dsh-auto-model-routing 0.4.1 → 0.4.2

package/CHANGELOG.md

@@ -1,5 +1,14 @@
 # Changelog
 
+## 0.4.2
+
+- AUTO now handles Claude and GPT model families only: selecting any other main model switches the session to MANUAL instead of failing the turn in the effort mapper.
+- Support the renamed `dsh-magpie-connector` Magpie route as an exact policy transport (replacing `dsh-magpie`); `dsh-magpie-connect` remains supported.
+- Fix the missing feedback entry on turns whose request configuration did not change: the main decision is marked applied from the effective request header instead of waiting for a new one.
+- Record two implicit feedback facts as separate ledger events keyed by session and decision: reply exposure and a later switch to MANUAL. No message text is stored; routing is unchanged.
+- The usage report adds a mutually exclusive implicit-signal breakdown per comparison group, with a fixed caveat that implicit no-objection is not the same as appropriate.
+- Sync the AUTO-selected effort through the official selection and simplify the picker display.
+
 ## 0.4.1
 
 - Add the on-demand `auto-model-routing-report` skill and read-only `lib/usage-report.js`: joins policy-v2 decisions to actual per-call usage by message ID, grouped by version × model × depth, with coverage gaps, sample thresholds and fixed caveats. Observational only; never writes, spawns or uses the network, and never prints message text, paths or session IDs.

package/README.md

@@ -55,7 +55,8 @@ Restart DSH and refresh the Web page. In **Settings → Auto model routing**, ch
 
 ### Main model stays fixed; children are selected separately
 
-- **Main chat:** `lower / baseline / higher` moves at most one supported effort step below/at/above the model's verified recommended starting point. The cap defaults to `xhigh`; AUTO never selects `max`. It does not change the main provider/model or write global model defaults.
+- **Main chat:** `lower / baseline / higher` moves at most one supported effort step below/at/above the model's verified recommended starting point. The cap defaults to `xhigh`; AUTO never selects `max`. It keeps the main provider/model and synchronizes the selected effort through official `selectModel`, which also saves global defaults.
+- **Claude/GPT families only:** AUTO manages models whose id (after any transport prefix) starts with `claude-` or `gpt-`. Selecting any other main model — DeepSeek, GLM, Grok, Gemini, Kimi, Qwen, … — switches the session to MANUAL and locks the chip: it shows an outlined info icon, and hovering the whole button explains the support scope with an issue/PR link to the repository. Switching back to a Claude/GPT model restores a clickable MANUAL chip; AUTO is never re-enabled automatically.
 - **Fresh children:** capability is `bounded / analytical / frontier`; effort is judged separately. Claude slots are Haiku / Sonnet / Opus 5.5; GPT slots are Luna / Sol / Sol. The current catalog, supported efforts and official child allowlist must all permit the route. Magpie uses its exact provider and full family-prefixed model IDs, not a fallback to local transports.
 - **Explicit choices win:** specifying a model or effort leaves only the other dimension to judge. Specifying both skips Jev and validates the route. Missing task scope stops delegation rather than guessing.
 - **Official creation binds the route:** fresh `subagent` and fresh `spawn_teammate` paths bind an exact child identity before its first request. `route_subagent` is an optional preview—not proof of application and not a user constraint to copy into a subsequent call. Fork and unbound creation paths are unsupported; there is no silent substitution.
@@ -74,6 +75,12 @@ Preset model menus expose documented versions, not live availability guarantees.
 
 **Certainty is not the selected probability.** Choice certainty is `(pmax − 1/n) / (1 − 1/n)`: with three candidates, a 70% top probability gives 55% certainty. The panel shows the service's returned certainty and full candidate distribution separately, including returned zeroes; missing values are marked unavailable. Neither number is task success rate.
 
