dsh-workbuddy-xdpool 1.8.1 → 1.8.3

package/CHANGELOG.md

@@ -4,6 +4,98 @@
 
 版本号遵循 [语义化版本](https://semver.org/lang/zh-CN/)。
 
+## 1.8.3 — 快过期的积分先花掉，失效的登录别再刷屏
+
+> 这一版有两件事。一件是新的第五种账号使用方式：让积分包最快过期的那个账号先上，别把积分放过期了。另一件是修一个看着很吓人的毛病——账号失效之后，卡片上会顶着一大段英文，把整个面板撑爆，而且每刷新一次就再撞一次墙。
+
+### 新增：第五种模式「先花快到期的」
+
+积分有两种：**月度包**每月自动刷新，永远不会过期；**一次性包**有明确截止日，到点没用完就归零。以前插件不知道这个区别，轮着用谁都是随机——结果就是某个账号的一次性积分包在角落里静静过期，而你还在用另一个账号刷。
+
+现在多了一个选项：**优先使用积分包最快过期的账号**。
+
+- 数据直接来自上游本来就返回的 `nearestExpiryMs` 字段，**没有为这个模式多发任何一次请求**。
+- 月度包不参与排序——它不会过期，算它「紧急」纯属瞎紧张。
+- 没读过积分的账号排在最后。理由很简单：「没看过」不等于「有截止日」，拿未知去压真实截止日是错的。
+- 积分花完或过期之后，那个截止日会被**擦掉**，不是一直记着。否则池子会死死钉在一个积分已经空了的账号上，跟这个模式的目的正好相反。
+- 读不到任何到期信息时，行为退回「优先」模式。放心开。
+
+默认仍然是「优先」，没有偷偷改任何人的设置。
+
+### 模式选择换了排版：五个并排，说明放下面
+
+第五个模式一加进来，原来那个「两列网格 + 最后一个跨整行」的排版就散了：五个格子掉成两行，最后一个还孤零零撑满一行，看着像掉出了网格。
+
+现在改成**一行可换行的芯片**：
+
+- 面板放得下的时候，五个模式**一行排开**，宽度自动均分。设置面板常见的 700px 宽度也还读作一行选择，而不是在 760px 处突然塌成两列。
+- 真的放不下了才换行，换行后最后那个会自动撑满整行——这是 flex 的天然行为，不需要任何「最后一项跨行」的补丁规则，也就不会出现「明明第一行还装得下、却把第五个挤到第二行」的怪事。
+- 每块芯片里**只剩模式名**。五句话塞进五个小方块会让选择器变成五段散文，所以说明改成**行下方一行字，只讲当前选中的那个模式**。
+
+### 修：网关 401 不再打印一整段英文
+
+有人的卡片上，三个失效账号每个都顶着这么一段：
+
+```
+the WorkBuddy gateway rejected this credential (http 401). This usually means
+the account is using a stale sign-in the upstream no longer accepts: sign in
+again in the WorkBuddy desktop app, then pick the account on the plugin card.
+Run `dsh-workbuddy-xdpool doctor` to list every credential found.
+```
+
+话说得没错，但**它出现在面板正文里**，三个账号就是三段，把卡片撑到没法看。
+
+现在：
+
+- 这段文字折叠成一个**红色短标记**，完整原文进鼠标悬停提示。要看细节还在，不想看的时候不占地方。
+- 更关键的是**它现在真的会被当成「登录已失效」**。以前只有「刷新 token 被拒」才会判死，而如果凭据文件里写的过期时间还在未来，刷新那一步根本不会跑——于是每次卡片刷新都老老实实再撞一次 401，撞完再显示一次红字。现在只要**网关返回那段 HTML 拒绝页**，就按失效处理：账号进冷却、退出轮换。
+- 失效的账号**不再多吃一次签到探测**。以前每个死账号刷新时会被打两次（积分一次、签到一次），现在只打一次——签到那步直接跳过。
+- 会**自我纠正**：只要某次积分读成功了，就说明登录其实是好的，死标立刻清掉。不用手动点任何按钮。
+
+### 关于那三个账号
+
+顺带说一句：本机实测那三个报 401 的账号**是真的失效了**，不是插件误判——强制刷新时上游明确回 `session_dead (http 401): 12153:refresh token is invalid`。要恢复它们，去 WorkBuddy 桌面端重新登录，再回卡片点「重新检测账号」。
+
+## 1.8.2 — 免费的那个被限流了，旁边那个收费的同款还能用
+
+> 免费模型被限流的时候，插件以前只会说一句「所有账号都在冷却」。这句话没错，但它让用户以为自己只剩一条路：等。其实同一个模型往往还有一条**收费的**产品行，而且在限流器眼里它们是两个不同的模型——免费行被关小黑屋的时候，收费行通常毫发无伤。这个版本就是把这个「另一条路」写出来。
+
+### 同一个模型，凭什么有两行
+
+上游把一条产品线按**多行**卖，每行有自己的 id、自己的价格、自己的限流额度，但**显示名字是一样的**。实测两个网关：
+
+| 网关 | 免费行 | 收费行 |
+| --- | --- | --- |
+| 国际版 | `deepseek-v4.1-flash`（x0.00） | `deepseek-v4.1-flash-sg`（x0.03） |
+| 国内版 | `hy3`（x0.00） | `hy3-x`（x0.05） |
+
+注意 `-sg` 和 `-x` 跟免费行**没有任何前缀关系**，能把它们认成一家的只有那个相同的显示名。所以匹配是按名字来的，不是按 id 猜的。
+
+### 现在会怎么提示
+
+- **模型列表**：免费行后面挂一个小灰标「付费同款 deepseek-v4.1-flash-sg（0.03x）」，鼠标悬停解释「免费档先被限流；同一个模型的付费行不受这个限制影响」。灰的、低调的——这一行现在仍然是免费的那个，标签只是告诉你退路在哪，不是催你花钱。
+- **429 报错**：以前是「every account is rate-limited for model deepseek-v4.1-flash (4/4 cooling, tried 0)」，现在后面会接一句「deepseek-v4.1-flash-sg is the same model on a paid row (x0.03 credits) and is rate-limited separately; switch to it to keep going」。
+- **账号卡片**：某个账号上「`hy4-preview` cooling to 10:14」这种小标后面，也会跟上「→ 收费行的 id（倍率）还能用」。
+
+### 刻意没做的事
+
+- **收费行不标标签。** 只有倍率等于 0 的行才找孪生兄弟。否则「GLM-5.3」这种本身收费、又恰好有个更便宜的闪速版的模型，也会被挂上一个标签，最后满屏都是提示，等于没提示。
+- **没有收费同款就不说话。** 独苗模型（整条线只有它自己）不会有任何多余文字，也不会留下一个孤零零的分隔符。
+- **不把另一个免费行当退路。** 两个免费行是一起被限流的，切过去纯属换个人排队，说了等于骗人。
+- **倍率未知的行不算数。** 上游没给价格的兄弟行不能被承诺成「那个收费的」——用户切过去以为知道价钱，结果未必。
+
+### 顺带说清楚：国际版没有「每天领 30 积分」
+
+上一版已经查过一遍，这次把话说完整。国际版（`www.workbuddy.ai`）**没有**签到活动，也**没有** `daily-checkin` 这条路由：
+
+- `POST /v2/billing/meter/checkin-activity-status` 返回 200，但 `active: false`（字段在，活动没开）；
+- `GET /v2/billing/meter/get-user-resource` 直接 404（国内版才有这个接口）；
+- `/v2/billing/meter/checkin`、`/daily-checkin-activity`、`/checkin-activity` 全是 404。
+
+国际版真正有的是另一套东西：**成长任务板**（5 条：完成一次对话、安装技能、绑定微信、召唤专家、使用模板）和 **streak 连续打卡**（7 天 / 14 天 / 28 天档位，分别给 0 / 50 / 150 积分）。但它的字段形状跟国内版不一样（没有 credit / energy / target / current），插件目前解析不了，所以**没有**为国际版开自动化——没有可领的东西，开了也只是每天准时去撞 404。
+
+国内版的签到是正常的，多账号逐个领取的链路本来就有，不受本次改动影响。
+
 ## 1.8.1 — 一个过期的登录状态，不再拖着整个池子陪跑
 
