end) || start < 0 || start > 24 || end < 0 || end > 24) {
-              throw bad(`「${match}」分时段时刻需为 0~24 的小时数`)
-            }
-            peaks.push({
-              start, end,
-              input: nonNeg(seg.input, `${match} 分时段输入单价`),
-              output: nonNeg(seg.output, `${match} 分时段输出单价`),
-              cacheRead: nonNeg(seg.cacheRead, `${match} 分时段缓存命中单价`),
-              cacheWrite: nonNeg(seg.cacheWrite, `${match} 分时段缓存写入单价`),
-            })
-          }
-          if (peaks.length) row.peaks = peaks
-          rows.push(row)
-        }
-        next.pricing = rows
-      }
-      if (input.fallback !== undefined) {
-        const f = input.fallback
-        if (!f || typeof f !== 'object' || Array.isArray(f)) throw bad('「兜底单价」格式不符')
-        next.fallback = {
-          input: nonNeg(f.input, '兜底输入单价'),
-          output: nonNeg(f.output, '兜底输出单价'),
-          cacheRead: nonNeg(f.cacheRead, '兜底缓存命中单价'),
-          cacheWrite: nonNeg(f.cacheWrite, '兜底缓存写入单价'),
-        }
-      }
-      try {
-        await settings.mutate(DOCK_NS, [{ op: 'set', path: ['tokenlog'], value: next }])
-      } catch (e) {
-        const err = new Error('保存单价被拒绝：' + ((e && e.message) || String(e)))
-        err.statusCode = 400
-        throw err
-      }
-      console.log('[dsh-dock] tokenlog pricing saved: rows=' + (Array.isArray(next.pricing) ? next.pricing.length : 0) +
-        ', fetchOfficial=' + next.fetchOfficial)
-      return next
-    }
-
-    // ===== 官网定价自动获取 =====
-    // 每 N 小时(默认 24)从官网抓取最新定价并解析 HTML 表格; 抓取/解析失败时使用内置默认值。
-    // 用户在面板配置的单价优先于抓取价, 因此抓取仅作兜底; 也可在面板关闭抓取。
-    let fetchedPricing = null  // Array<{ match, input, output, cacheRead, cacheWrite, currency }>
-    let lastFetchTime = 0
-
-    function getPricingFetchConfig() {
-      return { url: pricingUrl, intervalMs: fetchIntervalMs }
-    }
-
-    // 服务端出站请求前的 URL 校验(SSRF 防护): 仅 http/https, 且拒绝本机/环回/私有/保留地址。
-    function isPrivateOrReservedHost(host) {
-      const h = String(host || '').toLowerCase().replace(/^\[|\]$/g, '')
-      if (!h) return true
-      if (h === 'localhost' || h.endsWith('.localhost') || h.endsWith('.local') || h.endsWith('.internal')) return true
-      if (h.includes(':')) { // IPv6 字面量
-        return h === '::' || h === '::1' || h.startsWith('fe80') || h.startsWith('fc') || h.startsWith('fd')
-      }
-      const m = h.match(/^(\d{1,3})\.(\d{1,3})\.(\d{1,3})\.(\d{1,3})$/)
-      if (m) {
-        const a = Number(m[1]), b = Number(m[2])
-        return a === 0 || a === 10 || a === 127
-          || (a === 169 && b === 254)          // link-local
-          || (a === 172 && b >= 16 && b <= 31) // 私网
-          || (a === 192 && b === 168)          // 私网
-          || a >= 224                          // 组播/保留
-      }
-      return false
-    }
-
-    function assertAllowedFetchUrl(raw) {
-      let u
-      try { u = new URL(String(raw)) } catch { throw new Error('价格页地址不是合法 URL：' + raw) }
-      if (u.protocol !== 'http:' && u.protocol !== 'https:') throw new Error('价格页地址仅支持 http/https')
-      if (isPrivateOrReservedHost(u.hostname)) throw new Error('价格页地址不允许指向本机/内网/保留地址')
-      return u.toString()
-    }
-
-    // 从 HTML 中提取所有 <table> 的行列数据
-    function extractHTMLTables(html) {
-      const tables = []
-      const tableRe = /<table[^>]*>([\s\S]*?)<\/table>/gi
-      let tm
-      while ((tm = tableRe.exec(html)) !== null) {
-        const rows = []
-        const trRe = /<tr[^>]*>([\s\S]*?)<\/tr>/gi
-        let tr
-        while ((tr = trRe.exec(tm[1])) !== null) {
-          const cells = []
-          const tdRe = /<t[hd][^>]*>([\s\S]*?)<\/t[hd]>/gi
-          let td
-          while ((td = tdRe.exec(tr[1])) !== null) {
-            cells.push(td[1].replace(/<[^>]+>/g, '').replace(/&[a-z]+;/g, ' ').trim())
-          }
-          if (cells.length > 0) rows.push(cells)
-        }
-        if (rows.length >= 2) tables.push(rows)
-      }
-      return tables
-    }
-
-    // 从价格字符串提取数字(支持 "$0.27", "0.27", "¥1.5", "1.5元", "1,500" 等格式)
-    function parsePrice(s) {
-      if (!s) return null
-      const cleaned = String(s).replace(/,/g, '').replace(/￥|¥|元|\/.*$/g, '').trim()
-      const m = cleaned.match(/(\d+\.?\d*)/)
-      return m ? parseFloat(m[1]) : null
-    }
-
-    // 识别定价表列: 通过表头关键词定位各列
-    function identifyPricingColumns(headers) {
-      const h = headers.map(x => x.toLowerCase())
-      const find = (...patterns) => {
-        for (const p of patterns) {
-          const idx = h.findIndex(x => x.includes(p))
-          if (idx >= 0) return idx
-        }
-        return -1
-      }
-      return {
-        model: find('模型', 'model', '名称', 'name'),
-        inputMiss: find('缓存未命中', '未缓存', 'cache miss', 'uncached'),
-        inputHit: find('缓存命中', 'cache hit', 'cached'),
-        output: find('输出', 'output'),
-        peak: find('高峰', 'peak', 'busy', 'on-peak'),
-      }
-    }
-
-    // 从 HTML 解析 DeepSeek 官网定价
-    // 策略: 提取所有 <table>, 找到含"价格/缓存/输入/输出"关键词的表, 逐行提取模型名+价格;
-    // 识别 $ 标记区分 USD/CNY 币种; 无法解析时返回空数组(回退内置默认)。
-    function parsePricingFromHTML(html) {
-      const tables = extractHTMLTables(html)
-      const results = []
-      for (const rows of tables) {
-        const headers = rows[0]
-        const cols = identifyPricingColumns(headers)
-        // 至少需要模型名 + 一个价格列
-        if (cols.model < 0) continue
-        if (cols.inputMiss < 0 && cols.output < 0 && cols.inputHit < 0) continue
-        // 表头需与定价相关
-        const headerText = headers.join(' ')
-        if (!/价格|price|缓存|cache|输入|input|输出|output|tokens/i.test(headerText)) continue
-        for (let i = 1; i < rows.length; i++) {
-          const row = rows[i]
-          const name = (row[cols.model] || '').trim()
-          if (!name || /^(模型|model|名称|合计|total|---)/i.test(name)) continue
-          const input = cols.inputMiss >= 0 ? parsePrice(row[cols.inputMiss]) : null
-          const output = cols.output >= 0 ? parsePrice(row[cols.output]) : null
-          const cacheHit = cols.inputHit >= 0 ? parsePrice(row[cols.inputHit]) : null
-          if (input === null && output === null && cacheHit === null) continue
-          // 判断币种
-          const allText = row.join(' ')
-          const isUSD = /\$|USD|dollar/i.test(allText)
-          results.push({
-            match: name.toLowerCase(),
-            input: input || 0,
-            output: output || 0,
-            cacheRead: cacheHit || 0,
-            cacheWrite: 0, // DeepSeek 官方不收取缓存写入
-            currency: isUSD ? 'usd' : 'cny',
-          })
-        }
-      }
-      return results
-    }
-
-    // 从官网获取最新定价(异步, 失败静默回退内置默认)
-    async function fetchOfficialPricing() {
-      try {
-        const { url } = getPricingFetchConfig()
-        const safeUrl = assertAllowedFetchUrl(url)
-        console.log('[dsh-dock] tokenlog fetching official pricing from:', safeUrl)
-        const res = await fetch(safeUrl, {
-          headers: { 'User-Agent': 'dsh-dock-tokenlog/0.4', 'Accept': 'text/html,*/*' }
-        })
-        if (!res.ok) throw new Error('HTTP ' + res.status)
-        const html = await res.text()
-        console.log('[dsh-dock] tokenlog pricing page fetched, size:', html.length)
-        if (html.length < 500) {
-          console.warn('[dsh-dock] tokenlog page too small, likely SPA shell — using defaults')
-          return
-        }
-        const parsed = parsePricingFromHTML(html)
-        if (parsed.length > 0) {
-          fetchedPricing = parsed
-          lastFetchTime = Date.now()
-          console.log('[dsh-dock] tokenlog parsed', parsed.length, 'models from official pricing')
-        } else {
-          console.warn('[dsh-dock] tokenlog parsed 0 models from pricing page — check page structure, using defaults')
-        }
-      } catch (e) {
-        console.error('[dsh-dock] tokenlog fetch official pricing failed:', e && e.message, '— using built-in defaults')
-      }
-    }
-
-    // 读取 settings.yaml 的 llm-pi-ai 命名空间, 返回「现有提供商」路由名 -> [模型 id]。
-    // 用途: 前端提供商/模型下拉 = 现有配置 ∪ 历史记录, 去重合并 ——
-    //   - 没有任何记录时仍可选现有提供商;
-    //   - 提供商/模型已从配置删除时, 因有历史记录仍可选中(兼容两者)。
-    function readConfiguredProviders() {
-      const out = {}
-      try {