+### Magpie transports and picker display
+
+The default policy accepts both exact transports, `dsh-magpie-connector` and `dsh-magpie-connect`. They are not aliases: full model ids, live model capabilities, supported effort/caps, the current allowlist and the parent's captured allowlist all retain their exact transport identity. Unknown families need an explicit model. With a root AUTO session and explicit `policy: legacy`, both Magpie previews reject before Jev; MANUAL still returns `routed: false` without judging or state writes. Already bound children replay their saved route.
+
+AUTO applies the selected effort to the current main request and awaits official `selectModel` on the same provider/model when the effort changes. This also updates global defaults; the official picker shows one effort value. MANUAL requests pass through unchanged and bound children never write the main selection. A selection-sync failure is logged while the current request retains its selected effort. Magpie menu grouping still requires ohmydsh's managed five-package Host extension; stock rc.2 keeps a flat list. See [Magpie setup and display requirements](<../dsh-magpie-connector/README.md#model-menu-groups-and-effort>); local tests do not establish deployment or live GUI acceptance.
+
 ## Good to know
 
 - **Failures are visible.** Main judgment failure disables AUTO with a reason; child judgment/route failure stops dispatch. No local classifier, automatic retry, upgrade or cross-service fallback.
@@ -83,7 +90,6 @@ Preset model menus expose documented versions, not live availability guarantees.
 - **Local state:** configuration is under `$DSH_HOME/plugins/dsh-auto-model-routing/`; mode, child routes and feedback are under `$DSH_HOME/state/auto-model-routing/`. Feedback stores resource metadata, IDs, versions, returned distributions and timestamps—not task/reply text. Request event IDs allow later usage correlation; costs are not fabricated when unavailable. See [feedback and evaluation](docs/feedback-evaluation.md).
 - **No automatic learning.** Human allocation feedback is observational. It does not change routing or prove quality/cost benefits. Old messages without decision bindings are not backfilled; old records with missing evaluation fields remain unknown.
 - **Upgrade:** existing AUTO sessions use the new policy without a version switch or rewriting their stored mode. New chats still default to MANUAL unless configured otherwise. Historical four-tier mappings remain only for explicit `policy: "legacy"` rollback in plugin configuration; they are not the current policy or feedback menu.
-
 ## Architecture
 <!-- section: architecture -->
 
@@ -101,7 +107,6 @@ These isolated-host recordings use mock Jev/model responses and show the previou
 ![Legacy four-tier routing settings](docs/settings.gif)
 