 > 有人的卡片上，好几个账号下面顶着一大段红色英文，大意是「上游不认这个登录了，去桌面端重新登录」。话说得没错，可问题是：它只说了结论，没解决过程——那几个账号明明已经废了，每次请求还是要挨个撞一遍，撞得头破血流，旁边几个好端端的账号一次都轮不上。

package/README.md

@@ -50,6 +50,8 @@
 
 - **模型目录带标注**：列出池里能用的模型，附带积分倍率（如 `GLM-5.2 · x0.79`）、免费 / 限时免费 / 夜间折扣标签、能否收图和上下文窗口大小，全部跟随上游实时更新。
 
+- **被限流时告诉你退路**：免费档被 429 冷却时，卡片和报错都会点名**同名的付费行**（如国际版 `deepseek-v4.1-flash` 对 `deepseek-v4.1-flash-sg`、国内版 `hy3` 对 `hy3-x`）。上游按「行」分别限流，所以免费行被关小黑屋时，收费行通常还是通的。
+
 - **国内版 / 国际版自动适配**：按账号凭据里的登录域名自动选上游——国际版登录（`workbuddy.ai`）走 `www.workbuddy.ai`，国内版（默认）走 `copilot.tencent.com`。两种账号可以混在同一个池里。
 
 - **想手动做点啥也行**：卡片和命令行都提供「重新检测账号」「清除所有冷却」「每日签到」三个操作。

package/lib/bin.js

@@ -266,9 +266,18 @@ function isGatewayHtmlRejection(status, text) {
 	const head = text.slice(0, 512).toLowerCase();
 	return head.includes("<html") || head.includes("openresty") || head.includes("apisix");
 }
+/**
+* The sentence thrown for a gateway HTML rejection.
+*
+* Exported (as a prefix test) because callers outside this module have to tell
+* this specific failure apart from an ordinary network error: it is proof the
+* *credential* is no longer honoured, so the pool may mark the account dead
+* instead of re-probing it on every refresh.
+*/
+const GATEWAY_REJECTION_MESSAGE = "the WorkBuddy gateway rejected this credential (http 401).";
 async function readEnvelope(response) {
 	const text = await response.text();
-	if (isGatewayHtmlRejection(response.status, text)) throw new Error("the WorkBuddy gateway rejected this credential (http 401). This usually means the account is using a stale sign-in the upstream no longer accepts: sign in again in the WorkBuddy desktop app, then pick the account on the plugin card. Run `dsh-workbuddy-xdpool doctor` to list every credential found.");
+	if (isGatewayHtmlRejection(response.status, text)) throw new Error(`${GATEWAY_REJECTION_MESSAGE} This usually means the account is using a stale sign-in the upstream no longer accepts: sign in again in the WorkBuddy desktop app, then pick the account on the plugin card. Run \`dsh-workbuddy-xdpool doctor\` to list every credential found.`);
 	let parsed;
 	try {
 		parsed = JSON.parse(text);
@@ -2748,6 +2757,19 @@ var WorkBuddyAccountPool = class {
 	*/
 	creditBalances = /* @__PURE__ */ new Map();
 	/**
+	* Nearest expiry among each account's one-off credit packs, epoch ms.
+	*
+	* Credit packs are use-it-or-lose-it, so "which account should serve next"
+	* has a second honest answer besides "which one is idle": the one whose
+	* credits die soonest. Only one-off packs count — monthly packs refresh on
+	* their own cycle and are therefore never urgent — which is why this is
+	* written from the same `fetchCredits` reading that fills `creditBalances`.
+	*
+	* An absent entry means "never probed, or nothing is about to expire". Both
+	* are the same thing to `expiry` mode: not urgent, so it sorts last.
+	*/
+	creditExpiry = /* @__PURE__ */ new Map();
+	/**
 	* Last time each account served a request, epoch ms. Drives the idle term
 	* of the priority-mode weighting below: an account that just served loses to
 	* one that has been idle, so a small pool stops hammering a single account.
@@ -3050,6 +3072,33 @@ var WorkBuddyAccountPool = class {
 		return pool[pool.length - 1];
 	}
 	/**
+	* `expiry` mode: the account whose one-off credit packs die soonest.
+	*
+	* Credit packs are use-it-or-lose-it, so spending the dying ones first is
+	* strictly better than spreading the spend: an account that expires with
+	* credits left is money burnt, while an account whose packs have no deadline
+	* loses nothing by waiting. Two accounts expiring at the same instant fall
+	* back to the pool order, and an account with no known expiry sorts last —
+	* "we have not looked" must never outrank a real deadline.
+	*
+	* A pool where nobody has a known expiry therefore behaves exactly like
+	* priority, which is the honest degradation: without a reading there is
+	* nothing to sort by.
+	*/
+	pickByExpiry(pool) {
+		let best;
+		let bestAt = Number.POSITIVE_INFINITY;
+		for (const account of pool) {
+			const at = this.creditExpiry.get(account.id);
+			if (at === void 0) continue;
+			if (at < bestAt) {
+				best = account;
+				bestAt = at;
+			}
+		}
+		return best ?? pool[0];
+	}
+	/**
 	* Pick the account to serve a request.
 	*
 	* Two distributions, chosen by the `distribution` setting:
@@ -3064,6 +3113,8 @@ var WorkBuddyAccountPool = class {
 	* - **sticky**: one account per conversation, and a new conversation moves to
 	*   the next account in order. Keeps the upstream prompt cache warm within a
 	*   conversation while still spreading spend across conversations.
+	* - **expiry**: the account whose one-off credit packs expire soonest serves
+	*   first, so use-it-or-lose-it credits are spent before they vanish.
 	*
 	* In every mode an explicit user selection (`prefer`) heads the list, a
 	* cooling account is skipped for that model only, and an unrecognised setting
@@ -3092,7 +3143,7 @@ var WorkBuddyAccountPool = class {
 				const bound = this.affinityAccount(pool, conversationKey);
 				if (bound !== void 0 && await this.ensureFresh(bound)) return bound;
 			}
-			const account = this.distribution === "round-robin" || this.distribution === "sticky" ? this.pickRoundRobin(pool) : this.distribution === "balanced" ? this.pickByWeight(pool) : pool[0];
+			const account = this.distribution === "round-robin" || this.distribution === "sticky" ? this.pickRoundRobin(pool) : this.distribution === "balanced" ? this.pickByWeight(pool) : this.distribution === "expiry" ? this.pickByExpiry(pool) : pool[0];
 			if (account === void 0) return void 0;
 			if (!await this.ensureFresh(account)) continue;
 			if (this.distribution === "sticky" && conversationKey !== void 0 && conversationKey !== "") this.bindAffinity(conversationKey, account.id);
@@ -3134,11 +3185,26 @@ var WorkBuddyAccountPool = class {
 	* check has something to compare against. A reading for an unknown account is
 	* dropped: `scan()` rebuilds the account list and a stale id would otherwise
 	* accumulate forever.
+	*
+	* `nearestExpiryMs` is the same reading's nearest one-off pack deadline, or
+	* undefined when nothing is about to expire. It is written through, not
+	* merged: a pack that has been spent or has died disappears from the next
+	* reading, and keeping the stale deadline would pin the pool to an account
+	* whose credits are already gone — the opposite of what `expiry` mode wants.
 	*/
-	noteCredits(accountId, balance) {
+	noteCredits(accountId, balance, nearestExpiryMs) {
 		if (!Number.isFinite(balance)) return;
 		if (!this.accounts.some((account) => account.id === accountId)) return;
 		this.creditBalances.set(accountId, balance);
+		if (nearestExpiryMs === void 0 || !Number.isFinite(nearestExpiryMs)) {
+			this.creditExpiry.delete(accountId);
+			return;
+		}
+		this.creditExpiry.set(accountId, nearestExpiryMs);
+	}
+	/** Nearest one-off pack expiry for one account, or undefined when none known. */
+	creditExpiryOf(accountId) {
+		return this.creditExpiry.get(accountId);
 	}
 	/** Last known balance for one account, or undefined when never read. */
 	creditsOf(accountId) {
@@ -5243,8 +5309,9 @@ z.object({
 		"priority",
 		"round-robin",
 		"balanced",
-		"sticky"
-	]).default("priority").description("How requests are spread: priority (drain one), round-robin (in order), balanced (idle-weighted random), or sticky (one account per conversation, new conversations rotate)")),
+		"sticky",
+		"expiry"
+	]).default("priority").description("How requests are spread: priority (drain one), round-robin (in order), balanced (idle-weighted random), sticky (one account per conversation, new conversations rotate), or expiry (spend the account whose credit packs expire soonest)")),
 	disabledAccountIds: asVolatile(z.array(z.string()).default([]).description("Account ids excluded from the pool (empty = every discovered account participates)")),
 	creditReserves: asVolatile(z.dict(z.number().step(1).min(0)).default({}).description("Per-account credit floor: stop using an account once its balance reaches this value")),
 	enabledModelIds: asVolatile(z.array(z.string()).default([]).description("Legacy shared model-id list; used by a region that has no per-region selection yet")),