-        const settings = ctx.get('settings')
-        if (!settings || typeof settings.get !== 'function') return out
-        const sec = settings.get('llm-pi-ai')
-        const providers = sec && typeof sec === 'object' && sec.providers ? sec.providers : {}
-        for (const pid of Object.keys(providers || {})) {
-          const p = providers[pid]
-          if (!p || typeof p !== 'object') { out[pid] = []; continue }
-          const models = Array.isArray(p.models) ? p.models : []
-          const ids = []
-          for (const m of models) {
-            const id = m && typeof m === 'object' ? m.id : m
-            if (typeof id === 'string' && id) ids.push(id)
-          }
-          out[pid] = ids
-        }
-      } catch (e) {
-        console.error('[dsh-dock] tokenlog read configured providers failed', e && e.message)
-      }
-      return out
-    }
-
-    // ---------- 事件采集 ----------
-
-    // 预扫描 turn/end: 先建立 turn -> status/error 映射, 再生成记录
-    // (解决顺序处理导致的状态缺失: assistant/message 先于 turn/end 到达)
-    function preScanTurnEnds(sessionId, events) {
-      for (const e of events) {
-        if (e.type !== 'turn/end') continue
-        const t = e.data && e.data.turn
-        const r = e.data && e.data.reason
-        let status = 'completed'
-        if (r && typeof r === 'object') status = r.kind || 'completed'
-        else if (typeof r === 'string') status = r
-        turnStatus.set(sessionId + ':' + t, status)
-        let errInfo = null
-        if (r && typeof r === 'object' && r.error) {
-          const er = r.error
-          errInfo = {
-            statusCode: typeof er.status === 'number' ? er.status : 0,
-            errorCode: er.code || '',
-            errorMsg: er.message || '',
-          }
-        }
-        turnError.set(sessionId + ':' + t, errInfo)
-      }
-    }
-
-    function ingestEvent(sessionId, event) {
-      if (event.type === 'request/header') {
-        const h = event.data && event.data.header
-        if (h && h.config) {
-          headerCache.set(sessionId, {
-            provider: h.config.provider || '',
-            model: h.config.model || '',
-            effort: h.config.reasoningEffort || '',
-          })
-        }
-      }
-      if (event.type === 'turn/end') {
-        const t = event.data && event.data.turn
-        const r = event.data && event.data.reason
-        let status = 'completed'
-        if (r && typeof r === 'object') status = r.kind || 'completed'
-        else if (typeof r === 'string') status = r
-        turnStatus.set(sessionId + ':' + t, status)
-        let errInfo = null
-        if (r && typeof r === 'object' && r.error) {
-          const e = r.error
-          errInfo = {
-            statusCode: typeof e.status === 'number' ? e.status : 0,
-            errorCode: e.code || '',
-            errorMsg: e.message || '',
-          }
-        }
-        turnError.set(sessionId + ':' + t, errInfo)
-      }
-      if (event.type === 'step/start') {
-        stepStart.set(sessionId + ':' + event.data.turn + ':' + event.data.step, event.time)
-      }
-      if (event.type !== 'assistant/message') return null
-      const d = event.data || {}
-      let usage = d.usage
-      if (!usage || typeof usage !== 'object') usage = { inputTokens: 0, outputTokens: 0 }
-      const head = headerCache.get(sessionId) || { provider: '', model: '', effort: '' }
-      const turn = num(d.turn)
-      const step = num(d.step)
-      const startK = sessionId + ':' + turn + ':' + step
-      const startMs = stepStart.get(startK)
-      const llmMs = typeof startMs === 'number' ? Math.max(0, event.time - startMs) : null
-      const status = turnStatus.get(sessionId + ':' + turn) || 'in-progress'
-      const ei = turnError.get(sessionId + ':' + turn) || null
-      const input = num(usage.inputTokens)
-      const output = num(usage.outputTokens)
-      const cacheRead = num(usage.cacheReadTokens)
-      const cacheWrite = num(usage.cacheWriteTokens)
-      const reasoning = num(usage.reasoningTokens)
-      const priced = costOf(usage, head.model, event.time)
-      stepStart.delete(startK)
-      return {
-        id: sessionId + ':' + event.seq,
-        time: event.time,
-        sessionId,
-        provider: head.provider || '',
-        model: head.model || '',
-        apiKey: '',
-        inputTokens: input,
-        outputTokens: output,
-        cacheReadTokens: cacheRead,
-        cacheWriteTokens: cacheWrite,
-        reasoningTokens: reasoning,
-        billedInput: input + cacheRead + cacheWrite,
-        totalTokens: input + output + cacheRead + cacheWrite,
-        cacheHitPercent: (input + cacheRead) > 0 ? Math.round((cacheRead / (input + cacheRead)) * 100) : 0,
-        cost: priced.cost,
-        pricingSource: priced.source,
-        effort: head.effort || '',
-        status,
-        statusCode: ei ? ei.statusCode : 0,
-        errorCode: ei ? ei.errorCode : '',
-        errorMsg: ei ? ei.errorMsg : '',
-        llmMs,
-        turn,
-        step,
-      }
-    }
-
-    function ingestEvents(sessionId, events) {
-      const out = []
-      for (const e of events) {
-        const r = ingestEvent(sessionId, e)
-        if (r) out.push(r)
-      }
-      return out
-    }
-
-    // 历史扫描: 遍历所有会话, 读取完整日志, 重建索引
-    // 说明: 读取失败的多是旧版(v0→v1 迁移)、中途损坏或缺 seq 的会话——这类日志无法解码是常态,
-    // 不应视为异常。若逐条 console.error 会把终端刷屏, 改为统计失败数并只打印少量样例明细。
-    async function scanHistory() {
-      if (!sessionQuery) return
-      const failures = [] // { sid, msg } 保留前 MAX 条明细供排查
-      const MAX_FAIL_SAMPLES = 3
-      let failCount = 0
-      try {
-        const sessions = await sessionQuery.listSessions()
-        let added = 0
-        for (const s of sessions) {
-          const sid = s && s.header ? s.header.id : (s && s.id)
-          if (!sid || scannedSessions.has(sid)) continue
-          scannedSessions.add(sid)
-          try {
-            const snap = await sessionQuery.readSession(sid)
-            const events = snap && snap.events ? snap.events : []
-            // 两遍: 先建 turn/end 映射, 再生成记录
-            preScanTurnEnds(sid, events)
-            const recs = ingestEvents(sid, events)
-            for (const r of recs) records.set(r.id, r)
-            added += recs.length
-          } catch (e) {
-            failCount += 1
-            if (failures.length < MAX_FAIL_SAMPLES) failures.push(sid + ': ' + ((e && e.message) || String(e)))
-          }
-        }
-        // 失败会话统一汇总为一行(仅列前几条样例), 不逐条刷屏。
-        if (failCount > 0) {
-          console.warn('[dsh-dock] tokenlog 历史扫描: ' + failCount + ' 个会话未能读取' +
-            (failCount > MAX_FAIL_SAMPLES ? ', 前 ' + MAX_FAIL_SAMPLES + ' 条如下:' : ':') )
-          for (const f of failures) console.warn('  - ' + f)
-        }
-        console.log('[dsh-dock] tokenlog history scanned: +' + added + ' records, total ' + records.size)
-      } catch (e) {
-        console.error('[dsh-dock] tokenlog scanHistory failed', e && e.message)
-      }
-    }
-
-    function buildQuery(args) {
-      const q = args || {}
-      const from = typeof q.from === 'number' ? q.from : 0
-      const to = typeof q.to === 'number' ? q.to : Infinity
-      const provider = (q.provider || '').trim()
-      const model = (q.model || '').trim()
-      const status = (q.status || '').trim()
-      const effort = (q.effort || '').trim()
-      const sessionId = (q.sessionId || '').trim()
-      return [...records.values()].filter((r) => {
-        if (r.time < from || r.time > to) return false
-        if (provider && r.provider !== provider) return false
-        if (model && r.model !== model) return false
-        if (status && r.status !== status) return false
-        if (effort && r.effort !== effort) return false
-        if (sessionId && r.sessionId !== sessionId) return false
-        return true
-      })
-    }
-
-    function summarize(list, dim) {
-      const map = new Map()
-      for (const r of list) {
-        const key = r[dim] || '(none)'
-        const e = map.get(key)
-        if (e) {
-          e.calls += 1
-          e.inputTokens += r.inputTokens
-          e.outputTokens += r.outputTokens
