@thincoder/core 0.9.4 → 0.10.1

package/memory-status.mjs

@@ -0,0 +1,69 @@
+/**
+ * memory-status.mjs — 记忆 / 索引状态**只读出口**（R2 · 桌面功能对位批 · 台账 #412 正解）：三表计数一处出。
+ *
+ * 判据（父侧裁 —— 端侧不直读核内表；`docs/desktop/design/PROJECT.md` §10 N 行）：
+ * - **只读**：本档零写面（无 DDL ∕ DML ∕ 迁移）+ 零动作面（无 sync ∕ reindex ∕ 建库）；
+ *   句柄装配归调用面（`memory` 首参 = 已开句柄）。
+ * - **零端名**：键名 = 表名 ∕ 通用计数词，无端侧语汇 —— 各端同面同形（第二消费面即本出口本身）。
+ * - **origin 归一**：`origin` 给定 ⇒ 三表皆按**归一键**限定（核写缝一律归一 —— `memory/origin.mjs`；
+ *   读面不归一 ⇒ 读不到本项目行）；缺 ∕ 非串 ∕ 空串 ⇒ 全库计数（回执 `origin: null`）。
+ *
+ * 回执：
+ * ```
+ * { origin, tables: { files, code: { files, chunks }, doc: { files, chunks } },
+ *   totals: { files, chunks }, indexed, dbBytes, origins: [{ origin, code, doc }] }
+ * ```
+ * `tables.files` = 记忆层（`files` 表：project ∕ team 两层 markdown 记忆）行数；`code` ∕ `doc` 各 =
+ * 去重文件数（`COUNT(DISTINCT path)`）与分块数（`COUNT(*)`）；`totals` = code + doc 合计（端侧既有
+ * 索引读数同口径）；`indexed` = 合计文件数 > 0（「本项目已建索引」读数 —— 读数非承诺）。
+ * P1 新增读数（§6.14 面③）：`dbBytes` = 库文件 `statSync` 字节数（非文件型 ∕ 不可读 ⇒ null）；
+ * `origins` = 逐 origin 行数（code ∕ doc 两表并集，键 = 库内**原样** origin，按名排序）。
+ */
+import { statSync } from "node:fs"
+import { normalizeOrigin } from "./memory/origin.mjs"
+
+/** 单表计数：`key` 非 null ⇒ 追加 `WHERE origin = ?`（三表皆有 origin 列 —— v5 ∕ v8 起）。 */
+function countOf(memory, base, key) {
+  const sql = key === null ? base : `${base} WHERE origin = ?`
+  return key === null ? memory.db.prepare(sql).get() : memory.db.prepare(sql).get(key)
+}
+
+/** 三表计数只读出口（形见档头；`memory` = `createMemory()` 回执句柄）。 */
+export function memoryStatus(memory, { origin } = {}) {
+  const normalized = normalizeOrigin(origin)
+  const scope = typeof normalized === "string" && normalized.length > 0 ? normalized : null
+
+  const files = countOf(memory, "SELECT COUNT(*) AS n FROM files", scope)?.n ?? 0
+  const codeRow = countOf(memory, "SELECT COUNT(DISTINCT path) AS files, COUNT(*) AS chunks FROM code_chunks", scope) ?? {}
+  const docRow = countOf(memory, "SELECT COUNT(DISTINCT path) AS files, COUNT(*) AS chunks FROM doc_chunks", scope) ?? {}
+  const code = { files: codeRow.files ?? 0, chunks: codeRow.chunks ?? 0 }
+  const doc = { files: docRow.files ?? 0, chunks: docRow.chunks ?? 0 }
+  const totalFiles = code.files + doc.files
+
+  return {
+    origin: scope,
+    tables: { files, code, doc },
+    totals: { files: totalFiles, chunks: code.chunks + doc.chunks },
+    indexed: totalFiles > 0,
+    dbBytes: dbBytesOf(memory),
+    origins: originRows(memory),
+  }
+}
+
+/** 库文件字节数（P1——`statSync` 实读；非文件型 ∕ 不可读 ⇒ null）。 */
+function dbBytesOf(memory) {
+  try { return memory.dbPath ? statSync(memory.dbPath).size : null } catch { return null }
+}
+
+/** 逐 origin 行数（P1——code ∕ doc 两表并集；键 = 库内原样 origin，按名排序）。 */
+function originRows(memory) {
+  const map = new Map()
+  for (const [table, key] of [["code_chunks", "code"], ["doc_chunks", "doc"]]) {
+    for (const r of memory.db.prepare(`SELECT origin, COUNT(*) AS n FROM ${table} GROUP BY origin`).all()) {
+      const cur = map.get(r.origin) ?? { origin: r.origin, code: 0, doc: 0 }
+      cur[key] = Number(r.n)
+      map.set(r.origin, cur)
+    }
+  }
+  return [...map.values()].sort((a, b) => (a.origin < b.origin ? -1 : a.origin > b.origin ? 1 : 0))
+}

package/memory.mjs

@@ -19,3 +19,6 @@ export { gitSync, codeSync, markIndexedCommit, codeSearch, ensureCodeEmbeddings,
 