package/lib/client.js

@@ -182,8 +182,17 @@ window.__ModuleLoader__.load({
 /* "now: X" — repeats the answer in the header so the user does not have to
    scan four cards for the raised one. */
 .dsm-workbuddy-xdpool-dist-now{flex:none;padding:2px 10px;border-radius:999px;border:1px solid color-mix(in oklab, var(--dsw-alias-label-primary,#e6e6e6) 26%, transparent);background:color-mix(in oklab, var(--dsw-alias-label-primary,#e6e6e6) 10%, transparent);color:var(--dsw-alias-label-primary,#e6e6e6);font-size:12px;font-weight:600;line-height:18px;white-space:nowrap}
-.dsm-workbuddy-xdpool-dist-options{display:grid;grid-template-columns:repeat(2,minmax(0,1fr));gap:8px}
-.dsm-workbuddy-xdpool-dist-option{appearance:none;font:inherit;cursor:pointer;text-align:left;display:flex;flex-direction:column;justify-content:center;gap:4px;min-height:62px;border:1px solid color-mix(in oklab, var(--dsw-alias-border-l2,#3a3d45) 80%, transparent);border-radius:12px;padding:9px 13px;background:transparent;color:var(--dsw-alias-label-tertiary,#9aa0a8);transition:color .16s,border-color .16s,background .16s,box-shadow .16s}
+/* One row of equal chips, and let the row itself decide when it has to break.
+   flex-wrap plus a 112px basis keeps the five modes side by side for as long as
+   the card can hold them (a chip is ~112px: the longest label plus padding), so
+   a 700px panel still reads as one row of choices instead of a hard 760px cliff
+   dropping it to two columns. A flex line stretches its items to fill the row,
+   so when the chips finally do wrap, the odd last one spans the width on its own
+   — no last-child span rule, and therefore no way to accidentally strand the
+   last chip on a second row while the first row was still wide enough for all
+   five. */
+.dsm-workbuddy-xdpool-dist-options{display:flex;flex-wrap:wrap;gap:8px}
+.dsm-workbuddy-xdpool-dist-option{flex:1 1 112px;min-width:0;appearance:none;font:inherit;cursor:pointer;text-align:left;display:flex;flex-direction:column;justify-content:center;gap:3px;min-height:46px;border:1px solid color-mix(in oklab, var(--dsw-alias-border-l2,#3a3d45) 80%, transparent);border-radius:12px;padding:9px 11px;background:transparent;color:var(--dsw-alias-label-tertiary,#9aa0a8);transition:color .16s,border-color .16s,background .16s,box-shadow .16s}
 .dsm-workbuddy-xdpool-dist-option:hover:not(:disabled):not(.dsm-workbuddy-xdpool-dist-option-active){color:var(--dsw-alias-label-secondary,#c6c9d0);border-color:var(--dsw-alias-label-dimmed,#777);background:rgba(255,255,255,.03)}
 .dsm-workbuddy-xdpool-dist-option:focus-visible{outline:2px solid var(--dsw-alias-brand-primary,#5686fe);outline-offset:1px}
 /* Selected: raised onto layer-3 (the same surface the region tabs use when
@@ -192,11 +201,15 @@ window.__ModuleLoader__.load({
 .dsm-workbuddy-xdpool-dist-option-active{background:var(--dsw-alias-bg-layer-3,#2a2c33);border-color:color-mix(in oklab, var(--dsw-alias-label-primary,#e6e6e6) 34%, transparent);box-shadow:inset 3px 0 0 var(--dsw-alias-label-primary,#e6e6e6);color:var(--dsw-alias-label-primary,#e6e6e6)}
 .dsm-workbuddy-xdpool-dist-option-active .dsm-workbuddy-xdpool-dist-option-name{color:var(--dsw-alias-label-primary,#e6e6e6)}
 .dsm-workbuddy-xdpool-dist-option:disabled{cursor:default;opacity:.6}
-.dsm-workbuddy-xdpool-dist-option-top{display:flex;align-items:center;gap:7px;min-width:0}
-.dsm-workbuddy-xdpool-dist-option-name{font-size:12.5px;line-height:18px;font-weight:600;min-width:0;overflow:hidden;text-overflow:ellipsis;white-space:nowrap}
+.dsm-workbuddy-xdpool-dist-option-top{display:flex;align-items:center;gap:6px;min-width:0}
+/* The name is the only thing in a chip, so it may wrap rather than ellipsize:
+   a truncated mode name is unreadable, a two-line one is merely taller. */
+.dsm-workbuddy-xdpool-dist-option-name{font-size:12.5px;line-height:17px;font-weight:600;min-width:0;overflow-wrap:anywhere}
 /* "recommended" chip: quiet, and only on the mode this release argues for. */
-.dsm-workbuddy-xdpool-dist-option-badge{flex:none;margin-left:auto;padding:1px 7px;border-radius:999px;background:var(--dsw-alias-state-success-subtle,rgba(34,160,107,.14));color:var(--dsw-alias-state-success-primary,#22a06b);font-size:10px;font-weight:600;line-height:15px;white-space:nowrap}
-.dsm-workbuddy-xdpool-dist-option-hint{font-size:11px;line-height:16px;opacity:.85}
+.dsm-workbuddy-xdpool-dist-option-badge{flex:none;margin-left:auto;padding:1px 6px;border-radius:999px;background:var(--dsw-alias-state-success-subtle,rgba(34,160,107,.14));color:var(--dsw-alias-state-success-primary,#22a06b);font-size:10px;font-weight:600;line-height:15px;white-space:nowrap}
+/* One line of prose under the row explains the mode that is selected. Five
+   chips cannot each carry a sentence without becoming five paragraphs. */
+.dsm-workbuddy-xdpool-dist-hint{color:var(--dsw-alias-label-tertiary,#9aa0a8);font-size:11px;line-height:16px}
 .dsm-workbuddy-xdpool-usage-actions{display:flex;align-items:center;gap:8px;flex-wrap:wrap}
 