-          e.cacheReadTokens += r.cacheReadTokens
-          e.cacheWriteTokens += r.cacheWriteTokens
-          e.reasoningTokens += r.reasoningTokens
-          e.billedInput += r.billedInput
-          e.totalTokens += r.totalTokens
-          e.cost += r.cost
-          if (typeof r.llmMs === 'number') { e.llmMs += r.llmMs; e.timed += 1 }
-        } else {
-          map.set(key, {
-            key, calls: 1,
-            inputTokens: r.inputTokens, outputTokens: r.outputTokens,
-            cacheReadTokens: r.cacheReadTokens, cacheWriteTokens: r.cacheWriteTokens,
-            reasoningTokens: r.reasoningTokens, billedInput: r.billedInput,
-            totalTokens: r.totalTokens, cost: r.cost,
-            llmMs: typeof r.llmMs === 'number' ? r.llmMs : 0,
-            timed: typeof r.llmMs === 'number' ? 1 : 0,
-          })
-        }
-      }
-      return [...map.values()]
-        .map((e) => ({ ...e, cacheHitPct: e.billedInput > 0 ? Math.round((e.cacheReadTokens / e.billedInput) * 100) : 0 }))
-        .sort((a, b) => b.calls - a.calls)
-    }
-
-    function totals(list) {
-      const t = { calls: 0, inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheWriteTokens: 0, reasoningTokens: 0, billedInput: 0, totalTokens: 0, cost: 0, llmMs: 0, timed: 0 }
-      for (const r of list) {
-        t.calls += 1
-        t.inputTokens += r.inputTokens
-        t.outputTokens += r.outputTokens
-        t.cacheReadTokens += r.cacheReadTokens
-        t.cacheWriteTokens += r.cacheWriteTokens
-        t.reasoningTokens += r.reasoningTokens
-        t.billedInput += r.billedInput
-        t.totalTokens += r.totalTokens
-        t.cost += r.cost
-        if (typeof r.llmMs === 'number') { t.llmMs += r.llmMs; t.timed += 1 }
-      }
-      t.cacheHitPct = t.billedInput > 0 ? Math.round((t.cacheReadTokens / t.billedInput) * 100) : 0
-      return t
-    }
-
-    // ---------- RPC（webServer HTTP 路由，前缀 /dsh-dock/tokenlog/） ----------
-
-    // 实时监听 + 历史扫描: 必须等 sessionQuery 服务就绪。
-    // 注意: 不能在 setup 时用 ctx.get('sessionQuery') 软获取——profile 插件无 inject 声明,
-    // 可能早于 sessionQuery 提供方激活, 拿到 undefined 会让 监听+扫描 整条采集链路静默失效。
-    // 改用 ctx.inject(['sessionQuery'], ...): 服务已就绪则立即执行, 否则挂起等待。
-    let sessionQuery = null
-    disposers.push(ctx.inject(['sessionQuery'], (sqCtx) => {
-      sessionQuery = sqCtx.sessionQuery
-      // 服务就绪后加载 settings 单价配置(汇率 + 自定义单价 + 抓取开关), 供采集/重扫/导出使用
-      loadPricingConfig()
-      // 从官网获取最新定价(启动时 + 每 N 小时自动刷新; 面板关闭抓取时不发起请求)
-      if (fetchOfficialEnabled) fetchOfficialPricing()
-      const fetchTimer = setInterval(() => { if (fetchOfficialEnabled) fetchOfficialPricing() }, getPricingFetchConfig().intervalMs)
-      sqCtx.effect(() => () => clearInterval(fetchTimer))
-      // 实时监听: session/event (root scope 可收到所有 session 事件, 与官方 persistence 同款订阅)
-      disposers.push(sqCtx.on('session/event', (session, event) => {
-        try {
-          const sid = typeof session === 'object' && session ? (session.id || '') : String(session || '')
-          const rec = ingestEvent(sid, event)
-          if (rec) records.set(rec.id, rec)
-          // turn/end 到达后回填该 turn 已有记录的状态/错误(实时增量场景)
-          if (event.type === 'turn/end') {
-            const t = event.data && event.data.turn
-            const key = sid + ':' + t
-            const st = turnStatus.get(key)
-            const ei = turnError.get(key) || null
-            if (st) {
-              for (const r of records.values()) {
-                if (r.sessionId === sid && String(r.turn) === String(t)) {
-                  r.status = st
-                  r.statusCode = ei ? ei.statusCode : 0
-                  r.errorCode = ei ? ei.errorCode : ''
-                  r.errorMsg = ei ? ei.errorMsg : ''
-                }
-              }
-            }
-          }
-        } catch (err) {
-          console.error('[dsh-dock] tokenlog ingest error', err)
-        }
-      }))
-      // 服务就绪后立即做一次历史全量扫描
-      scanHistory()
-    }))
-
-    // webServer 就绪后注册 /dsh-dock/tokenlog 前缀路由(ctx.inject 保证就绪后再执行)
-    disposers.push(ctx.inject(['webServer'], (wsCtx) => {
-      wsCtx.effect(() => wsCtx.webServer.register({
-        kind: 'prefix',
-        path: '/dsh-dock/tokenlog',
-        async handler(req, res) {
-          try {
-            const url = new URL(req.url || '/', 'http://dsh.internal')
-            const method = url.pathname.replace(/^\/dsh-dock\/tokenlog\/?/, '').split('/')[0] || ''
-            const payload = await readBody(req)
-            if (method === 'query') {
-              loadPricingConfig()
-              // 费用按当前单价即时重算(withCost): 面板改价后立即反映到统计与明细, 无需重扫历史。
-              const list = buildQuery(payload || {}).map(withCost)
-              const rows = [...list].sort((a, b) => b.time - a.time)
-              const all = [...records.values()]
-              const cfg = readConfiguredProviders()
-              // 提供商/模型下拉 = 现有配置(llm-pi-ai.providers) ∪ 历史记录, 去重排序。
-              // 兼容两种场景: 无记录时仍可选现有提供商; 提供商/模型已删除时因记录仍可选中。
-              const providerSet = new Set(all.map((r) => r.provider).filter(Boolean))
-              const modelSet = new Set(all.map((r) => r.model).filter(Boolean))
-              for (const pid of Object.keys(cfg)) if (pid) providerSet.add(pid)
-              for (const ids of Object.values(cfg)) for (const id of ids) modelSet.add(id)
-              // 提供商 -> 模型 映射(供前端联动筛选: 选中提供商后模型下拉只显示其下模型)
-              const modelsByProvider = {}
-              const initMbp = (p) => { if (!modelsByProvider[p]) modelsByProvider[p] = [] }
-              for (const p of providerSet) initMbp(p)
-              for (const r of all) {
-                if (!r.model) continue
-                const p = r.provider || '(none)'
-                initMbp(p)
-                if (!modelsByProvider[p].includes(r.model)) modelsByProvider[p].push(r.model)
-              }
-              for (const [pid, ids] of Object.entries(cfg)) {
-                initMbp(pid)
-                for (const id of ids) if (!modelsByProvider[pid].includes(id)) modelsByProvider[pid].push(id)
-              }
-              for (const p of Object.keys(modelsByProvider)) modelsByProvider[p].sort()
-              return sendJson(res, 200, {
-                ok: true,
-                data: {
-                  records: rows,
-                  counts: { total: records.size, matching: rows.length },
-                  providers: [...providerSet].sort(),
-                  models: [...modelSet].sort(),
-                  modelsByProvider,
-                  statuses: [...new Set(all.map((r) => r.status).filter(Boolean))],
-                  efforts: [...new Set(all.map((r) => r.effort).filter(Boolean))],
-                  sessionIds: [...new Set(all.map((r) => r.sessionId).filter(Boolean))].sort(),
-                  rateUsdCny: usdCnyRate,
-                  pricingInfo: {
-                    hasCustom: userPricing.length > 0,
-                    customCount: userPricing.length,
-                    fetchOfficial: fetchOfficialEnabled,
-                    sources: [...new Set(list.map((r) => r.pricingSource).filter(Boolean))].map((s) => PRICING_SOURCE_LABELS[s] || s),
-                  },
-                  totals: totals(list),
-                  summary: payload && payload.dim ? summarize(list, payload.dim) : [],
-                }
-              })
-            }
-            if (method === 'pricing') {
-              loadPricingConfig()
-              return sendJson(res, 200, { ok: true, data: readPricingConfig() })
-            }
-            if (method === 'setpricing') {
-              await savePricingConfig(payload || {})
-              loadPricingConfig()
-              return sendJson(res, 200, { ok: true, data: readPricingConfig() })
-            }
-            if (method === 'export') {
-              loadPricingConfig()
-              const list = buildQuery(payload || {}).map(withCost)
-              // 列序对齐界面习惯（缓存命中在前）：… 命中 → 未命中 → 输出 → 写入
-              const header = ['time', 'provider', 'model', 'apiKey', 'cacheReadTokens', 'inputTokens', 'outputTokens', 'cacheWriteTokens', 'reasoningTokens', 'billedInput', 'cacheHitPercent', 'totalTokens', 'costCny', 'effort', 'status', 'statusCode', 'errorCode', 'errorMsg', 'llmMs', 'sessionId', 'turn', 'step']