 </details>
-
 ## Development
 
 ```sh

package/README.zh.md

@@ -55,7 +55,8 @@ dsh plugin --profile web add prgrmrwy/dsh-auto-model-routing
 
 ### 主模型不换，子任务独立选配
 
-- **主会话**：`lower / baseline / higher` 相对于具体模型已验证的推荐起点，在受支持序列降低一档、保持或提高一档。缺省上限 `xhigh`，自动不选 `max`；不修改主 provider/model，也不改写全局模型默认值。
+- **主会话**：`lower / baseline / higher` 相对于具体模型已验证的推荐起点，在受支持序列降低一档、保持或提高一档。缺省上限 `xhigh`，自动不选 `max`；主 provider/model 不变；通过官方 `selectModel` 同步选定 effort，同时保存全局默认值。
+- **仅支持 Claude / GPT 系列**：AUTO 只管理模型名（去掉 transport 前缀后）以 `claude-` 或 `gpt-` 开头的主模型。切换到其他主模型（DeepSeek、GLM、Grok、Gemini、Kimi、Qwen 等）会自动落为 MANUAL 并锁定芯片：芯片显示描边 info 图标，悬停整个按钮说明支持范围并提供仓库的 issue/PR 链接。切回 Claude / GPT 系列后芯片恢复可点击且保持 MANUAL，不会自动重新启用 AUTO。
 - **fresh 子任务**：能力需求为 `bounded / analytical / frontier`，推理投入独立判断。Claude 能力槽为 Haiku / Sonnet / Opus 5.5；GPT 为 Luna / Sol / Sol。实际路由必须同时满足当前目录、支持的 effort 和官方子智能体允许列表。Magpie 保留真实 provider 与家族前缀完整模型 ID，不回退到本地入口。
 - **尊重显式选择**：只指定模型或投入时，仅判断另一维；两者都指定则不请求 Jev，只校验路由。缺少任务目标或范围时停止委派，不猜。
 - **官方创建绑定路由**：fresh `subagent` 和 fresh `spawn_teammate` 在子任务首次请求前绑定精确身份。`route_subagent` 只是可选预览，不代表已应用，也不能把推荐复制成用户约束。fork 和未绑定创建路径暂不支持，不静默替换。
@@ -74,6 +75,12 @@ dsh plugin --profile web add prgrmrwy/dsh-auto-model-routing
 
 **判定确定度不等于选中概率**：Choice 公式为 `(pmax − 1/n) / (1 − 1/n)`，三个候选中最高概率 70% 对应确定度 55%。面板分别显示服务返回的确定度和完整分布，保留返回的 0%，缺值标未提供。两者都不是任务成功率。
 
+### Magpie transport 与选择器展示
+
+默认策略接受 `dsh-magpie-connector`、`dsh-magpie-connect` 两个精确 transport，不互为 alias：完整 model id、实时模型能力、支持 effort/cap、当前允许列表与父 captured allowlist 都按原 transport 校验。未知家族需显式 model。根会话 AUTO 且显式 `policy: legacy` 时，两种 Magpie 预览均在 Jev 前拒绝；MANUAL 仍直接返回 `routed: false`，不判档、不写状态。已绑定子实例恢复原路线。
+
+AUTO 将选定 effort 应用于本次主请求；effort 改变时，等待官方 `selectModel` 同步同一 provider/model，并允许保存全局默认值。官方选择器只显示一个 effort。MANUAL 请求原样通过，已绑定子会话不写主选择；同步失败会记录告警，本次请求仍保留选定 effort。Magpie 菜单分组仍需 ohmydsh 受管五包 Host 扩展，stock rc.2 保持 flat 列表。见 [Magpie 配置与展示前提](<../dsh-magpie-connector/README.zh.md#模型菜单分组与-effort>)；本地测试不等于已部署或 live GUI 验收。
+
 ## 须知
 
 - **失败可见**：主判断失败会停用 AUTO 并显示原因，子判断或路由失败则停止派发。没有本地分类器、自动重试、自动升级或跨服务降级。
@@ -101,7 +108,6 @@ dsh plugin --profile web add prgrmrwy/dsh-auto-model-routing
 ![旧版四档路由设置](docs/settings.gif)
 
 </details>
-
 ## 开发验证
 
 ```sh

package/docs/feedback-evaluation.md

@@ -14,5 +14,22 @@
 - 以requestSeq和消息ID关联实际usage/延迟/返工证据；拿不到价格或token就未知。用户主观反馈不等于正确性oracle，适合率也不能单独证明质量或节省。
 - 按需观察报表见包内 skill `auto-model-routing-report`（`lib/usage-report.js`，只读）；仍不做自动阈值调整、自动训练或换模重试。
 
+## 隐式事实与分布（2026-10-26）
+
+运行时只追加 `{type:'fact', version:1, sessionId, decisionId, fact, at}`。`fact` 仅为 `exposed` 或 `manual-switch`；按决策/类型保留首次事实，时间由 Host 生成，不收正文。曝光要求回复操作栏在前台、聚焦页面连续完整可见 3 秒；切 MANUAL 在清除运行态之前取当前主决策，只在切换成功后记录。写入失败只记录日志，不阻断主线。
+
+回合结束与继续不重复写账本，按需报告从已有日志推导。决策所在回合取最后一次关联调用的实际回合号，不是归属的用户回合。第一个后续用户回合的开始时间为继续时间；曝光时间必须早于继续时间，否则视为事后翻看，不算有效曝光。没有继续时曝光仍有效，但不能得到隐式无异议。
+
+每个比较子组的 `implicit.counts` 五类互斥，合计等于决策数：
+1. `explicit`：有显式评价；`implicit.explicit` 分类别，冲突单列。
+2. `negative`：无显式评价，且用户中止了决策最后一次关联调用所在回合，或切换 MANUAL；`implicit.negative` 分原因，两原因可重叠。
+3. `implicit-ok`：有效曝光、正常完成且用户继续，无负向事实。只表示这些行为同时出现，不证明资源分配合适。
+4. `inconclusive`：有效曝光但其它条件不足，包括未继续、结束未知或父会话取消。
+5. `unobserved`：没有有效曝光。
+
+`implicit.exposedExplicit` 是有效曝光且有显式评价的参照计数。曝光不证明读过；用户中止或切 MANUAL 也可能因为换任务，不证明策略有错。旧数据不回填曝光，未知不能当正例。隐式信号、显式评价和将来的机器评审均应分开，不自动调策略。
+
+D012 修复后，实际配置不变、不产生新 `request/header` 的回合也会绑定决策。`snapshot.requestSeq` 指向当时生效的请求头（可能来自更早回合），`requestTurn` 是应用时的当前回合。上线后记录覆盖率会提高；这是补缺口，不是策略改善，不能直接比较修复前后的适合率。
+
 ## 确定度
 UI称“判定确定度”，Choice公式为`(pmax - 1/n)/(1 - 1/n)`；n为该维候选数。三选一最高概率70%得到55%，不等于选中概率或任务成功率。插件直接展示服务返回值，分布独立保留，不从四舍五入后的百分比反推。官方定义：https://docs.typesafe.ai/confidence.md 。