 /* Account list (each account = a labeled subpanel, same as dingminhua). */
@@ -256,6 +269,10 @@ window.__ModuleLoader__.load({
 .dsm-workbuddy-xdpool-account-meta-dead{color:var(--dsw-alias-state-error-primary,#ef4444)}
 .dsm-workbuddy-xdpool-account-modelcool{display:flex;flex-wrap:wrap;gap:6px;margin-top:2px}
 .dsm-workbuddy-xdpool-account-modelcool-chip{display:inline-flex;align-items:center;gap:4px;padding:1px 8px;border-radius:999px;font-size:11px;line-height:18px;background:var(--dsw-alias-state-warning-subtle,rgba(217,119,6,.12));color:var(--dsw-alias-state-warning-primary,#d97706)}
+/* "→ deepseek-v4.1-flash-sg (0.03x) still works" inside a cooling chip. Keeps
+   the warning tint of its parent but drops the weight: it is the way out, not
+   a second problem. */
+.dsm-workbuddy-xdpool-account-modelcool-twin{color:var(--dsw-alias-label-secondary,#c6c9d0);cursor:help}
 .dsm-workbuddy-xdpool-account-error{margin:0;color:var(--dsw-alias-state-error-primary,#ef4444);font-size:13px;line-height:20px}
 
 /* Two-column stats: packages on the left, total + check-in on the right. */
@@ -307,6 +324,10 @@ window.__ModuleLoader__.load({
 /* "活动至 10-31": the campaign's end date. Muted — it is context for planning,
    not a claim about the current price. */
 .dsm-workbuddy-xdpool-model-meta-promo{color:var(--dsw-alias-label-tertiary,#9aa0a8);font-size:11px;line-height:16px;opacity:.85}
+/* "付费同款 deepseek-v4.1-flash-sg (0.03x)": the way out when the free row is
+   rate-limited. Muted like the other meta chips — it is an alternative, not a
+   warning, and the row it sits on is still the free one. */
+.dsm-workbuddy-xdpool-model-meta-twin{color:var(--dsw-alias-label-tertiary,#9aa0a8);font-size:11px;line-height:16px;opacity:.85;cursor:help}
 /* Unreadable credential files: a warning tint, since it explains a smaller pool
    and "encrypted" is actionable (start the app once). */
 .dsm-workbuddy-xdpool-skipped{margin:6px 0 0;padding:7px 10px;border-radius:8px;background:var(--dsw-alias-state-warning-subtle,rgba(217,119,6,.12));display:flex;flex-direction:column;gap:3px}
@@ -362,9 +383,8 @@ window.__ModuleLoader__.load({
   .dsm-workbuddy-xdpool-checkin{align-items:stretch}
   .dsm-workbuddy-xdpool-checkin-meta{align-items:flex-start}
   .dsm-workbuddy-xdpool-checkin-bonus{text-align:left}
-  /* One mode per row: at this width a two-up grid leaves ~150px per card, and
-     the hint sentence wraps to four lines. */
-  .dsm-workbuddy-xdpool-dist-options{grid-template-columns:minmax(0,1fr)}
+  /* The chips wrap on their own (auto-fit), so the only thing left to do here
+     is stop reserving desktop height for rows that are now one chip tall. */
   .dsm-workbuddy-xdpool-dist-option{min-height:0}
 }
 