-              const esc = (v) => {
-                const s = String(v === undefined || v === null ? '' : v)
-                return /[,"\n]/.test(s) ? '"' + s.replace(/"/g, '""') + '"' : s
-              }
-              const lines = [header.map(esc).join(',')]
-              for (const r of list) lines.push(header.map((h) => esc(h === 'costCny' ? ((Number(r.cost) || 0) * usdCnyRate) : r[h])).join(','))
-              return sendJson(res, 200, { ok: true, data: { csv: lines.join('\n'), count: list.length } })
-            }
-            if (method === 'scan') {
-              await scanHistory()
-              return sendJson(res, 200, { ok: true, data: { total: records.size } })
-            }
-            if (method === 'hello') {
-              return sendJson(res, 200, { ok: true, data: { ok: true, records: records.size, scanned: scannedSessions.size } })
-            }
-            return sendJson(res, 404, { ok: false, error: { code: 'method-not-found', message: 'unknown method: ' + method, details: {} } })
-          } catch (e) {
-            // 参数校验类错误按 4xx 返回(如单价非法), 其余为 500
-            const status = e && e.statusCode ? e.statusCode : 500
-            return sendJson(res, status, { ok: false, error: { code: status === 500 ? 'internal' : 'bad-request', message: (e && e.message) || String(e), details: {} } })
-          }
-        }
-      }), 'dsh-dock tokenlog: /dsh-dock/tokenlog HTTP route')
-    }))
-
-    return dispose
-  },
-}
+// dsh-dock · 功能模块【用量记录】· 宿主半部（v0.4.0，移植自 @wycto/dsh-token-usage lib/index.js）
+//
+// 记录 DSH 会话日志中的每次 LLM 调用（assistant/message 的 usage），
+// 支持历史全量扫描 + 实时 session/event 增量；RPC 经 webServer HTTP 路由
+//   POST /dsh-dock/tokenlog/query|export|scan|hello|pricing|setpricing
+// 向浏览器半部提供查询/统计/导出，以及单价的读取与保存。
+// export 返回结构化明细 { rows, rateUsdCny, count }（不是 CSV 文本）：中文表头与取值格式由客户端
+// 按「调用明细」表统一生成，保证导出与页面显示一致；费用在导出时同样按当前单价重算。
+//
+// 数据模型（标量，无 Host 对象引用）：
+// {
+//   id, time, sessionId, provider, model, apiKey(掩码),
+//   inputTokens, outputTokens, cacheReadTokens, cacheWriteTokens, reasoningTokens,
+//   billedInput, totalTokens, cacheHitPercent, cost, pricingSource, effort, status, llmMs, turn, step
+// }
+//
+// 单价配置保存在 settings 的 dsh-dock 命名空间 tokenlog 段（schema 见 src/host-core.js）：
+//   dsh-dock:
+//     tokenlog:
+//       usdCnyRate: 7.2
+//       fetchOfficial: true
+//       pricing:
+//         - { match: deepseek-v4-flash, input: 1.5, output: 4.5, cacheRead: 0.05, cacheWrite: 0,
+//             peaks: [{ start: 9, end: 14, input: 3, output: 9, cacheRead: 0.1, cacheWrite: 0 }] }
+// 定价优先级：用户自定义单价 > 官网抓取刊例价 > 内置默认表 > 兜底单价。
+// 分时段价支持多段（start>end 跨零点、start=end 全天、小数小时表示半点）；旧单段 peak 自动归一。
+// 费用在查询/导出时按当前单价即时重算，因此改价无需重扫历史。
+import { DOCK_NS, sendJson, readBody } from '../../src/host-core.js'
+
+export const feature = {
+  id: 'tokenlog',
+  name: '用量记录',
+  description: '记录全部 LLM API 调用并统计 Token 用量与花费',
+  defaultEnabled: false,
+  setup(ctx) {
+    const disposers = []
+    const dispose = () => {
+      while (disposers.length > 0) {
+        const fn = disposers.pop()
+        try { if (typeof fn === 'function') fn() } catch { /* 停用清理失败不阻断 */ }
+      }
+    }
+
+    // 用 Map<id, record> 存记录: id = sessionId:seq 天然唯一。
+    // 实时监听与历史扫描存在重叠窗口(活跃会话既被实时 push 又被扫描全量读),
+    // 用 set 幂等去重, 避免同一条 LLM 调用被统计两次(此前用数组 push 导致"两倍")。
+    const records = new Map()
+    let scannedSessions = new Set()
+    const headerCache = new Map()
+    const turnError = new Map()
+    const turnStatus = new Map()
+    const stepStart = new Map()
+
+    function num(v) { return typeof v === 'number' && Number.isFinite(v) ? v : 0 }
+
+    // ===== 定价与金额估算 =====
+    // 定价表以「人民币元 / 百万 tokens」为基准(DeepSeek/智谱官方刊例即人民币), 金额先算 CNY,
+    // 对外记录的 cost 保持 USD，供旧版客户端兼容；当前客户端和 CSV 统一展示人民币。
+    // 单价来源与优先级(resolvePricing):
+    //   1) 用户在面板「单价设置」保存的自定义单价(settings dsh-dock.tokenlog.pricing)——最高优先;
+    //   2) 官网抓取刊例价(fetchOfficial 关闭时不参与);
+    //   3) 内置默认表 DEFAULT_PRICING;
+    //   4) 兜底单价(settings dsh-dock.tokenlog.fallback)。
+    // 费用在查询/导出时即时重算(见 withCost), 所以改价立即生效、无需重扫历史。
+    const DEFAULT_PRICING_URL = 'https://api-docs.deepseek.com/zh-cn/quick_start/pricing'
+    const DEFAULT_FETCH_INTERVAL_HOURS = 24
+    const DEFAULT_FALLBACK = { input: 2.16, output: 6.48, cacheRead: 0.43, cacheWrite: 4.32 }
+    /** 计价来源标签(随记录下发, 供界面标注)。 */
+    const PRICING_SOURCE_LABELS = { custom: '自定义', official: '官网', builtin: '内置', fallback: '兜底' }
+    let usdCnyRate = 7.2
+    let fallbackPricing = { match: '*', ...DEFAULT_FALLBACK }
+    // 内置定价表(人民币元 / 百万 tokens), 按 model 子串匹配; 未匹配走 fallback。
+    // 支持分时段定价: 条目可带 peaks=[{start,end,input,output,cacheRead,cacheWrite}, ...],
+    // 命中某段(插件运行机器本地时间)用该段价, 其余时段用基准价; 支持多段与跨零点(start>end)。
+    // 官方来源: DeepSeek-V4 系列 2026-08-17 生效的分时段刊例(高峰每日 9:00-14:00, 空闲为高峰一半)。
+    const DEFAULT_PRICING = [
+      { match: 'deepseek-v4-flash', input: 1.5, output: 4.5, cacheRead: 0.05, cacheWrite: 0,
+        peaks: [{ start: 9, end: 14, input: 3.0, output: 9.0, cacheRead: 0.10, cacheWrite: 0 }] },
+      { match: 'deepseek-v4-pro', input: 4.5, output: 13.5, cacheRead: 0.15, cacheWrite: 0,
+        peaks: [{ start: 9, end: 14, input: 9.0, output: 27.0, cacheRead: 0.30, cacheWrite: 0 }] },
+      { match: 'deepseek-v3', input: 3.6, output: 7.2, cacheRead: 0.72, cacheWrite: 7.2 },
+      { match: 'deepseek-chat', input: 1.94, output: 7.92, cacheRead: 0.5, cacheWrite: 7.92 },
+      { match: 'deepseek-reasoner', input: 3.96, output: 15.77, cacheRead: 1.01, cacheWrite: 7.92 },
+      { match: 'nemotron', input: 1.08, output: 2.88, cacheRead: 0.22, cacheWrite: 2.16 },
+      { match: 'glm', input: 2.88, output: 8.64, cacheRead: 0.58, cacheWrite: 7.2 },
+    ]
+    let pricingTable = DEFAULT_PRICING
+    // 用户自定义单价(最高优先)与官网抓取开关/地址——均由 loadPricingConfig() 从 settings 刷新。
+    let userPricing = []
+    let fetchOfficialEnabled = true
+    let pricingUrl = DEFAULT_PRICING_URL
+    let fetchIntervalMs = DEFAULT_FETCH_INTERVAL_HOURS * 3600000
+
+    function normalizePricing(match, e) {
+      const n = (x) => (typeof x === 'number' && Number.isFinite(x) ? x : 0)
+      const row = { match, input: n(e.input), output: n(e.output), cacheRead: n(e.cacheRead), cacheWrite: n(e.cacheWrite) }
+      // 分时段价支持多段: peaks=[{start,end,input,output,cacheRead,cacheWrite}, ...]。
+      // 兼容旧的单段写法 peak={...}(以及旧版本存盘数据), 统一归一到 peaks。
+      const src = Array.isArray(e.peaks) && e.peaks.length ? e.peaks : (e.peak && typeof e.peak === 'object' ? [e.peak] : [])
+      const peaks = []
+      for (const s of src) {
+        if (!s || typeof s !== 'object') continue
+        peaks.push({
+          start: n(s.start), end: n(s.end),
+          input: n(s.input), output: n(s.output),
+          cacheRead: n(s.cacheRead), cacheWrite: n(s.cacheWrite),
+        })
+      }
+      if (peaks.length) row.peaks = peaks
+      return row
+    }
+
+    /** 浅拷贝分时段数组（RPC 下发与落盘都用它，避免调用方改到内存态）。 */
+    function clonePeaks(peaks) {
+      return Array.isArray(peaks) ? peaks.map((s) => Object.assign({}, s)) : []
+    }
+
+    // 读 settings 的 dsh-dock.tokenlog 段并刷新模块级定价状态。幂等, 可重复调用。
+    // 取值做防御性归一(schema 已给默认, 但 settings 未挂载/异常时仍需兜底)。
+    function loadPricingConfig() {
+      try {
+        const settings = ctx.get('settings')
+        const root = settings && typeof settings.get === 'function' ? settings.get(DOCK_NS) : null
+        const t = root && typeof root === 'object' && root.tokenlog && typeof root.tokenlog === 'object' ? root.tokenlog : null
+        if (!t) return
+        if (typeof t.usdCnyRate === 'number' && t.usdCnyRate > 0) usdCnyRate = t.usdCnyRate
+        if (typeof t.fetchOfficial === 'boolean') fetchOfficialEnabled = t.fetchOfficial
+        if (typeof t.pricingUrl === 'string' && t.pricingUrl) pricingUrl = t.pricingUrl
+        if (typeof t.pricingFetchIntervalHours === 'number' && t.pricingFetchIntervalHours > 0) {
+          fetchIntervalMs = t.pricingFetchIntervalHours * 3600000
+        }
+        userPricing = Array.isArray(t.pricing)
+          ? t.pricing
+            .filter((e) => e && typeof e === 'object' && typeof e.match === 'string' && e.match)