package/lib/client/index.d.ts

@@ -1,6 +1,11 @@
 import type { Context } from '@deepseek-ai/cordis';
 import { type Decision, type Rpc, type Toast } from './control.js';
 import { type JevCategory } from '../hints.js';
+interface AutoSupport {
+    supported: boolean;
+    provider?: string;
+    model?: string;
+}
 interface View {
     mode: 'AUTO' | 'MANUAL';
     policy?: string;
@@ -13,14 +18,17 @@ interface View {
     };
     decision?: Decision;
     decisionToast?: boolean;
+    autoSupport?: AutoSupport;
 }
+type SubscribeSelection = (sessionId: string, onChange: () => void) => () => void;
 interface Props {
     sessionId: string;
     rpc: Rpc;
+    subscribeSelection?: SubscribeSelection;
 }
 export declare const name = "dsh-auto-model-routing/client";
 export declare const inject: string[];
-export declare function RoutingChip({ sessionId, rpc, initialView, initiallyOpen, initialToast }: Props & {
+export declare function RoutingChip({ sessionId, rpc, initialView, initiallyOpen, initialToast, subscribeSelection }: Props & {
     initialView?: View;
     initiallyOpen?: boolean;
     initialToast?: Toast;

package/lib/client/index.js

@@ -2,13 +2,14 @@ import { createElement as h, useEffect, useRef, useState } from 'react';
 import { createPortal } from 'react-dom';
 import { callRouting, chipOrder, nextToast, panelOpenAfter, TOAST_MS, toastText } from './control.js';
 import { CATEGORY_HINTS } from '../hints.js';
-import { IconGaugeOutlineRegular, Menu, Tooltip } from '@deepseek-ai/dsh-client-ui-primitives';
+import { IconInfoOutlineRegular, IconGaugeOutlineRegular, Menu, Tooltip } from '@deepseek-ai/dsh-client-ui-primitives';
 import { LabelButton } from './label.js';
 import * as settingsUi from '@deepseek-ai/dsh-client-ui-primitives';
 import { RoutingSettings } from './settings-section.js';
 import { installRoutingNavIcon } from './nav-icon.js';
 export const name = 'dsh-auto-model-routing/client';
 export const inject = ['slots', 'connection'];
+const REPO_URL = 'https://github.com/prgrmrwy/dsh-auto-model-routing';
 // Same look as the official model picker trigger (dsh-client-ui-model-selection), values read from its bundle.
 const CSS = `
 .amr-root{position:relative;min-width:0;display:inline-flex}
@@ -17,6 +18,11 @@ const CSS = `
 .amr-trigger:focus-visible{box-shadow:0 0 0 2px var(--dsw-focus-ring-color,var(--dsw-alias-state-business-primary))}
 .amr-trigger:disabled{color:var(--dsw-alias-label-dimmed);cursor:default}
 .amr-icon{flex:none;color:var(--dsw-alias-label-caption)}
+.amr-trigger[data-locked=true]{color:var(--dsw-alias-label-dimmed);cursor:default}
+.amr-trigger[data-locked=true]:hover{background:0 0}
+.amr-info{color:var(--dsw-alias-label-caption)}
+.amr-card a{color:var(--dsw-alias-state-business-primary);text-decoration:none;overflow-wrap:anywhere}
+.amr-card a:hover{text-decoration:underline}
 .amr-trigger[data-disabled=true],.amr-trigger[data-disabled=true] .amr-icon{color:var(--dsw-alias-state-error-primary)}
 .amr-panel{position:absolute;z-index:1100;bottom:100%;left:0;padding-bottom:6px}
 .amr-card{width:280px;padding:10px 12px;border-radius:var(--dsw-radius-md);background:var(--dsw-alias-bg,#fff);color:var(--dsw-alias-label-primary);box-shadow:var(--dsw-elevation-prominent);font-size:12px;line-height:18px}
@@ -28,9 +34,9 @@ const CSS = `
 `;
 // Left-right swap arrows at the picker chevron's size; colour follows currentColor.
 function SwapIcon() {