@@ -510,6 +530,43 @@ window.__ModuleLoader__.load({
 			return status?.kind === "free" || status?.kind === "night";
 		}
 		//#endregion
+		//#region src/siblings.ts
+		/**
+		* The cheapest PAID model sharing `model`'s display name, or undefined.
+		*
+		* Cheapest rather than first, because the point of the hint is "there is a
+		* cheap way out": offering the most expensive twin would make switching look
+		* like a trap. A sibling whose price is unknown is skipped — an unpriced row
+		* cannot be promised as "the paid one".
+		*/
+		function paidSiblingOf(model, models) {
+			let best;
+			for (const candidate of models) {
+				if (candidate.id === model.id) continue;
+				if (candidate.name !== model.name) continue;
+				const multiplier = candidate.multiplier;
+				if (multiplier === void 0 || multiplier <= 0) continue;
+				if (best === void 0 || multiplier < best.multiplier) best = {
+					id: candidate.id,
+					name: candidate.name,
+					multiplier
+				};
+			}
+			return best;
+		}
+		/**
+		* The paid sibling of a model, but ONLY when the model itself is free.
+		*
+		* The guard is the whole point: a `x0.79` row with a `x0.06` cousin is not a
+		* "free model with a paid twin", it is just a more expensive row, and labelling
+		* it would be noise on most of the list. The hint answers one question — "the
+		* free one is limited, what now".
+		*/
+		function paidAlternativeFor(model, models) {
+			if (model.multiplier !== 0) return void 0;
+			return paidSiblingOf(model, models);
+		}
+		//#endregion
 		//#region src/client/PoolCard.tsx
 		/**
 		* WorkBuddy XD Pool card contributed to DSH Plugin configuration.
@@ -630,22 +687,25 @@ window.__ModuleLoader__.load({
 			return String(value);
 		}
 		/**
-		* The four ways the pool can hand out accounts.
+		* The five ways the pool can hand out accounts.
 		*
 		* Order is the reading order of the picker: the recommended mode first, then
-		* the two cache-friendly extremes, then the two spreading modes.
+		* the two cache-friendly extremes, then the two spreading modes. The row is
+		* laid out one chip per mode, so this list length is also the column count.
 		*/
 		const DIST_OPTIONS = [
 			"sticky",
 			"priority",
 			"balanced",
-			"round-robin"
+			"round-robin",
+			"expiry"
 		];
 		/** Localized name of one distribution mode. */
 		function distLabel(option, t) {
 			if (option === "sticky") return t?.("row.distSticky") ?? "Per conversation";
 			if (option === "priority") return t?.("row.distPriority") ?? "Priority";
 			if (option === "balanced") return t?.("row.distBalanced") ?? "Balanced";
+			if (option === "expiry") return t?.("row.distExpiry") ?? "Expiring first";
 			return t?.("row.distRoundRobin") ?? "Round-robin";
 		}
 		/** One-line explanation of what a mode does to caching and spend. */
@@ -653,6 +713,7 @@ window.__ModuleLoader__.load({
 			if (option === "sticky") return t?.("row.distStickyHint") ?? "";
 			if (option === "priority") return t?.("row.distPriorityHint") ?? "";
 			if (option === "balanced") return t?.("row.distBalancedHint") ?? "";
+			if (option === "expiry") return t?.("row.distExpiryHint") ?? "";
 			return t?.("row.distRoundRobinHint") ?? "";
 		}
 		/** Build the draft from the server's selection + catalog flags. */
@@ -1356,47 +1417,51 @@ window.__ModuleLoader__.load({
 										className: "dsm-workbuddy-xdpool-dist",
 										role: "radiogroup",
 										"aria-label": t?.("row.distTitle") ?? "Account usage",
-										children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
-											className: "dsm-workbuddy-xdpool-dist-head",
-											children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-												className: "dsm-workbuddy-xdpool-dist-title",
-												children: t?.("row.distTitle") ?? "Account usage"
-											}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-												className: "dsm-workbuddy-xdpool-dist-now",
-												children: distLabel(status.distribution ?? "priority", t)
-											})]
-										}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
-											className: "dsm-workbuddy-xdpool-dist-options",
-											children: DIST_OPTIONS.map((option) => {
-												const active = (status.distribution ?? "priority") === option;
-												const label = distLabel(option, t);
-												const hint = distHint(option, t);
-												return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("button", {
-													type: "button",
-													role: "radio",
-													"aria-checked": active,
-													title: hint,
-													disabled: !modelsEditable,
-													className: "dsm-workbuddy-xdpool-dist-option" + (active ? " dsm-workbuddy-xdpool-dist-option-active" : ""),
-													onClick: () => {
-														setDistribution(option);
-													},
-													children: [/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
-														className: "dsm-workbuddy-xdpool-dist-option-top",
-														children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-															className: "dsm-workbuddy-xdpool-dist-option-name",
-															children: label
-														}), option === "sticky" ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-															className: "dsm-workbuddy-xdpool-dist-option-badge",
-															children: t?.("row.distRecommended") ?? "Recommended"
-														}) : null]
-													}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-														className: "dsm-workbuddy-xdpool-dist-option-hint",
-														children: hint
-													})]
-												}, option);
+										children: [
+											/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
+												className: "dsm-workbuddy-xdpool-dist-head",
+												children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+													className: "dsm-workbuddy-xdpool-dist-title",
+													children: t?.("row.distTitle") ?? "Account usage"
+												}), /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+													className: "dsm-workbuddy-xdpool-dist-now",
+													children: distLabel(status.distribution ?? "priority", t)
+												})]
+											}),
+											/* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
+												className: "dsm-workbuddy-xdpool-dist-options",
+												children: DIST_OPTIONS.map((option) => {
+													const active = (status.distribution ?? "priority") === option;
+													const label = distLabel(option, t);
+													const hint = distHint(option, t);
+													return /* @__PURE__ */ (0, react_jsx_runtime.jsx)("button", {
+														type: "button",
+														role: "radio",
+														"aria-checked": active,
+														title: hint,
+														disabled: !modelsEditable,
+														className: "dsm-workbuddy-xdpool-dist-option" + (active ? " dsm-workbuddy-xdpool-dist-option-active" : ""),
+														onClick: () => {
+															setDistribution(option);
+														},
+														children: /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+															className: "dsm-workbuddy-xdpool-dist-option-top",
+															children: [/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+																className: "dsm-workbuddy-xdpool-dist-option-name",
+																children: label
+															}), option === "sticky" ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+																className: "dsm-workbuddy-xdpool-dist-option-badge",
+																children: t?.("row.distRecommended") ?? "Recommended"
+															}) : null]
+														})
+													}, option);
+												})
+											}),
+											/* @__PURE__ */ (0, react_jsx_runtime.jsx)("p", {
+												className: "dsm-workbuddy-xdpool-dist-hint",
+												children: distHint(status.distribution ?? "priority", t)
 											})
-										})]
+										]
 									})
 								]
 							}), /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
@@ -1617,6 +1682,7 @@ window.__ModuleLoader__.load({
 							})(),
 							status?.accounts.map((account) => /* @__PURE__ */ (0, react_jsx_runtime.jsx)(AccountBlock, {