+            .map((e) => normalizePricing(String(e.match).toLowerCase(), e))
+          : []
+        if (t.fallback && typeof t.fallback === 'object') fallbackPricing = normalizePricing('*', t.fallback)
+      } catch (e) {
+        console.error('[dsh-dock] tokenlog read pricing config failed', e && e.message)
+      }
+    }
+
+    // 分时段命中: start<end 为普通区间 [start,end); start>end 表示跨零点(如 23~7 点);
+    // start===end 视为全天。多段命中时取第一段(按配置顺序)。
+    function pickPeakSegment(peaks, hour) {
+      if (!Array.isArray(peaks) || hour < 0) return null
+      for (const s of peaks) {
+        const a = Number(s.start), b = Number(s.end)
+        if (!Number.isFinite(a) || !Number.isFinite(b)) continue
+        const hit = a === b ? true : (a < b ? (hour >= a && hour < b) : (hour >= a || hour < b))
+        if (hit) return s
+      }
+      return null
+    }
+
+    // 命中定价行: 用户自定义 → 官网抓取(可关闭) → 内置表 → 兜底; 再按调用时刻取分时价。
+    // 官网获取的数据若为 USD 则自动折成 CNY(内置表已是 CNY)。返回单价(元/百万 tokens)与来源标签。
+    function resolvePricing(model, timeMs) {
+      const m = String(model || '').toLowerCase()
+      let row = null
+      let source = 'fallback'
+      // 1) 用户自定义单价: 最高优先(官网抓取价常与实际计费不符, 用户填了就以其为准)
+      for (const r of userPricing) {
+        if (m.includes(r.match)) { row = r; source = 'custom'; break }
+      }
+      // 2) 官网自动获取的定价(48h 内有效, 超时回退内置; 可在设置里关闭抓取)
+      if (!row && fetchOfficialEnabled) {
+        const fetched = fetchedPricing
+        if (fetched && fetched.length > 0 && (Date.now() - lastFetchTime < fetchIntervalMs * 2)) {
+          for (const r of fetched) {
+            if (m.includes(r.match)) { row = r; source = 'official'; break }
+          }
+        }
+      }
+      const fromFetch = source === 'official'
+      // 3) 内置表
+      if (!row) {
+        for (const r of pricingTable) {
+          if (m.includes(r.match)) { row = r; source = 'builtin'; break }
+        }
+      }
+      // 4) 兜底
+      if (!row) { row = fallbackPricing; source = 'fallback' }
+      // 分时段判断(本地时间, 支持多段与跨零点; 用小数小时以便 08:30 这类半点边界); 未命中用基准价
+      let h = -1
+      if (typeof timeMs === 'number' && Number.isFinite(timeMs)) {
+        const d = new Date(timeMs)
+        h = d.getHours() + d.getMinutes() / 60
+      }
+      const seg = pickPeakSegment(row.peaks, h)
+      const prices = seg
+        ? { input: seg.input, output: seg.output, cacheRead: seg.cacheRead, cacheWrite: seg.cacheWrite }
+        : { input: row.input, output: row.output, cacheRead: row.cacheRead, cacheWrite: row.cacheWrite }
+      // 币种转换: 官网抓取的数据若是 USD 则折成 CNY(内置表已是 CNY, 无需转换)
+      if (fromFetch && row.currency === 'usd') {
+        prices.input *= usdCnyRate
+        prices.output *= usdCnyRate
+        prices.cacheRead *= usdCnyRate
+        prices.cacheWrite *= usdCnyRate
+      }
+      return { ...prices, source }
+    }
+
+    // 金额估算: 表为人民币价 → 先算 CNY, 再折成 USD(= CNY / usdCnyRate) 供客户端/CSV 使用。
+    // 返回 { cost(USD), source(计价来源) }。
+    function costOf(usage, model, timeMs) {
+      const p = resolvePricing(model, timeMs)
+      const cny = (
+        num(usage.inputTokens) * p.input +
+        num(usage.cacheReadTokens) * p.cacheRead +
+        num(usage.cacheWriteTokens) * p.cacheWrite +
+        num(usage.outputTokens) * p.output
+      ) / 1e6
+      return { cost: cny / usdCnyRate, source: p.source }
+    }
+
+    // 按当前单价重算一条记录的费用(浅拷贝)：面板改价后无需重扫历史即可即时生效。
+    function withCost(r) {
+      const c = costOf(r, r.model, r.time)
+      return Object.assign({}, r, { cost: c.cost, pricingSource: c.source })
+    }
+
+    // 面板「单价设置」读取：返回当前生效的单价表 + 兜底/汇率/抓取开关 + 已出现过的模型
+    // (含已配置 Provider 的模型, 供用户按模型逐条配置单价; 未配置的模型可手输匹配串)。
+    function readPricingConfig() {
+      const cfg = readConfiguredProviders()
+      const seen = new Set()
+      for (const r of records.values()) if (r.model) seen.add(r.model)
+      for (const ids of Object.values(cfg)) for (const id of ids) if (id) seen.add(id)
+      // 每个模型的当前计价来源(供面板提示哪些模型走了兜底价, 提示用户去配单价)
+      const sourcesByModel = {}
+      for (const m of seen) sourcesByModel[m] = PRICING_SOURCE_LABELS[resolvePricing(m, Date.now()).source] || ''
+      return {
+        usdCnyRate,
+        fetchOfficial: fetchOfficialEnabled,
+        pricingUrl,
+        pricingFetchIntervalHours: Math.round(fetchIntervalMs / 3600000),
+        pricing: userPricing.map((r) => Object.assign({}, r, { peaks: clonePeaks(r.peaks) })),
+        fallback: Object.assign({}, fallbackPricing),
+        builtin: pricingTable.map((r) => ({ match: r.match, input: r.input, output: r.output, cacheRead: r.cacheRead, cacheWrite: r.cacheWrite, peaks: clonePeaks(r.peaks) })),
+        models: [...seen].sort(),
+        sourcesByModel,
+      }
+    }
+
+    // 保存面板提交的单价配置: 校验 → 合并进 dsh-dock.tokenlog(保留 pricingUrl 等未在面板暴露
+    // 的字段) → 落盘。保存后由调用方 loadPricingConfig() 刷新内存, 使改价立即影响费用估算。
+    async function savePricingConfig(input) {
+      const settings = ctx.get('settings')
+      if (!settings || typeof settings.get !== 'function' || typeof settings.mutate !== 'function') {
+        throw new Error('settings 服务不可用，单价无法持久化')
+      }
+      const bad = (msg) => { const err = new Error(msg); err.statusCode = 400; return err }
+      const nonNeg = (v, label) => {
+        const n = Number(v)
+        if (!Number.isFinite(n) || n < 0) throw bad(`「${label}」需为不小于 0 的数字`)
+        return n
+      }
+      const root = settings.get(DOCK_NS)
+      const current = root && typeof root === 'object' && root.tokenlog && typeof root.tokenlog === 'object' ? root.tokenlog : {}
+      const next = Object.assign({}, current)
+      if (input.usdCnyRate !== undefined) {
+        const r = Number(input.usdCnyRate)
+        if (!Number.isFinite(r) || r <= 0 || r > 1000) throw bad('「USD→CNY 汇率」需为 0~1000 的正数')
+        next.usdCnyRate = r
+      }
+      if (input.fetchOfficial !== undefined) next.fetchOfficial = !!input.fetchOfficial
+      if (input.pricing !== undefined) {
+        if (!Array.isArray(input.pricing)) throw bad('「模型单价」需为数组')
+        const rows = []
+        for (const raw of input.pricing) {
+          if (!raw || typeof raw !== 'object' || Array.isArray(raw)) throw bad('模型单价条目格式不符')
+          const match = String(raw.match || '').trim().toLowerCase()
+          if (!match) throw bad('模型单价条目的「模型匹配」不能为空')
+          if (match === '*') throw bad('模型匹配不能是 *；未匹配模型请填「兜底单价」')
+          const row = {
+            match,
+            input: nonNeg(raw.input, `${match} 输入单价`),
+            output: nonNeg(raw.output, `${match} 输出单价`),
+            cacheRead: nonNeg(raw.cacheRead, `${match} 缓存命中单价`),
+            cacheWrite: nonNeg(raw.cacheWrite, `${match} 缓存写入单价`),
+          }
+          // 分时段价(可多段): 每段 start/end 为 0~24 的本地小时(支持 8.5 表示 08:30);
+          // start>end 表示跨零点, start===end 表示全天。
+          const rawPeaks = Array.isArray(raw.peaks) ? raw.peaks : []
+          const peaks = []
+          for (const seg of rawPeaks) {
+            if (!seg || typeof seg !== 'object' || Array.isArray(seg)) throw bad(`「${match}」分时段条目格式不符`)
+            const start = Number(seg.start), end = Number(seg.end)
+            if (!Number.isFinite(start) || !Number.isFinite(end) || start < 0 || start > 24 || end < 0 || end > 24) {
+              throw bad(`「${match}」分时段时刻需为 0~24 的小时数`)
+            }
+            peaks.push({
+              start, end,
+              input: nonNeg(seg.input, `${match} 分时段输入单价`),
+              output: nonNeg(seg.output, `${match} 分时段输出单价`),
+              cacheRead: nonNeg(seg.cacheRead, `${match} 分时段缓存命中单价`),
+              cacheWrite: nonNeg(seg.cacheWrite, `${match} 分时段缓存写入单价`),
+            })
+          }
+          if (peaks.length) row.peaks = peaks
+          rows.push(row)
+        }
+        next.pricing = rows
+      }
+      if (input.fallback !== undefined) {
+        const f = input.fallback
+        if (!f || typeof f !== 'object' || Array.isArray(f)) throw bad('「兜底单价」格式不符')
+        next.fallback = {
+          input: nonNeg(f.input, '兜底输入单价'),
+          output: nonNeg(f.output, '兜底输出单价'),
+          cacheRead: nonNeg(f.cacheRead, '兜底缓存命中单价'),
+          cacheWrite: nonNeg(f.cacheWrite, '兜底缓存写入单价'),