-    return h('svg', { className: 'amr-icon', width: 14, height: 14, viewBox: '0 0 16 16', fill: 'none', stroke: 'currentColor', strokeWidth: 1.4, strokeLinecap: 'round', strokeLinejoin: 'round', 'aria-hidden': true }, h('path', { d: 'M3 5.5h9.5M10 3l2.5 2.5L10 8' }), h('path', { d: 'M13 10.5H3.5M6 8l-2.5 2.5L6 13' }));
+    return h('svg', { className: 'amr-icon amr-swap', width: 14, height: 14, viewBox: '0 0 16 16', fill: 'none', stroke: 'currentColor', strokeWidth: 1.4, strokeLinecap: 'round', strokeLinejoin: 'round', 'aria-hidden': true }, h('path', { d: 'M3 5.5h9.5M10 3l2.5 2.5L10 8' }), h('path', { d: 'M13 10.5H3.5M6 8l-2.5 2.5L6 13' }));
 }
-export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false, initialToast }) {
+export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false, initialToast, subscribeSelection }) {
     const [view, setView] = useState(initialView);
     const [open, setOpen] = useState(initiallyOpen);
     const [busy, setBusy] = useState(false);
@@ -77,6 +83,28 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
         const hide = setTimeout(() => setToast(undefined), TOAST_MS);
         return () => clearTimeout(hide);
     }, [toast]);
+    const refreshRef = useRef(() => { });
+    // Optional acceleration from the official picker store; retried on every state read so a late service still binds.
+    const subscription = useRef(undefined);
+    const subscribeRef = useRef(undefined);
+    subscribeRef.current = subscribeSelection;
+    const ensureSubscription = () => {
+        const subscribe = subscribeRef.current;
+        if (!subscribe || subscription.current?.sessionId === sessionId)
+            return;
+        subscription.current?.stop();
+        subscription.current = undefined;
+        let handle;
+        try {
+            const stop = subscribe(sessionId, () => { if (handle)
+                clearTimeout(handle); handle = setTimeout(() => refreshRef.current(), 80); });
+            subscription.current = { sessionId, stop: () => { stop(); if (handle)
+                    clearTimeout(handle); } };
+        }
+        catch {
+            subscription.current = undefined;
+        }
+    };
     useEffect(() => {
         let active = true;
         primed.current = false;
@@ -87,11 +115,20 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
             viewSession.current = sessionId;
             setView(value);
             setError('');
+            ensureSubscription();
         } }, reason => { if (active)
             setError(String(reason)); }); };
+        refreshRef.current = refresh;
         refresh();
         const poll = setInterval(refresh, 3000);
-        return () => { active = false; clearInterval(poll); clearTimeout(timer.current); };
+        return () => {
+            active = false;
+            clearInterval(poll);
+            clearTimeout(timer.current);
+            subscription.current?.stop();
+            subscription.current = undefined;
+            refreshRef.current = () => { };
+        };
     }, [rpc, sessionId]);
     const on = (event) => () => {
         clearTimeout(timer.current);
@@ -119,19 +156,64 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
     };
     const off = view?.disabled;
     const hint = off ? CATEGORY_HINTS[off.category] ?? off.category : undefined;
+    const locked = view?.autoSupport?.supported === false;
+    const support = view?.autoSupport;
+    const lockedModel = `${support?.provider ?? '未知 provider'} / ${support?.model ?? '未知模型'}`;
     // The panel sits inside the root (padding bridges the gap), so moving onto it keeps it open.