 								account,
+								models: status?.models ?? [],
 								...checkinBusyId === void 0 ? {} : { checkinBusyId },
 								onClaimCheckin: (accountId) => {
 									claimCheckin(accountId);
@@ -1723,6 +1789,7 @@ window.__ModuleLoader__.load({
 							children: [...status?.models ?? []].sort((a, b) => Number(isFreeNow(b, now, activeRegion)) - Number(isFreeNow(a, now, activeRegion))).map((model) => /* @__PURE__ */ (0, react_jsx_runtime.jsx)(ModelRow, {
 								model,
 								region: activeRegion,
+								paidTwin: paidAlternativeFor(model, status?.models ?? []),
 								t,
 								draft: modelDraft[model.id] ?? {
 									enabled: model.enabled,
@@ -1739,7 +1806,7 @@ window.__ModuleLoader__.load({
 			});
 		}
 		/** One account block: label + status tag + meta + optional credit panels. */
-		function AccountBlock({ account, t, checkinBusyId, onClaimCheckin, accountBusyId, onToggleDisabled, onIgnoreAccount, onSaveCreditReserve, reserveBusyId }) {
+		function AccountBlock({ account, models, t, checkinBusyId, onClaimCheckin, accountBusyId, onToggleDisabled, onIgnoreAccount, onSaveCreditReserve, reserveBusyId }) {
 			const isDisabled = account.disabled === true;
 			const isCooling = account.cooling === true;
 			const isDead = account.credentialDead === true;
@@ -1812,13 +1879,24 @@ window.__ModuleLoader__.load({
 								}) : null,
 								modelCooldowns.length > 0 ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("div", {
 									className: "dsm-workbuddy-xdpool-account-modelcool",
-									children: modelCooldowns.map((mc) => /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
-										className: "dsm-workbuddy-xdpool-account-modelcool-chip",
-										children: t?.("row.modelCooling", {
-											model: mc.modelId,
-											time: formatDateTime(mc.until)
-										}) ?? `${mc.modelId} cooling to ${formatDateTime(mc.until)}`
-									}, mc.modelId))
+									children: modelCooldowns.map((mc) => {
+										const coolingModel = models.find((model) => model.id === mc.modelId);
+										const twin = coolingModel === void 0 ? void 0 : paidAlternativeFor(coolingModel, models);
+										return /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("span", {
+											className: "dsm-workbuddy-xdpool-account-modelcool-chip",
+											children: [t?.("row.modelCooling", {
+												model: mc.modelId,
+												time: formatDateTime(mc.until)
+											}) ?? `${mc.modelId} cooling to ${formatDateTime(mc.until)}`, twin === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+												className: "dsm-workbuddy-xdpool-account-modelcool-twin",
+												title: t?.("row.paidTwinHint") ?? "Same model, paid row",
+												children: t?.("row.rateLimitedPaidTwin", {
+													id: twin.id,
+													rate: twin.multiplier.toFixed(2)
+												}) ?? `→ ${twin.id} (${twin.multiplier.toFixed(2)}x) still works`
+											})]
+										}, mc.modelId);
+									})
 								}) : null
 							]
 						}),
@@ -2044,7 +2122,8 @@ window.__ModuleLoader__.load({
 							className: "dsm-workbuddy-xdpool-checkin",
 							children: account.checkinError !== void 0 ? /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 								className: "dsm-workbuddy-xdpool-checkin-error",
-								children: account.checkinError
+								title: account.checkinError,
+								children: t?.("row.checkinUnavailable") ?? "check-in unavailable"
 							}) : checkin === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsxs)(react_jsx_runtime.Fragment, { children: [
 								/* @__PURE__ */ (0, react_jsx_runtime.jsxs)("div", {
 									className: "dsm-workbuddy-xdpool-checkin-meta",
@@ -2085,7 +2164,7 @@ window.__ModuleLoader__.load({
 		* context radios stay disabled rather than pretending an edit took hold. The
 		* draft lives in the parent, so this component only ever reports intent.
 		*/
-		function ModelRow({ model, region, t, draft, editable, onToggle, onToggleImage, onBudget }) {
+		function ModelRow({ model, region, paidTwin, t, draft, editable, onToggle, onToggleImage, onBudget }) {
 			const now = /* @__PURE__ */ new Date();
 			const tag = tagFor(model, now, region);
 			const promo = promoStatusFor(model, now, region);
@@ -2184,6 +2263,14 @@ window.__ModuleLoader__.load({
 							className: "dsm-workbuddy-xdpool-model-meta-promo",
 							children: promoUntil
 						}),
+						paidTwin === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+							className: "dsm-workbuddy-xdpool-model-meta-twin",
+							title: t?.("row.paidTwinHint") ?? "Same model, paid row",
+							children: t?.("row.paidTwin", {
+								id: paidTwin.id,
+								rate: paidTwin.multiplier.toFixed(2)
+							}) ?? `paid twin ${paidTwin.id} (${paidTwin.multiplier.toFixed(2)}x)`
+						}),
 						/* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 							className: "dsm-workbuddy-xdpool-model-cap",
 							children: t?.("row.modelOutput", { size: formatCapacity(model.maxOutputTokens) }) ?? `out ${formatCapacity(model.maxOutputTokens)}`
@@ -2315,6 +2402,7 @@ window.__ModuleLoader__.load({
 			"row.checkinStreakBonus": "day {days} bonus +{credit}",
 			"row.checkinClaimedReward": "Claimed +{credit} credits",
 			"row.checkinError": "Check-in failed: {message}",
+			"row.checkinUnavailable": "check-in unavailable",
 			"row.checkinAllHint": "Collect every account’s daily reward here — no need to switch accounts first.",
 			"row.modelsTitle": "Models",
 			"row.modelsSummary": "{count} model(s) in the live catalog",
@@ -2345,6 +2433,9 @@ window.__ModuleLoader__.load({
 			"row.nightFreeNow": "free until {time}",
 			"row.nightFreeLater": "free from {time}",
 			"row.promoUntil": "promo to {date}",
+			"row.paidTwin": "paid twin {id} ({rate}x)",
+			"row.paidTwinHint": "The free tier is rate-limited first; this same model has a paid row that keeps working",
+			"row.rateLimitedPaidTwin": "Tip: {id} is the same model on a paid row ({rate}x credits) and is limited separately — switch to it to keep going",
 			"row.skippedFiles": "{count} credential file(s) could not be read",
 			"row.skippedEncrypted": "encrypted — start WorkBuddy once so its key can be read",
 			"row.skippedUnreadable": "file could not be read",
@@ -2362,6 +2453,8 @@ window.__ModuleLoader__.load({
 			"row.distBalancedHint": "Draw at random, favouring the account idle longest",
 			"row.distSticky": "Per conversation",
 			"row.distStickyHint": "One account per chat, new chats take the next account — keeps the prompt cache warm",
+			"row.distExpiry": "Expiring first",
+			"row.distExpiryHint": "Spend the account whose credit packs expire soonest — use them before they are lost",
 			"row.distRecommended": "Recommended",
 			"row.resetCooldowns": "Clear all cooldowns",
 			"row.resetCooldownsBusy": "Clearing…",
@@ -2473,6 +2566,7 @@ window.__ModuleLoader__.load({
 			"row.checkinStreakBonus": "第 {days} 天额外 +{credit}",
 			"row.checkinClaimedReward": "已领取 +{credit} 积分",
 			"row.checkinError": "签到失败：{message}",
+			"row.checkinUnavailable": "签到不可用",
 			"row.checkinAllHint": "这里可以为每个账号分别领取每日签到奖励，无需先切换账号。",
 			"row.modelsTitle": "模型",
 			"row.modelsSummary": "实时目录中 {count} 个模型",
@@ -2503,6 +2597,9 @@ window.__ModuleLoader__.load({
 			"row.nightFreeNow": "免费至 {time}",
 			"row.nightFreeLater": "{time} 起免费",
 			"row.promoUntil": "活动至 {date}",
+			"row.paidTwin": "付费同款 {id}（{rate}x）",
+			"row.paidTwinHint": "免费档先被限流；同一个模型的付费行不受这个限制影响",
+			"row.rateLimitedPaidTwin": "提示：{id} 是同一个模型的付费行（{rate}x 积分），限流是分开算的 —— 切过去就能继续用",
 			"row.skippedFiles": "有 {count} 个凭据文件读不出来",
 			"row.skippedEncrypted": "已加密 —— 启动一次 WorkBuddy 才能取到密钥",
 			"row.skippedUnreadable": "文件读不出来",
@@ -2520,6 +2617,8 @@ window.__ModuleLoader__.load({
 			"row.distBalancedHint": "随机抽取，闲置越久的账号被选中概率越高",
 			"row.distSticky": "每对话固定",
 			"row.distStickyHint": "一个对话只用一个账号，新对话自动换下一个账号——保住上游缓存，命中率和速度都更好",
+			"row.distExpiry": "先花快到期的",
+			"row.distExpiryHint": "优先使用积分包最快过期的账号——把要过期的积分花掉，别浪费",
 			"row.distRecommended": "推荐",
 			"row.resetCooldowns": "清除所有冷却",
 			"row.resetCooldownsBusy": "正在清除…",