+        }
+      }
+      try {
+        await settings.mutate(DOCK_NS, [{ op: 'set', path: ['tokenlog'], value: next }])
+      } catch (e) {
+        const err = new Error('保存单价被拒绝：' + ((e && e.message) || String(e)))
+        err.statusCode = 400
+        throw err
+      }
+      console.log('[dsh-dock] tokenlog pricing saved: rows=' + (Array.isArray(next.pricing) ? next.pricing.length : 0) +
+        ', fetchOfficial=' + next.fetchOfficial)
+      return next
+    }
+
+    // ===== 官网定价自动获取 =====
+    // 每 N 小时(默认 24)从官网抓取最新定价并解析 HTML 表格; 抓取/解析失败时使用内置默认值。
+    // 用户在面板配置的单价优先于抓取价, 因此抓取仅作兜底; 也可在面板关闭抓取。
+    let fetchedPricing = null  // Array<{ match, input, output, cacheRead, cacheWrite, currency }>
+    let lastFetchTime = 0
+
+    function getPricingFetchConfig() {
+      return { url: pricingUrl, intervalMs: fetchIntervalMs }
+    }
+
+    // 服务端出站请求前的 URL 校验(SSRF 防护): 仅 http/https, 且拒绝本机/环回/私有/保留地址。
+    function isPrivateOrReservedHost(host) {
+      const h = String(host || '').toLowerCase().replace(/^\[|\]$/g, '')
+      if (!h) return true
+      if (h === 'localhost' || h.endsWith('.localhost') || h.endsWith('.local') || h.endsWith('.internal')) return true
+      if (h.includes(':')) { // IPv6 字面量
+        return h === '::' || h === '::1' || h.startsWith('fe80') || h.startsWith('fc') || h.startsWith('fd')
+      }
+      const m = h.match(/^(\d{1,3})\.(\d{1,3})\.(\d{1,3})\.(\d{1,3})$/)
+      if (m) {
+        const a = Number(m[1]), b = Number(m[2])
+        return a === 0 || a === 10 || a === 127
+          || (a === 169 && b === 254)          // link-local
+          || (a === 172 && b >= 16 && b <= 31) // 私网
+          || (a === 192 && b === 168)          // 私网
+          || a >= 224                          // 组播/保留
+      }
+      return false
+    }
+
+    function assertAllowedFetchUrl(raw) {
+      let u
+      try { u = new URL(String(raw)) } catch { throw new Error('价格页地址不是合法 URL：' + raw) }
+      if (u.protocol !== 'http:' && u.protocol !== 'https:') throw new Error('价格页地址仅支持 http/https')
+      if (isPrivateOrReservedHost(u.hostname)) throw new Error('价格页地址不允许指向本机/内网/保留地址')
+      return u.toString()
+    }
+
+    // 从 HTML 中提取所有 <table> 的行列数据
+    function extractHTMLTables(html) {
+      const tables = []
+      const tableRe = /<table[^>]*>([\s\S]*?)<\/table>/gi
+      let tm
+      while ((tm = tableRe.exec(html)) !== null) {
+        const rows = []
+        const trRe = /<tr[^>]*>([\s\S]*?)<\/tr>/gi
+        let tr
+        while ((tr = trRe.exec(tm[1])) !== null) {
+          const cells = []
+          const tdRe = /<t[hd][^>]*>([\s\S]*?)<\/t[hd]>/gi
+          let td
+          while ((td = tdRe.exec(tr[1])) !== null) {
+            cells.push(td[1].replace(/<[^>]+>/g, '').replace(/&[a-z]+;/g, ' ').trim())
+          }
+          if (cells.length > 0) rows.push(cells)
+        }
+        if (rows.length >= 2) tables.push(rows)
+      }
+      return tables
+    }
+
+    // 从价格字符串提取数字(支持 "$0.27", "0.27", "¥1.5", "1.5元", "1,500" 等格式)
+    function parsePrice(s) {
+      if (!s) return null
+      const cleaned = String(s).replace(/,/g, '').replace(/￥|¥|元|\/.*$/g, '').trim()
+      const m = cleaned.match(/(\d+\.?\d*)/)
+      return m ? parseFloat(m[1]) : null
+    }
+
+    // 识别定价表列: 通过表头关键词定位各列
+    function identifyPricingColumns(headers) {
+      const h = headers.map(x => x.toLowerCase())
+      const find = (...patterns) => {
+        for (const p of patterns) {
+          const idx = h.findIndex(x => x.includes(p))
+          if (idx >= 0) return idx
+        }
+        return -1
+      }
+      return {
+        model: find('模型', 'model', '名称', 'name'),
+        inputMiss: find('缓存未命中', '未缓存', 'cache miss', 'uncached'),
+        inputHit: find('缓存命中', 'cache hit', 'cached'),
+        output: find('输出', 'output'),
+        peak: find('高峰', 'peak', 'busy', 'on-peak'),
+      }
+    }
+
+    // 从 HTML 解析 DeepSeek 官网定价
+    // 策略: 提取所有 <table>, 找到含"价格/缓存/输入/输出"关键词的表, 逐行提取模型名+价格;
+    // 识别 $ 标记区分 USD/CNY 币种; 无法解析时返回空数组(回退内置默认)。
+    function parsePricingFromHTML(html) {
+      const tables = extractHTMLTables(html)
+      const results = []
+      for (const rows of tables) {
+        const headers = rows[0]
+        const cols = identifyPricingColumns(headers)
+        // 至少需要模型名 + 一个价格列
+        if (cols.model < 0) continue
+        if (cols.inputMiss < 0 && cols.output < 0 && cols.inputHit < 0) continue
+        // 表头需与定价相关
+        const headerText = headers.join(' ')
+        if (!/价格|price|缓存|cache|输入|input|输出|output|tokens/i.test(headerText)) continue
+        for (let i = 1; i < rows.length; i++) {
+          const row = rows[i]
+          const name = (row[cols.model] || '').trim()
+          if (!name || /^(模型|model|名称|合计|total|---)/i.test(name)) continue
+          const input = cols.inputMiss >= 0 ? parsePrice(row[cols.inputMiss]) : null
+          const output = cols.output >= 0 ? parsePrice(row[cols.output]) : null
+          const cacheHit = cols.inputHit >= 0 ? parsePrice(row[cols.inputHit]) : null
+          if (input === null && output === null && cacheHit === null) continue
+          // 判断币种
+          const allText = row.join(' ')
+          const isUSD = /\$|USD|dollar/i.test(allText)
+          results.push({
+            match: name.toLowerCase(),
+            input: input || 0,
+            output: output || 0,
+            cacheRead: cacheHit || 0,
+            cacheWrite: 0, // DeepSeek 官方不收取缓存写入
+            currency: isUSD ? 'usd' : 'cny',
+          })
+        }
+      }
+      return results
+    }
+
+    // 从官网获取最新定价(异步, 失败静默回退内置默认)
+    async function fetchOfficialPricing() {
+      try {
+        const { url } = getPricingFetchConfig()
+        const safeUrl = assertAllowedFetchUrl(url)
+        console.log('[dsh-dock] tokenlog fetching official pricing from:', safeUrl)
+        const res = await fetch(safeUrl, {
+          headers: { 'User-Agent': 'dsh-dock-tokenlog/0.4', 'Accept': 'text/html,*/*' }
+        })
+        if (!res.ok) throw new Error('HTTP ' + res.status)
+        const html = await res.text()
+        console.log('[dsh-dock] tokenlog pricing page fetched, size:', html.length)
+        if (html.length < 500) {
+          console.warn('[dsh-dock] tokenlog page too small, likely SPA shell — using defaults')
+          return
+        }
+        const parsed = parsePricingFromHTML(html)
+        if (parsed.length > 0) {
+          fetchedPricing = parsed
+          lastFetchTime = Date.now()
+          console.log('[dsh-dock] tokenlog parsed', parsed.length, 'models from official pricing')
+        } else {
+          console.warn('[dsh-dock] tokenlog parsed 0 models from pricing page — check page structure, using defaults')
+        }
+      } catch (e) {
+        console.error('[dsh-dock] tokenlog fetch official pricing failed:', e && e.message, '— using built-in defaults')
+      }
+    }
+
+    // 读取 settings.yaml 的 llm-pi-ai 命名空间, 返回「现有提供商」路由名 -> [模型 id]。
+    // 用途: 前端提供商/模型下拉 = 现有配置 ∪ 历史记录, 去重合并 ——
+    //   - 没有任何记录时仍可选现有提供商;
+    //   - 提供商/模型已从配置删除时, 因有历史记录仍可选中(兼容两者)。
+    function readConfiguredProviders() {
+      const out = {}
+      try {
+        const settings = ctx.get('settings')
+        if (!settings || typeof settings.get !== 'function') return out
+        const sec = settings.get('llm-pi-ai')
+        const providers = sec && typeof sec === 'object' && sec.providers ? sec.providers : {}
+        for (const pid of Object.keys(providers || {})) {
+          const p = providers[pid]
+          if (!p || typeof p !== 'object') { out[pid] = []; continue }
+          const models = Array.isArray(p.models) ? p.models : []
+          const ids = []
+          for (const m of models) {
+            const id = m && typeof m === 'object' ? m.id : m
+            if (typeof id === 'string' && id) ids.push(id)
+          }
+          out[pid] = ids
+        }
+      } catch (e) {
+        console.error('[dsh-dock] tokenlog read configured providers failed', e && e.message)
+      }
+      return out
+    }
+
+    // ---------- 事件采集 ----------
+
+    // 预扫描 turn/end: 先建立 turn -> status/error 映射, 再生成记录