     const place = anchor.current ?? { left: (typeof window === 'undefined' ? 0 : window.innerWidth / 2), top: (typeof window === 'undefined' ? 0 : window.innerHeight) - 96 };
+    const panel = locked
+        ? [
+            h('p', { key: 'locked-title' }, h('strong', null, 'AUTO 不支持当前模型')),
+            h('p', { key: 'locked-model' }, `${lockedModel} 不在自动路由支持范围，已切换为 MANUAL。`),
+            h('p', { key: 'locked-note' }, '目前仅支持 Claude / GPT 系列模型；其他模型策略欢迎提交 issue 或 PR：'),
+            h('p', { key: 'locked-link' }, h('a', { href: REPO_URL, target: '_blank', rel: 'noopener noreferrer' }, REPO_URL)),
+            h('p', { key: 'locked-hint', className: 'amr-muted' }, '切回 Claude / GPT 系列模型后，可重新点击启用 AUTO。'),
+            error && h('p', { key: 'error', role: 'alert', className: 'amr-error' }, error),
+        ]
+        : [
+            h('p', { key: 'title' }, h('strong', null, view?.mode === 'AUTO' ? '自动调整推理强度' : '手动：不自动调整')),
+            h('p', { key: 'output' }, `本回合插件输出：${view?.effort ?? '未改写'} · ${view?.reason ?? '尚无决定'}`),
+            view?.decision && h('div', { key: 'latest', className: 'amr-latest' }, h('p', null, h('strong', null, '最近决策')), ...toastText(view.decision).map((line, index) => h('p', { key: index, className: index ? 'amr-muted' : undefined }, line))),
+            h('p', { key: 'policy', className: 'amr-muted' }, view?.policy === 'policy-v2' ? `AUTO · ${view.depth ?? '尚无判断'}；主模型不变，推理投入按请求应用；子任务自动选配。` : 'AUTO 的决定会同步到右侧模型选择；要手动指定 effort，请先切到 MANUAL。'),
+            off && h('p', { key: 'disabled', role: 'status', className: 'amr-error' }, `Jev 停用（${off.category}）：${hint}。${new Date(off.at).toLocaleString()}。点击 AUTO 重试。`),
+            h('p', { key: 'data', className: 'amr-muted' }, 'AUTO 会把脱敏任务文本发送给设置中所选的决策服务判档。'),
+            error && h('p', { key: 'error', role: 'alert', className: 'amr-error' }, error),
+        ];
     return h('span', { ref: (el) => { rootRef.current = el; remember(); }, className: 'amr-root', onMouseEnter: on('enter'), onMouseLeave: on('leave'), onFocus: on('focus'), onBlur: on('blur') }, h('style', null, CSS), h('button', {
-        type: 'button', className: 'amr-trigger', disabled: busy || !view, onClick: () => { void toggle(); },
-        'data-disabled': off ? 'true' : undefined, style: { border: 'none' },
-        title: off ? `Jev 判档已停用：${hint}。点击 AUTO 重试` : '点击切换自动推理强度',
-        'aria-label': `推理强度模式：${view?.mode ?? '加载中'}，点击切换`,
-    }, h('span', null, off ? `${view?.mode} · Jev 停用` : view?.mode ?? '…'), h(SwapIcon)), toast && !open && (typeof document === 'undefined'
+        type: 'button', className: 'amr-trigger', disabled: busy || !view, onClick: () => { if (locked)
+            return; void toggle(); },
+        'data-disabled': off ? 'true' : undefined, 'data-locked': locked ? 'true' : undefined,
+        'aria-disabled': locked ? 'true' : undefined, style: { border: 'none' },
+        title: locked ? 'AUTO 仅支持 Claude / GPT 系列模型；悬停查看说明' : off ? `Jev 判档已停用：${hint}。点击 AUTO 重试` : '点击切换自动推理强度',
+        'aria-label': locked ? `推理强度模式：${view?.mode ?? 'MANUAL'}（AUTO 不支持当前模型，不可切换）` : `推理强度模式：${view?.mode ?? '加载中'}，点击切换`,
+    }, h('span', null, locked ? (view?.mode ?? 'MANUAL') : off ? `${view?.mode} · Jev 停用` : view?.mode ?? '…'), locked ? h(IconInfoOutlineRegular, { size: 14, className: 'amr-icon amr-info' }) : h(SwapIcon)), toast && !open && (typeof document === 'undefined'
         ? h('div', { className: 'amr-toast', role: 'status', style: { pointerEvents: 'none' } }, ...toast.lines.map((line, index) => h('div', { key: index, className: index ? 'amr-muted' : undefined }, line)))
