dsh-recent-usage-metrics 0.3.1 → 0.3.3

package/CHANGELOG.md

@@ -1,3 +1,17 @@
+## 0.3.3
+
+- 输出 Token 列的悬浮提示里，输入 token 改为带千分位的完整数字（`370,683 tok`），
+  不再是缩写。单元格本身仍缩写以保持对齐，提示存在的意义正是补回被缩写省掉的精度。
+
+## 0.3.2
+
+- 修复：速度列的悬浮提示常常拿不到模型名，或显示成别的模型。
+  原来从统计折叠的 `request/header` 事件推断，但那种事件只在模型或工具集变化时才写——
+  真实会话里 3139 次请求只对应 24 条，覆盖率约 1%，其余步骤只能借用几小时前的旧值。
+- 改为读轨迹域的 assistant 节点：它把 `assistant/message.source` 原样复制成
+  `providerMetadata`，每步都有，是客户端半能拿到的唯一逐步事实。
+- 没有轨迹节点的步骤保持空值，不再借用旧 header 猜测。
+
 ## 0.3.1
 
 - 修复：切换模型后的第一步，抽屉速度列的悬浮提示显示了上一个模型。

package/README.en.md

@@ -151,7 +151,7 @@ Every drawer column and every header carries a hint:
 | --- | --- |
 | Clock column | the full timestamp `11-06 10:47:41.828` |
 | Speed column | the model that served that row |
-| Tokens column | input tokens and cache hit rate |
+| Tokens column | input tokens (with thousands separators, e.g. `370,683 tok`) and cache hit rate |
 | Headers | what the column measures (the `#` header draws the four state glyphs) |
 
 The hints are a real overlay rather than `title`: `title` can only show plain text, and the

package/README.md

@@ -135,7 +135,7 @@ DSH 的极客级请求指标。它在 **设置 → 通用 → 性能与用量**
 | --- | --- |
 | 时间列 | 完整时间戳 `11-06 10:47:41.828` |
 | 速度列 | 该行使用的模型 |
-| 输出 Token 列 | 输入 token 与缓存命中率 |
+| 输出 Token 列 | 输入 token（带千分位，如 `370,683 tok`）与缓存命中率 |
 | 表头 | 该列的含义（`#` 一列用图标展示四种状态） |
 
 提示是自建的浮层而不是 `title`：`title` 只能显示纯文本，而 `#` 列的说明要画出图标。

package/client/client.js

@@ -1172,72 +1172,31 @@ window.__ModuleLoader__.load({
     /**
      * Which model answered each step, keyed `turn:step`.
      *
-     * The fold puts no provider/model on a step — `stepMetrics` carries only derived figures and
-     * has no `startSeq` either — so the model is read from the turn's `request-header` event
-     * rows, one per routed request.
+     * Read from the Trajectory domain's assistant nodes, which carry `providerMetadata` copied
+     * straight from `assistant/message.source`. That is the only per-step source of truth the
+     * client half can reach:
      *
-     * Two things make a naive "newest header before the step started" rule wrong:
+     * - The statistics fold puts no provider/model on a step, and its `request-header` rows are
+     *   written only when the routed model or tool set CHANGES — 24 of them for 3139 requests in
+     *   a real session, so they describe about 1% of steps and mislabel the rest.
+     * - The Chat snapshot's assistant nodes drop the source entirely.
      *
-     * 1. A header is written a few milliseconds AFTER its own `step/start` (measured: 4 ms), so
-     *    `header.time <= step.startTime` misses the header that belongs to this request and
-     *    falls back to the PREVIOUS request's model. That showed the wrong provider on the first
-     *    step after a model switch.
-     * 2. Not every step issues a new header: consecutive steps reuse the routed model, and then
-     *    the newest earlier header is the correct answer.
-     *
-     * So the window is tried first — every header from this step up to the next step's start —
-     * and the newest earlier header is the fallback. Measured against `assistant/message.source`
-     * on a real 3119-step log: the time-only rule matched 99.8%, this one matches 100%.
-     *
-     * `model/selection` is deliberately not used: it records what the user PICKED and can precede
-     * the request that actually carries it, so it would name a model that was not called.
+     * A compaction is not covered here: it has its own `compaction.model`.
      */
-    function modelIndexFromStats(statsSnapshot) {
-      const turns = statsSnapshot?.turns
+    function modelIndexFromTrajectory(trajectorySnapshot) {
       const index = new Map()
-      if (!Array.isArray(turns)) return index
-      for (const turn of turns) {
-        if (!Array.isArray(turn.steps)) continue
-        const headers = []
-        if (Array.isArray(turn.events)) {
-          for (const event of turn.events) {
-            if (event?.kind !== 'request-header') continue
-            const at = numberOrNull(event.time)
-            if (at === null) continue
-            if (typeof event.model !== 'string' && typeof event.provider !== 'string') continue
-            headers.push({ time: at, provider: event.provider ?? null, model: event.model ?? null })
-          }
-          headers.sort((left, right) => left.time - right.time)
-        }
-        if (headers.length === 0) continue
-
-        const starts = []
-        for (const step of turn.steps) {
-          const at = numberOrNull(step.startTime)
-          if (at === null || typeof step.step !== 'number') continue
-          starts.push({ at, step: step.step })
-        }
-        if (starts.length === 0) continue
-        starts.sort((left, right) => left.at - right.at)
-
-        for (let i = 0; i < starts.length; i += 1) {
-          const from = starts[i].at
-          const until = i + 1 < starts.length ? starts[i + 1].at : Number.POSITIVE_INFINITY
-          let chosen = null
-          for (const header of headers) {
-            if (header.time < from) continue
-            if (header.time >= until) break
-            chosen = header
-          }
-          if (chosen === null) {
-            for (const header of headers) {
-              if (header.time < from) chosen = header
-              else break
-            }
-          }
-          if (chosen === null) continue
-          index.set(`${turn.turn}:${starts[i].step}`, chosen)
-        }
+      const nodes = trajectorySnapshot?.eventNodes
+      if (!Array.isArray(nodes)) return index
+      for (const node of nodes) {
+        if (node === null || typeof node !== 'object') continue
+        if (node.kind !== 'assistant') continue
+        const meta = node.providerMetadata
+        if (meta === null || typeof meta !== 'object') continue
+        const model = typeof meta.model === 'string' && meta.model !== '' ? meta.model : null
+        const provider = typeof meta.provider === 'string' && meta.provider !== '' ? meta.provider : null
+        if (model === null && provider === null) continue
+        if (typeof node.turn !== 'number' || typeof node.step !== 'number') continue
+        index.set(`${node.turn}:${node.step}`, { provider, model })
       }
       return index
     }
@@ -1295,9 +1254,8 @@ window.__ModuleLoader__.load({
      * *after* selection: capping here at ten and then dropping the live step left the drawer
      * showing nine rows for as long as a request was streaming.
      */
-    function historyFromStats(statsSnapshot) {
+    function historyFromStats(statsSnapshot, models) {
       const steps = statsSteps(statsSnapshot)
-      const models = modelIndexFromStats(statsSnapshot)
       const rows = []
       for (const entry of steps) {
         // A step only has a row once it has started; `step/start` is what supplies this.
@@ -1305,7 +1263,7 @@ window.__ModuleLoader__.load({
         if (!statsStepHasFigures(entry.metrics)) continue
         const row = readingOfStatsStep(entry)
         // Attribute the request to the model that served it, for the speed column's tooltip.
-        const owner = models.get(row.key)
+        const owner = models === undefined ? undefined : models.get(row.key)
         if (owner !== undefined) {
           row.model = owner.model
           row.provider = owner.provider
@@ -1571,6 +1529,29 @@ window.__ModuleLoader__.load({
       return approximate ? t('approxTokensUnit', { value: rounded }) : t('tokensUnit', { value: rounded })
     }
 