package/lib/index.d.ts

@@ -845,7 +845,7 @@ export declare function workbuddyAccountId(credential: Pick<WorkBuddyCredential,
 /** Every directory the pool should scan, in probe order. */
 export declare function candidateAuthDirs(env?: NodeJS.ProcessEnv): string[];
 /** How the pool chooses which account serves the next request. */
-type AccountDistribution = 'priority' | 'round-robin' | 'balanced' | 'sticky';
+type AccountDistribution = 'priority' | 'round-robin' | 'balanced' | 'sticky' | 'expiry';
 interface AccountPoolOptions {
   /** Logger for discovery and rotation events. */
   logger?: {
@@ -880,6 +880,10 @@ interface AccountPoolOptions {
    * - `sticky`: one account per conversation, and a new conversation moves to
    *   the next account in order. Keeps the upstream prompt cache warm inside a
    *   conversation while still spreading spend across conversations.
+   * - `expiry`: the account whose one-off credit packs expire soonest serves
+   *   first, so use-it-or-lose-it credits are spent before they die. Accounts
+   *   with no known expiry sort last, so a pool that has never been probed
+   *   behaves like `priority`.
    */
   distribution?: AccountDistribution;
 }
@@ -964,6 +968,19 @@ export declare class WorkBuddyAccountPool {
    * strand a healthy pool, and the first 402 still cools it as before.
    */
   private creditBalances;
+  /**
+   * Nearest expiry among each account's one-off credit packs, epoch ms.
+   *
+   * Credit packs are use-it-or-lose-it, so "which account should serve next"
+   * has a second honest answer besides "which one is idle": the one whose
+   * credits die soonest. Only one-off packs count — monthly packs refresh on
+   * their own cycle and are therefore never urgent — which is why this is
+   * written from the same `fetchCredits` reading that fills `creditBalances`.
+   *
+   * An absent entry means "never probed, or nothing is about to expire". Both
+   * are the same thing to `expiry` mode: not urgent, so it sorts last.
+   */
+  private creditExpiry;
   /**
    * Last time each account served a request, epoch ms. Drives the idle term
    * of the priority-mode weighting below: an account that just served loses to
@@ -1078,6 +1095,21 @@ export declare class WorkBuddyAccountPool {
    * spreads the very first picks instead of always returning index 0.
    */
   private pickByWeight;
+  /**
+   * `expiry` mode: the account whose one-off credit packs die soonest.
+   *
+   * Credit packs are use-it-or-lose-it, so spending the dying ones first is
+   * strictly better than spreading the spend: an account that expires with
+   * credits left is money burnt, while an account whose packs have no deadline
+   * loses nothing by waiting. Two accounts expiring at the same instant fall
+   * back to the pool order, and an account with no known expiry sorts last —
+   * "we have not looked" must never outrank a real deadline.
+   *
+   * A pool where nobody has a known expiry therefore behaves exactly like
+   * priority, which is the honest degradation: without a reading there is
+   * nothing to sort by.
+   */
+  private pickByExpiry;
   /**
    * Pick the account to serve a request.
    *
@@ -1093,6 +1125,8 @@ export declare class WorkBuddyAccountPool {
    * - **sticky**: one account per conversation, and a new conversation moves to
    *   the next account in order. Keeps the upstream prompt cache warm within a
    *   conversation while still spreading spend across conversations.
+   * - **expiry**: the account whose one-off credit packs expire soonest serves
+   *   first, so use-it-or-lose-it credits are spent before they vanish.
    *
    * In every mode an explicit user selection (`prefer`) heads the list, a
    * cooling account is skipped for that model only, and an unrecognised setting
@@ -1128,8 +1162,16 @@ export declare class WorkBuddyAccountPool {
    * check has something to compare against. A reading for an unknown account is
    * dropped: `scan()` rebuilds the account list and a stale id would otherwise
    * accumulate forever.
+   *
+   * `nearestExpiryMs` is the same reading's nearest one-off pack deadline, or
+   * undefined when nothing is about to expire. It is written through, not
+   * merged: a pack that has been spent or has died disappears from the next
+   * reading, and keeping the stale deadline would pin the pool to an account
+   * whose credits are already gone — the opposite of what `expiry` mode wants.
    */
-  noteCredits(accountId: string, balance: number): void;
+  noteCredits(accountId: string, balance: number, nearestExpiryMs?: number): void;
+  /** Nearest one-off pack expiry for one account, or undefined when none known. */
+  creditExpiryOf(accountId: string): number | undefined;
   /** Last known balance for one account, or undefined when never read. */
   creditsOf(accountId: string): number | undefined;
   /** The credit floor the user set for one account; 0 when unset. */
@@ -1607,7 +1649,7 @@ interface PoolWebAutomationEarnings {
  */
 type PoolRegion = 'cn' | 'global';
 /** How the pool spreads requests across its accounts. */
-type PoolDistribution = 'priority' | 'round-robin' | 'balanced' | 'sticky';
+type PoolDistribution = 'priority' | 'round-robin' | 'balanced' | 'sticky' | 'expiry';
 /**
  * The schedule every automation job falls back to.
  *
@@ -2579,10 +2621,13 @@ export interface Config {
    *   conversation throws away the upstream prompt cache (it is per tenant), so
    *   this keeps the cache warm while still spreading spend across
    *   conversations.
+   * - `expiry` picks the account whose one-off credit packs expire soonest, so
+   *   use-it-or-lose-it credits are spent before they die. Accounts with no
+   *   known expiry sort last, so an unprobed pool degrades to `priority`.
    *
    * Absent reads as `priority`.
    */
-  distribution?: 'priority' | 'round-robin' | 'balanced' | 'sticky';
+  distribution?: 'priority' | 'round-robin' | 'balanced' | 'sticky' | 'expiry';
   /**
    * Account ids switched off on the card. A disabled account is never picked
    * to serve a request, but it stays in the pool and on the card so it can be