-        : createPortal(h('div', { className: 'amr-toast', role: 'status', style: { pointerEvents: 'none', left: place.left, top: place.top } }, h('style', null, CSS), ...toast.lines.map((line, index) => h('div', { key: index, className: index ? 'amr-muted' : undefined }, line))), document.body)), open && h('section', { className: 'amr-panel', 'aria-label': '自动路由状态' }, h('div', { className: 'amr-card' }, h('p', null, h('strong', null, view?.mode === 'AUTO' ? '自动调整推理强度' : '手动：不自动调整')), h('p', null, `本回合插件输出：${view?.effort ?? '未改写'} · ${view?.reason ?? '尚无决定'}`), view?.decision && h('div', { className: 'amr-latest' }, h('p', null, h('strong', null, '最近决策')), ...toastText(view.decision).map((line, index) => h('p', { key: index, className: index ? 'amr-muted' : undefined }, line))), h('p', { className: 'amr-muted' }, view?.policy === 'policy-v2' ? `AUTO · ${view.depth ?? '尚无判断'}；主模型不变，推理投入按请求应用；子任务自动选配。` : 'AUTO 的决定会同步到右侧模型选择；要手动指定 effort，请先切到 MANUAL。'), off && h('p', { role: 'status', className: 'amr-error' }, `Jev 停用（${off.category}）：${hint}。${new Date(off.at).toLocaleString()}。点击 AUTO 重试。`), h('p', { className: 'amr-muted' }, 'AUTO 会把脱敏任务文本发送给设置中所选的决策服务判档。'), error && h('p', { role: 'alert', className: 'amr-error' }, error))));
+        : createPortal(h('div', { className: 'amr-toast', role: 'status', style: { pointerEvents: 'none', left: place.left, top: place.top } }, h('style', null, CSS), ...toast.lines.map((line, index) => h('div', { key: index, className: index ? 'amr-muted' : undefined }, line))), document.body)), open && h('section', { className: 'amr-panel', 'aria-label': '自动路由状态' }, h('div', { className: 'amr-card' }, ...panel)));
 }
 const labelUi = { Tooltip, Menu, Icon: IconGaugeOutlineRegular };
 export function apply(ctx) {
+    // Optional acceleration only: the official picker's per-session store triggers an immediate state read.
+    // When the service is absent (or appears late) the chip's 3s poll remains the source of truth.
+    const pickerSubscriber = { current: undefined };
+    const subscribeSelection = (id, onChange) => pickerSubscriber.current ? pickerSubscriber.current(id, onChange) : () => { };
+    try {
+        const inject = ctx.inject;
+        if (typeof inject === 'function') {
+            inject.call(ctx, ['modelDirectories'], (scope) => {
+                const service = scope.modelDirectories;
+                if (!service)
+                    return;
+                pickerSubscriber.current = (id, onChange) => { try {
+                    return service.directoryFor(id).store.subscribe(onChange);
+                }
+                catch {
+                    return () => { };
+                } };
+                return () => { pickerSubscriber.current = undefined; };
+            });
+        }
+    }
+    catch { /* polling stays the fallback */ }
     ctx.slots.inject('settings.section', () => ctx.slots.register({
         name: 'settings.section', id: 'auto-model-routing', order: 300,
         label: () => '自动模型路由',
@@ -145,6 +227,6 @@ export function apply(ctx) {
     }, LabelButton));
     ctx.slots.inject('conversation.input.right', () => ctx.slots.register({
         name: 'conversation.input.right', id: 'auto-model-routing', order: chipOrder,
-        inject: (sessionId) => ({ sessionId, rpc: ctx.connection.rpc }),
+        inject: (sessionId) => ({ sessionId, rpc: ctx.connection.rpc, subscribeSelection }),
     }, RoutingChip));
 }