+    /**
+     * A whole token count with thousands separators, for a hover hint.
+     *
+     * The cell itself has to stay narrow, so it abbreviates (`4.9k tok`). The hint has room, and
+     * a reader comparing an input of 370683 against one of 1914 needs the exact digits — `370k`
+     * and `371k` would hide the difference.
+     *
+     * The separator is a comma in both languages this plugin ships, so it is written out rather
+     * than derived from the host locale: that keeps the hint identical in zh and en, and avoids
+     * depending on a locale tag the slot props do not provide.
+     */
+    function tokensExact(value, t) {
+      if (value === null || value === undefined || !Number.isFinite(value)) return t('none')
+      const digits = String(Math.round(Math.abs(value)))
+      let grouped = ''
+      for (let i = 0; i < digits.length; i += 1) {
+        const fromEnd = digits.length - i
+        grouped += digits[i]
+        if (fromEnd > 1 && (fromEnd - 1) % 3 === 0) grouped += ','
+      }
+      return t('tokensUnit', { value: value < 0 ? `-${grouped}` : grouped })
+    }
+
     function clockText(at) {
       const date = new Date(at)
       if (Number.isNaN(date.getTime())) return '—'
@@ -1692,7 +1673,11 @@ window.__ModuleLoader__.load({
     /** The tokens column's hint: the input side, which the cell itself does not show. */
     function tokensHintText(row, t) {
       const none = t('none')
-      const input = row.context === null || row.context === undefined ? none : tokensText(row.context, false, t)
+      /*
+       * The input is spelled out with thousands separators; the cell above it abbreviates. The
+       * hint exists precisely for the cases where the abbreviation loses information.
+       */
+      const input = row.context === null || row.context === undefined ? none : tokensExact(row.context, t)
       const hit =
         row.cacheHit === null || row.cacheHit === undefined
           ? none
@@ -2999,8 +2984,11 @@ window.__ModuleLoader__.load({
        * what made the drawer lag the log.
        */
       const statsRows = useMemo(
-        () => (sessionStats === undefined ? undefined : historyFromStats(sessionStats)),
-        [sessionStats],
+        () =>
+          sessionStats === undefined
+            ? undefined
+            : historyFromStats(sessionStats, modelIndexFromTrajectory(trajectory)),
+        [sessionStats, trajectory],
       )
       /*
        * Fallback for a composition without the statistics target: derive rows from the Chat

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-recent-usage-metrics",
-  "version": "0.3.1",
+  "version": "0.3.3",
   "description": "Geek level for the Performance & usage setting: a live reading of the newest request (first-token latency, speed, tokens, total, amortised) plus a drawer listing the last ten, all derived from the shipped Statistics view data",
   "repository": {
     "type": "git",