package/lib/index.js

@@ -269,9 +269,22 @@ function isGatewayHtmlRejection(status, text) {
 	const head = text.slice(0, 512).toLowerCase();
 	return head.includes("<html") || head.includes("openresty") || head.includes("apisix");
 }
+/**
+* The sentence thrown for a gateway HTML rejection.
+*
+* Exported (as a prefix test) because callers outside this module have to tell
+* this specific failure apart from an ordinary network error: it is proof the
+* *credential* is no longer honoured, so the pool may mark the account dead
+* instead of re-probing it on every refresh.
+*/
+const GATEWAY_REJECTION_MESSAGE = "the WorkBuddy gateway rejected this credential (http 401).";
+/** Was this failure a gateway HTML rejection (a stale sign-in, not a hiccup)? */
+function isGatewayRejectionError(error) {
+	return (error instanceof Error ? error.message : String(error)).startsWith(GATEWAY_REJECTION_MESSAGE);
+}
 async function readEnvelope(response) {
 	const text = await response.text();
-	if (isGatewayHtmlRejection(response.status, text)) throw new Error("the WorkBuddy gateway rejected this credential (http 401). This usually means the account is using a stale sign-in the upstream no longer accepts: sign in again in the WorkBuddy desktop app, then pick the account on the plugin card. Run `dsh-workbuddy-xdpool doctor` to list every credential found.");
+	if (isGatewayHtmlRejection(response.status, text)) throw new Error(`${GATEWAY_REJECTION_MESSAGE} This usually means the account is using a stale sign-in the upstream no longer accepts: sign in again in the WorkBuddy desktop app, then pick the account on the plugin card. Run \`dsh-workbuddy-xdpool doctor\` to list every credential found.`);
 	let parsed;
 	try {
 		parsed = JSON.parse(text);
@@ -2856,6 +2869,19 @@ var WorkBuddyAccountPool = class {
 	*/
 	creditBalances = /* @__PURE__ */ new Map();
 	/**
+	* Nearest expiry among each account's one-off credit packs, epoch ms.
+	*
+	* Credit packs are use-it-or-lose-it, so "which account should serve next"
+	* has a second honest answer besides "which one is idle": the one whose
+	* credits die soonest. Only one-off packs count — monthly packs refresh on
+	* their own cycle and are therefore never urgent — which is why this is
+	* written from the same `fetchCredits` reading that fills `creditBalances`.
+	*
+	* An absent entry means "never probed, or nothing is about to expire". Both
+	* are the same thing to `expiry` mode: not urgent, so it sorts last.
+	*/
+	creditExpiry = /* @__PURE__ */ new Map();
+	/**
 	* Last time each account served a request, epoch ms. Drives the idle term
 	* of the priority-mode weighting below: an account that just served loses to
 	* one that has been idle, so a small pool stops hammering a single account.
@@ -3158,6 +3184,33 @@ var WorkBuddyAccountPool = class {
 		return pool[pool.length - 1];
 	}
 	/**
+	* `expiry` mode: the account whose one-off credit packs die soonest.
+	*
+	* Credit packs are use-it-or-lose-it, so spending the dying ones first is
+	* strictly better than spreading the spend: an account that expires with
+	* credits left is money burnt, while an account whose packs have no deadline
+	* loses nothing by waiting. Two accounts expiring at the same instant fall
+	* back to the pool order, and an account with no known expiry sorts last —
+	* "we have not looked" must never outrank a real deadline.
+	*
+	* A pool where nobody has a known expiry therefore behaves exactly like
+	* priority, which is the honest degradation: without a reading there is
+	* nothing to sort by.
+	*/
+	pickByExpiry(pool) {
+		let best;
+		let bestAt = Number.POSITIVE_INFINITY;
+		for (const account of pool) {
+			const at = this.creditExpiry.get(account.id);
+			if (at === void 0) continue;
+			if (at < bestAt) {
+				best = account;
+				bestAt = at;
+			}
+		}
+		return best ?? pool[0];
+	}
+	/**
 	* Pick the account to serve a request.
 	*
 	* Two distributions, chosen by the `distribution` setting:
@@ -3172,6 +3225,8 @@ var WorkBuddyAccountPool = class {
 	* - **sticky**: one account per conversation, and a new conversation moves to
 	*   the next account in order. Keeps the upstream prompt cache warm within a
 	*   conversation while still spreading spend across conversations.
+	* - **expiry**: the account whose one-off credit packs expire soonest serves
+	*   first, so use-it-or-lose-it credits are spent before they vanish.
 	*
 	* In every mode an explicit user selection (`prefer`) heads the list, a
 	* cooling account is skipped for that model only, and an unrecognised setting
@@ -3200,7 +3255,7 @@ var WorkBuddyAccountPool = class {
 				const bound = this.affinityAccount(pool, conversationKey);
 				if (bound !== void 0 && await this.ensureFresh(bound)) return bound;
 			}
-			const account = this.distribution === "round-robin" || this.distribution === "sticky" ? this.pickRoundRobin(pool) : this.distribution === "balanced" ? this.pickByWeight(pool) : pool[0];
+			const account = this.distribution === "round-robin" || this.distribution === "sticky" ? this.pickRoundRobin(pool) : this.distribution === "balanced" ? this.pickByWeight(pool) : this.distribution === "expiry" ? this.pickByExpiry(pool) : pool[0];
 			if (account === void 0) return void 0;
 			if (!await this.ensureFresh(account)) continue;
 			if (this.distribution === "sticky" && conversationKey !== void 0 && conversationKey !== "") this.bindAffinity(conversationKey, account.id);
@@ -3242,11 +3297,26 @@ var WorkBuddyAccountPool = class {
 	* check has something to compare against. A reading for an unknown account is
 	* dropped: `scan()` rebuilds the account list and a stale id would otherwise
 	* accumulate forever.
+	*
+	* `nearestExpiryMs` is the same reading's nearest one-off pack deadline, or
+	* undefined when nothing is about to expire. It is written through, not
+	* merged: a pack that has been spent or has died disappears from the next
+	* reading, and keeping the stale deadline would pin the pool to an account
+	* whose credits are already gone — the opposite of what `expiry` mode wants.
 	*/
-	noteCredits(accountId, balance) {
+	noteCredits(accountId, balance, nearestExpiryMs) {
 		if (!Number.isFinite(balance)) return;
 		if (!this.accounts.some((account) => account.id === accountId)) return;
 		this.creditBalances.set(accountId, balance);
+		if (nearestExpiryMs === void 0 || !Number.isFinite(nearestExpiryMs)) {
+			this.creditExpiry.delete(accountId);
+			return;
+		}
+		this.creditExpiry.set(accountId, nearestExpiryMs);
+	}
+	/** Nearest one-off pack expiry for one account, or undefined when none known. */
+	creditExpiryOf(accountId) {
+		return this.creditExpiry.get(accountId);
 	}
 	/** Last known balance for one account, or undefined when never read. */
 	creditsOf(accountId) {
@@ -5687,6 +5757,43 @@ function injectSummary(compacted, summary) {
 	];
 }
 //#endregion
+//#region src/siblings.ts
+/**
+* The cheapest PAID model sharing `model`'s display name, or undefined.
+*
+* Cheapest rather than first, because the point of the hint is "there is a
+* cheap way out": offering the most expensive twin would make switching look
+* like a trap. A sibling whose price is unknown is skipped — an unpriced row
+* cannot be promised as "the paid one".
+*/
+function paidSiblingOf(model, models) {
+	let best;
+	for (const candidate of models) {
+		if (candidate.id === model.id) continue;
+		if (candidate.name !== model.name) continue;
+		const multiplier = candidate.multiplier;
+		if (multiplier === void 0 || multiplier <= 0) continue;
+		if (best === void 0 || multiplier < best.multiplier) best = {
+			id: candidate.id,
+			name: candidate.name,
+			multiplier
+		};
+	}
+	return best;
+}
+/**
+* The paid sibling of a model, but ONLY when the model itself is free.
+*
+* The guard is the whole point: a `x0.79` row with a `x0.06` cousin is not a
+* "free model with a paid twin", it is just a more expensive row, and labelling
+* it would be noise on most of the list. The hint answers one question — "the
+* free one is limited, what now".
+*/
+function paidAlternativeFor(model, models) {
+	if (model.multiplier !== 0) return void 0;
+	return paidSiblingOf(model, models);
+}
+//#endregion
 //#region src/shim.ts
 /**
 * Loopback OpenAI-compatible endpoint with multi-account failover.
@@ -5872,6 +5979,22 @@ function createWorkBuddyShim(options) {
 	const server = createServer((req, res) => {
 		handle(req, res);
 	});
+	/**
+	* "The same model has a paid row" sentence for a 429 body, or undefined.
+	*
+	* Reads the catalog the shim was built with, so the hint names a row this
+	* gateway ACTUALLY serves — a CN-only id in a global 429 would be worse than
+	* no hint at all. Absent when the cooling model is itself paid, or when the
+	* family has no priced twin.
+	*/
+	function paidTwinHint(modelId) {
+		const models = catalog.current();
+		const model = models.find((candidate) => candidate.id === modelId);
+		if (model === void 0) return void 0;
+		const twin = paidAlternativeFor(model, models);
+		if (twin === void 0) return void 0;
+		return `${twin.id} is the same model on a paid row (x${twin.multiplier.toFixed(2)} credits) and is rate-limited separately; switch to it to keep going`;
+	}
 	const ready = new Promise((resolve, reject) => {
 		server.once("listening", () => resolve());
 		server.once("error", reject);
@@ -5902,7 +6025,7 @@ function createWorkBuddyShim(options) {
 		lastBalanceAt.set(account.id, now);
 		try {
 			const credits = await client.fetchCredits(account.credential);
-			pool.noteCredits(account.id, credits.total);
+			pool.noteCredits(account.id, credits.total, credits.nearestExpiryMs);
 		} catch {}
 	}
 	async function handle(req, res) {
@@ -5983,7 +6106,8 @@ function createWorkBuddyShim(options) {
 				if (why.reason === "cooling") {
 					const subject = modelId === void 0 ? "every WorkBuddy account is rate-limited" : `every account is rate-limited for model ${modelId}`;
 					const upstream = last === void 0 ? "" : ` — ${last.message.slice(0, 200)}`;
-					writeOpenAIError(res, 429, "soft_rate", `${subject} (${why.cooling}/${why.total} cooling, tried ${tried.length})${tried.length === 0 ? "" : `: ${tried.join(" → ")}`}${upstream}`);
+					const twin = modelId === void 0 ? void 0 : paidTwinHint(modelId);
+					writeOpenAIError(res, 429, "soft_rate", `${subject} (${why.cooling}/${why.total} cooling, tried ${tried.length})${tried.length === 0 ? "" : `: ${tried.join(" → ")}`}${upstream}` + (twin === void 0 ? "" : ` — ${twin}`));
 					return;
 				}
 				if (why.reason === "disabled") {
@@ -6560,9 +6684,11 @@ async function poolWebStatus(deps, region = "cn") {
 		const earned = deps.scheduler?.().earningsToday[account.id];
 		if (earned !== void 0) Object.assign(row, { automationToday: earned });
 		if (!row.cooling && row.credentialDead !== true) {
+			let deadNow = false;
 			try {
 				const credits = await deps.client.fetchCredits(account.credential);
-				deps.pool.noteCredits(account.id, credits.total);
+				deps.pool.noteCredits(account.id, credits.total, credits.nearestExpiryMs);
+				deps.pool.clearCredentialDead(account.id);
 				Object.assign(row, { credits: {
 					total: credits.total,
 					packages: credits.packages,
@@ -6570,9 +6696,14 @@ async function poolWebStatus(deps, region = "cn") {
 					...credits.nearestExpiryMs === void 0 ? {} : { nearestExpiryMs: credits.nearestExpiryMs }
 				} });
 			} catch (error) {
+				if (isGatewayRejectionError(error)) {
+					deps.pool.penalizeCredentialDead(account.id);
+					deadNow = true;
+					Object.assign(row, { credentialDead: true });
+				}
 				Object.assign(row, { creditsError: safeMessage(error) });
 			}
-			try {
+			if (!deadNow) try {
 				const checkin = await deps.client.fetchCheckinStatus(account.credential);
 				const web = {
 					active: checkin.active,
@@ -7317,8 +7448,9 @@ const Config = z.object({
 		"priority",
 		"round-robin",
 		"balanced",
-		"sticky"
-	]).default("priority").description("How requests are spread: priority (drain one), round-robin (in order), balanced (idle-weighted random), or sticky (one account per conversation, new conversations rotate)")),
+		"sticky",
+		"expiry"
+	]).default("priority").description("How requests are spread: priority (drain one), round-robin (in order), balanced (idle-weighted random), sticky (one account per conversation, new conversations rotate), or expiry (spend the account whose credit packs expire soonest)")),
 	disabledAccountIds: asVolatile(z.array(z.string()).default([]).description("Account ids excluded from the pool (empty = every discovered account participates)")),
 	creditReserves: asVolatile(z.dict(z.number().step(1).min(0)).default({}).description("Per-account credit floor: stop using an account once its balance reaches this value")),
 	enabledModelIds: asVolatile(z.array(z.string()).default([]).description("Legacy shared model-id list; used by a region that has no per-region selection yet")),

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-workbuddy-xdpool",
   "displayName": "DSH WorkBuddy XD Pool",
   "description": "Merge every locally signed-in WorkBuddy account into DeepSeek Harness as one auto-failing-over model pool (multi-account rotation, live credits, daily check-in and model catalog).",
-  "version": "1.8.1",
+  "version": "1.8.3",
   "license": "MIT",
   "author": "XDTrees",
   "homepage": "https://github.com/XDTrees/dsh-workbuddy-xdpool#readme",