+    // (解决顺序处理导致的状态缺失: assistant/message 先于 turn/end 到达)
+    function preScanTurnEnds(sessionId, events) {
+      for (const e of events) {
+        if (e.type !== 'turn/end') continue
+        const t = e.data && e.data.turn
+        const r = e.data && e.data.reason
+        let status = 'completed'
+        if (r && typeof r === 'object') status = r.kind || 'completed'
+        else if (typeof r === 'string') status = r
+        turnStatus.set(sessionId + ':' + t, status)
+        let errInfo = null
+        if (r && typeof r === 'object' && r.error) {
+          const er = r.error
+          errInfo = {
+            statusCode: typeof er.status === 'number' ? er.status : 0,
+            errorCode: er.code || '',
+            errorMsg: er.message || '',
+          }
+        }
+        turnError.set(sessionId + ':' + t, errInfo)
+      }
+    }
+
+    function ingestEvent(sessionId, event) {
+      if (event.type === 'request/header') {
+        const h = event.data && event.data.header
+        if (h && h.config) {
+          headerCache.set(sessionId, {
+            provider: h.config.provider || '',
+            model: h.config.model || '',
+            effort: h.config.reasoningEffort || '',
+          })
+        }
+      }
+      if (event.type === 'turn/end') {
+        const t = event.data && event.data.turn
+        const r = event.data && event.data.reason
+        let status = 'completed'
+        if (r && typeof r === 'object') status = r.kind || 'completed'
+        else if (typeof r === 'string') status = r
+        turnStatus.set(sessionId + ':' + t, status)
+        let errInfo = null
+        if (r && typeof r === 'object' && r.error) {
+          const e = r.error
+          errInfo = {
+            statusCode: typeof e.status === 'number' ? e.status : 0,
+            errorCode: e.code || '',
+            errorMsg: e.message || '',
+          }
+        }
+        turnError.set(sessionId + ':' + t, errInfo)
+      }
+      if (event.type === 'step/start') {
+        stepStart.set(sessionId + ':' + event.data.turn + ':' + event.data.step, event.time)
+      }
+      if (event.type !== 'assistant/message') return null
+      const d = event.data || {}
+      let usage = d.usage
+      if (!usage || typeof usage !== 'object') usage = { inputTokens: 0, outputTokens: 0 }
+      const head = headerCache.get(sessionId) || { provider: '', model: '', effort: '' }
+      const turn = num(d.turn)
+      const step = num(d.step)
+      const startK = sessionId + ':' + turn + ':' + step
+      const startMs = stepStart.get(startK)
+      const llmMs = typeof startMs === 'number' ? Math.max(0, event.time - startMs) : null
+      const status = turnStatus.get(sessionId + ':' + turn) || 'in-progress'
+      const ei = turnError.get(sessionId + ':' + turn) || null
+      const input = num(usage.inputTokens)
+      const output = num(usage.outputTokens)
+      const cacheRead = num(usage.cacheReadTokens)
+      const cacheWrite = num(usage.cacheWriteTokens)
+      const reasoning = num(usage.reasoningTokens)
+      const priced = costOf(usage, head.model, event.time)
+      stepStart.delete(startK)
+      return {
+        id: sessionId + ':' + event.seq,
+        time: event.time,
+        sessionId,
+        provider: head.provider || '',
+        model: head.model || '',
+        apiKey: '',
+        inputTokens: input,
+        outputTokens: output,
+        cacheReadTokens: cacheRead,
+        cacheWriteTokens: cacheWrite,
+        reasoningTokens: reasoning,
+        billedInput: input + cacheRead + cacheWrite,
+        totalTokens: input + output + cacheRead + cacheWrite,
+        cacheHitPercent: (input + cacheRead) > 0 ? Math.round((cacheRead / (input + cacheRead)) * 100) : 0,
+        cost: priced.cost,
+        pricingSource: priced.source,
+        effort: head.effort || '',
+        status,
+        statusCode: ei ? ei.statusCode : 0,
+        errorCode: ei ? ei.errorCode : '',
+        errorMsg: ei ? ei.errorMsg : '',
+        llmMs,
+        turn,
+        step,
+      }
+    }
+
+    function ingestEvents(sessionId, events) {
+      const out = []
+      for (const e of events) {
+        const r = ingestEvent(sessionId, e)
+        if (r) out.push(r)
+      }
+      return out
+    }
+
+    // 历史扫描: 遍历所有会话, 读取完整日志, 重建索引
+    // 说明: 读取失败的多是旧版(v0→v1 迁移)、中途损坏或缺 seq 的会话——这类日志无法解码是常态,
+    // 不应视为异常。若逐条 console.error 会把终端刷屏, 改为统计失败数并只打印少量样例明细。
+    async function scanHistory() {
+      if (!sessionQuery) return
+      const failures = [] // { sid, msg } 保留前 MAX 条明细供排查
+      const MAX_FAIL_SAMPLES = 3
+      let failCount = 0
+      try {
+        const sessions = await sessionQuery.listSessions()
+        let added = 0
+        for (const s of sessions) {
+          const sid = s && s.header ? s.header.id : (s && s.id)
+          if (!sid || scannedSessions.has(sid)) continue
+          scannedSessions.add(sid)
+          try {
+            const snap = await sessionQuery.readSession(sid)
+            const events = snap && snap.events ? snap.events : []
+            // 两遍: 先建 turn/end 映射, 再生成记录
+            preScanTurnEnds(sid, events)
+            const recs = ingestEvents(sid, events)
+            for (const r of recs) records.set(r.id, r)
+            added += recs.length
+          } catch (e) {
+            failCount += 1
+            if (failures.length < MAX_FAIL_SAMPLES) failures.push(sid + ': ' + ((e && e.message) || String(e)))
+          }
+        }
+        // 失败会话统一汇总为一行(仅列前几条样例), 不逐条刷屏。
+        if (failCount > 0) {
+          console.warn('[dsh-dock] tokenlog 历史扫描: ' + failCount + ' 个会话未能读取' +
+            (failCount > MAX_FAIL_SAMPLES ? ', 前 ' + MAX_FAIL_SAMPLES + ' 条如下:' : ':') )
+          for (const f of failures) console.warn('  - ' + f)
+        }
+        console.log('[dsh-dock] tokenlog history scanned: +' + added + ' records, total ' + records.size)
+      } catch (e) {
+        console.error('[dsh-dock] tokenlog scanHistory failed', e && e.message)
+      }
+    }
+
+    function buildQuery(args) {
+      const q = args || {}
+      const from = typeof q.from === 'number' ? q.from : 0
+      const to = typeof q.to === 'number' ? q.to : Infinity
+      const provider = (q.provider || '').trim()
+      const model = (q.model || '').trim()
+      const status = (q.status || '').trim()
+      const effort = (q.effort || '').trim()
+      const sessionId = (q.sessionId || '').trim()
+      return [...records.values()].filter((r) => {
+        if (r.time < from || r.time > to) return false
+        if (provider && r.provider !== provider) return false
+        if (model && r.model !== model) return false
+        if (status && r.status !== status) return false
+        if (effort && r.effort !== effort) return false
+        if (sessionId && r.sessionId !== sessionId) return false
+        return true
+      })
+    }
+
+    function summarize(list, dim) {
+      const map = new Map()
+      for (const r of list) {
+        const key = r[dim] || '(none)'
+        const e = map.get(key)
+        if (e) {
+          e.calls += 1
+          e.inputTokens += r.inputTokens
+          e.outputTokens += r.outputTokens
+          e.cacheReadTokens += r.cacheReadTokens
+          e.cacheWriteTokens += r.cacheWriteTokens
+          e.reasoningTokens += r.reasoningTokens
+          e.billedInput += r.billedInput
+          e.totalTokens += r.totalTokens
+          e.cost += r.cost
+          if (typeof r.llmMs === 'number') { e.llmMs += r.llmMs; e.timed += 1 }
+        } else {
+          map.set(key, {
+            key, calls: 1,
+            inputTokens: r.inputTokens, outputTokens: r.outputTokens,
+            cacheReadTokens: r.cacheReadTokens, cacheWriteTokens: r.cacheWriteTokens,
+            reasoningTokens: r.reasoningTokens, billedInput: r.billedInput,
+            totalTokens: r.totalTokens, cost: r.cost,
+            llmMs: typeof r.llmMs === 'number' ? r.llmMs : 0,
+            timed: typeof r.llmMs === 'number' ? 1 : 0,
+          })
+        }
+      }
+      return [...map.values()]
+        .map((e) => ({ ...e, cacheHitPct: e.billedInput > 0 ? Math.round((e.cacheReadTokens / e.billedInput) * 100) : 0 }))