package/lib/client/label.d.ts

@@ -45,5 +45,5 @@ export declare function LabelButton({ sessionId, messageId, rpc, ui }: {
     messageId: string;
     rpc: Rpc;
     ui: LabelUi;
-}): import("react").DetailedReactHTMLElement<import("react").HTMLAttributes<HTMLElement>, HTMLElement> | null;
+}): import("react").DetailedReactHTMLElement<import("react").HTMLAttributes<HTMLSpanElement>, HTMLSpanElement> | null;
 export {};

package/lib/client/label.js

@@ -19,6 +19,7 @@ export function LabelButton({ sessionId, messageId, rpc, ui }) {
     const [view, setView] = useState(undefined);
     const [open, setOpen] = useState(false);
     const [error, setError] = useState('');
+    const exposureTarget = useRef(null);
     const alive = useRef(true);
     useEffect(() => () => { alive.current = false; }, []);
     useEffect(() => {
@@ -29,6 +30,40 @@ export function LabelButton({ sessionId, messageId, rpc, ui }) {
             setView(value); }, () => { });
         return () => { active = false; };
     }, [rpc, sessionId, messageId]);
+    useEffect(() => {
+        const target = exposureTarget.current;
+        if (view?.policy !== 'policy-v2' || !target || typeof IntersectionObserver === 'undefined')
+            return;
+        let visible = false, sent = false;
+        let timer;
+        const update = () => {
+            clearTimeout(timer);
+            timer = undefined;
+            if (!sent && visible && document.visibilityState === 'visible' && document.hasFocus()) {
+                timer = setTimeout(() => {
+                    if (!visible || document.visibilityState !== 'visible' || !document.hasFocus())
+                        return;
+                    sent = true;
+                    void rpc.call('/api', 'auto-model-routing.observe', { sessionId, messageId }).catch(() => { });
+                }, 3000);
+            }
+        };
+        const observer = new IntersectionObserver(entries => {
+            visible = entries.some(entry => entry.isIntersecting && entry.intersectionRatio >= 1);
+            update();
+        }, { threshold: 1 });
+        observer.observe(target);
+        document.addEventListener('visibilitychange', update);
+        window.addEventListener('focus', update);
+        window.addEventListener('blur', update);
+        return () => {
+            clearTimeout(timer);
+            observer.disconnect();
+            document.removeEventListener('visibilitychange', update);
+            window.removeEventListener('focus', update);
+            window.removeEventListener('blur', update);
+        };
+    }, [view?.policy, rpc, sessionId, messageId]);
     if (!view)
         return null;
     const choose = (label) => {
@@ -46,7 +81,7 @@ export function LabelButton({ sessionId, messageId, rpc, ui }) {
             setError('反馈未保存，请重试'); });
     };
     const button = h('button', { type: 'button', className: 'amr-label-action', 'aria-label': tooltipText(view), onClick: () => setOpen(value => !value) }, h(ui.Icon));
-    return h('span', null, h('style', null, CSS), h(ui.Menu, {
+    return h('span', { ref: exposureTarget }, h('style', null, CSS), h(ui.Menu, {
         open, side: 'top', portal: true, onClose: () => setOpen(false),
         // The official menu draws the check on the selected (labelled) row.
         ...((allocation ? view.feedback : view.label) ? { selectedId: (allocation ? view.feedback : view.label) } : {}),