+        .sort((a, b) => b.calls - a.calls)
+    }
+
+    function totals(list) {
+      const t = { calls: 0, inputTokens: 0, outputTokens: 0, cacheReadTokens: 0, cacheWriteTokens: 0, reasoningTokens: 0, billedInput: 0, totalTokens: 0, cost: 0, llmMs: 0, timed: 0 }
+      for (const r of list) {
+        t.calls += 1
+        t.inputTokens += r.inputTokens
+        t.outputTokens += r.outputTokens
+        t.cacheReadTokens += r.cacheReadTokens
+        t.cacheWriteTokens += r.cacheWriteTokens
+        t.reasoningTokens += r.reasoningTokens
+        t.billedInput += r.billedInput
+        t.totalTokens += r.totalTokens
+        t.cost += r.cost
+        if (typeof r.llmMs === 'number') { t.llmMs += r.llmMs; t.timed += 1 }
+      }
+      t.cacheHitPct = t.billedInput > 0 ? Math.round((t.cacheReadTokens / t.billedInput) * 100) : 0
+      return t
+    }
+
+    // ---------- RPC（webServer HTTP 路由，前缀 /dsh-dock/tokenlog/） ----------
+
+    // 实时监听 + 历史扫描: 必须等 sessionQuery 服务就绪。
+    // 注意: 不能在 setup 时用 ctx.get('sessionQuery') 软获取——profile 插件无 inject 声明,
+    // 可能早于 sessionQuery 提供方激活, 拿到 undefined 会让 监听+扫描 整条采集链路静默失效。
+    // 改用 ctx.inject(['sessionQuery'], ...): 服务已就绪则立即执行, 否则挂起等待。
+    let sessionQuery = null
+    disposers.push(ctx.inject(['sessionQuery'], (sqCtx) => {
+      sessionQuery = sqCtx.sessionQuery
+      // 服务就绪后加载 settings 单价配置(汇率 + 自定义单价 + 抓取开关), 供采集/重扫/导出使用
+      loadPricingConfig()
+      // 从官网获取最新定价(启动时 + 每 N 小时自动刷新; 面板关闭抓取时不发起请求)
+      if (fetchOfficialEnabled) fetchOfficialPricing()
+      const fetchTimer = setInterval(() => { if (fetchOfficialEnabled) fetchOfficialPricing() }, getPricingFetchConfig().intervalMs)
+      sqCtx.effect(() => () => clearInterval(fetchTimer))
+      // 实时监听: session/event (root scope 可收到所有 session 事件, 与官方 persistence 同款订阅)
+      disposers.push(sqCtx.on('session/event', (session, event) => {
+        try {
+          const sid = typeof session === 'object' && session ? (session.id || '') : String(session || '')
+          const rec = ingestEvent(sid, event)
+          if (rec) records.set(rec.id, rec)
+          // turn/end 到达后回填该 turn 已有记录的状态/错误(实时增量场景)
+          if (event.type === 'turn/end') {
+            const t = event.data && event.data.turn
+            const key = sid + ':' + t
+            const st = turnStatus.get(key)
+            const ei = turnError.get(key) || null
+            if (st) {
+              for (const r of records.values()) {
+                if (r.sessionId === sid && String(r.turn) === String(t)) {
+                  r.status = st
+                  r.statusCode = ei ? ei.statusCode : 0
+                  r.errorCode = ei ? ei.errorCode : ''
+                  r.errorMsg = ei ? ei.errorMsg : ''
+                }
+              }
+            }
+          }
+        } catch (err) {
+          console.error('[dsh-dock] tokenlog ingest error', err)
+        }
+      }))
+      // 服务就绪后立即做一次历史全量扫描
+      scanHistory()
+    }))
+
+    // webServer 就绪后注册 /dsh-dock/tokenlog 前缀路由(ctx.inject 保证就绪后再执行)
+    disposers.push(ctx.inject(['webServer'], (wsCtx) => {
+      wsCtx.effect(() => wsCtx.webServer.register({
+        kind: 'prefix',
+        path: '/dsh-dock/tokenlog',
+        async handler(req, res) {
+          try {
+            const url = new URL(req.url || '/', 'http://dsh.internal')
+            const method = url.pathname.replace(/^\/dsh-dock\/tokenlog\/?/, '').split('/')[0] || ''
+            const payload = await readBody(req)
+            if (method === 'query') {
+              loadPricingConfig()
+              // 费用按当前单价即时重算(withCost): 面板改价后立即反映到统计与明细, 无需重扫历史。
+              const list = buildQuery(payload || {}).map(withCost)
+              const rows = [...list].sort((a, b) => b.time - a.time)
+              const all = [...records.values()]
+              const cfg = readConfiguredProviders()
+              // 提供商/模型下拉 = 现有配置(llm-pi-ai.providers) ∪ 历史记录, 去重排序。
+              // 兼容两种场景: 无记录时仍可选现有提供商; 提供商/模型已删除时因记录仍可选中。
+              const providerSet = new Set(all.map((r) => r.provider).filter(Boolean))
+              const modelSet = new Set(all.map((r) => r.model).filter(Boolean))
+              for (const pid of Object.keys(cfg)) if (pid) providerSet.add(pid)
+              for (const ids of Object.values(cfg)) for (const id of ids) modelSet.add(id)
+              // 提供商 -> 模型 映射(供前端联动筛选: 选中提供商后模型下拉只显示其下模型)
+              const modelsByProvider = {}
+              const initMbp = (p) => { if (!modelsByProvider[p]) modelsByProvider[p] = [] }
+              for (const p of providerSet) initMbp(p)
+              for (const r of all) {
+                if (!r.model) continue
+                const p = r.provider || '(none)'
+                initMbp(p)
+                if (!modelsByProvider[p].includes(r.model)) modelsByProvider[p].push(r.model)
+              }
+              for (const [pid, ids] of Object.entries(cfg)) {
+                initMbp(pid)
+                for (const id of ids) if (!modelsByProvider[pid].includes(id)) modelsByProvider[pid].push(id)
+              }
+              for (const p of Object.keys(modelsByProvider)) modelsByProvider[p].sort()
+              return sendJson(res, 200, {
+                ok: true,
+                data: {
+                  records: rows,
+                  counts: { total: records.size, matching: rows.length },
+                  providers: [...providerSet].sort(),
+                  models: [...modelSet].sort(),
+                  modelsByProvider,
+                  statuses: [...new Set(all.map((r) => r.status).filter(Boolean))],
+                  efforts: [...new Set(all.map((r) => r.effort).filter(Boolean))],
+                  sessionIds: [...new Set(all.map((r) => r.sessionId).filter(Boolean))].sort(),
+                  rateUsdCny: usdCnyRate,
+                  pricingInfo: {
+                    hasCustom: userPricing.length > 0,
+                    customCount: userPricing.length,
+                    fetchOfficial: fetchOfficialEnabled,
+                    sources: [...new Set(list.map((r) => r.pricingSource).filter(Boolean))].map((s) => PRICING_SOURCE_LABELS[s] || s),
+                  },
+                  totals: totals(list),
+                  summary: payload && payload.dim ? summarize(list, payload.dim) : [],
+                }
+              })
+            }
+            if (method === 'pricing') {
+              loadPricingConfig()
+              return sendJson(res, 200, { ok: true, data: readPricingConfig() })
+            }
+            if (method === 'setpricing') {
+              await savePricingConfig(payload || {})
+              loadPricingConfig()
+              return sendJson(res, 200, { ok: true, data: readPricingConfig() })
+            }
+            if (method === 'export') {
+              loadPricingConfig()
+              const list = buildQuery(payload || {}).map(withCost)
+              // export 返回结构化明细（不是 CSV 文本）：客户端按「明细表列序 + 中文表头」统一
+              // 格式化成 CSV，取值格式与页面显示一致；rateUsdCny 供客户端把 cost(USD) 折人民币。
+              return sendJson(res, 200, {
+                ok: true,
+                data: { rows: list, rateUsdCny: usdCnyRate, count: list.length },
+              })
+            }
+            if (method === 'scan') {
+              await scanHistory()
+              return sendJson(res, 200, { ok: true, data: { total: records.size } })
+            }
+            if (method === 'hello') {
+              return sendJson(res, 200, { ok: true, data: { ok: true, records: records.size, scanned: scannedSessions.size } })
+            }
+            return sendJson(res, 404, { ok: false, error: { code: 'method-not-found', message: 'unknown method: ' + method, details: {} } })
+          } catch (e) {
+            // 参数校验类错误按 4xx 返回(如单价非法), 其余为 500
+            const status = e && e.statusCode ? e.statusCode : 500
+            return sendJson(res, status, { ok: false, error: { code: status === 500 ? 'internal' : 'bad-request', message: (e && e.message) || String(e), details: {} } })
+          }
+        }
+      }), 'dsh-dock tokenlog: /dsh-dock/tokenlog HTTP route')
+    }))
+
+    return dispose
+  },
+}