 // doc sync + search + memoryTools
 export { docSync, docSearch, ensureDocEmbeddings, docSearchTool, memoryTools } from "./memory/docs.mjs"
+
+// status read-only export (three-table counts single point — R2)
+export { memoryStatus } from "./memory-status.mjs"

package/model-specs.mjs

@@ -17,9 +17,9 @@
  * partialMode:       Kimi/Qwen Partial Mode truncation continuation (assistant message with partial:true)
  * prefixMode:        DeepSeek Prefix Completion truncation continuation (uses /beta endpoint, with prefix:true)
  * multimodal:        whether multimodal (image/vision input supported)
- * cacheMode:         context caching mode: "auto"=automatic / "prompt"=needs explicit / "none"=unsupported
  * thinkApi:          thinking API type: "type"=thinking.type field / "effort"=reasoning_effort field
  * thinkEnabledValue: when thinkApi is "type", the value used to enable thinking (default "enabled"; MiniMax uses "adaptive")
+ * thinkAlwaysOn:     server-enforced thinking (true = always on — no off path; single source = MODEL-SPECS.md §16.3)
  * reasoningEcho:     reasoning_content cross-turn echo strategy: "required"=must echo (error if missing) / "optional"=echo optional (default: don't echo)
  * reasoningEffortEnum: valid reasoning_effort enum values (if undeclared, no validation — passed through as-is)
  * tempRange:         valid temperature range [min, max] (if undeclared, no clamping)
@@ -30,67 +30,99 @@ const MODEL_SPECS = [
   // beta, automatic disk cache, vision. The two legacy names below are RETIRED — the server
   // still accepts them and serves them from V4.1-Flash today (no switch window), so their
   // rows carry the V4.1-Flash parameters (v4-flash gains multimodal).
-  ["deepseek-flash",    { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-flash",    { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
 // deepseek-v4.1-flash = qwen-plan channel name for DeepSeek V4.1-Flash (token-plan GET /models verified 2026-09-15) — same fields as the row above; ".1" ≠ "-" so plain prefix lookup misses the existing rows (D-PR25)
-  ["deepseek-v4.1-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
-  // deepseek-v4-pro (V4-Pro-0813): fields unchanged — vision capability NOT verified, so no
-  // multimodal (conservative). From 2026-09-14 12:00 Beijing all requests route to V4.1-Flash.
-  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2] }],
+  ["deepseek-v4.1-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  // deepseek-v4-pro (V4-Pro-0813): fields unchanged — 视觉面 **未探**（不声明 multimodal——保守；
+  // 补探登记见 MODEL-SPECS.md §13.4）. From 2026-09-14 12:00 Beijing all requests route to V4.1-Flash.
+  ["deepseek-v4-pro",   { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2] }],
   // deepseek-v4-flash: RETIRED name — still accepted, served by V4.1-Flash today → aligned row
-  ["deepseek-v4-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-v4-flash", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
   // deepseek-v4-flash-vision-exp: retired experimental vision name — still accepted (V4.1-Flash)
-  ["deepseek-v4-flash-vision-exp", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
+  ["deepseek-v4-flash-vision-exp", { context: 1_000_000, maxOutput: 384_000, thinking: true,  prefixMode: true,  thinkApi: "type", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 2], multimodal: true }],
   // Kimi series
-  ["kimi-k3",           { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["kimi-k3",           { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
   // Qwen router prefixes model IDs with provider namespace: kimi/kimi-k3 → kimi-k3 (IK7K4V)
-  ["kimi/kimi-k3",      { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["kimi/kimi-k3",      { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
   // Kimi For Coding endpoint uses the short model ID "k3" (same specs as kimi-k3) — IK5VGJ
-  ["k3",                { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true, cacheMode: "auto",  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["k3",                { context: 1_000_000, maxOutput: 131_072, thinking: true,  partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  // kimi-for-coding / k3-256k / kimi-for-coding-highspeed（Kimi For Coding 端点三名 · 2026-09-25 批 §14.2 #7-9 · 活测 = §1.4 服务端 200 /models）：
+  // context = **活测**逐行；maxOutput = **族沿用 k3 行**（点名来源行——本注即标级）；effort 枚举 ["low","high","max"] = **活测**，
+  // 默认档 max（kimi-for-coding）/ high（k3-256k）落端差表；机制位（partialMode / thinkApi / reasoningEcho）**族沿用 k3 行**
+  // （有则沿、无则不声明）；能力位按 §1.4 现盘口径（image ✓ ⇒ multimodal ✓）。highspeed = 尺寸行 + image ✓（§1.4）——
+  // 活测**无 effort 块** ⇒ thinking / 枚举 / 默认档均不声明；与 kimi-k2.7-code-highspeed（128_000）分值分行（防合并）。
+  ["kimi-for-coding-highspeed", { context: 262_144,   maxOutput: 131_072, multimodal: true }],
+  ["kimi-for-coding",           { context: 1_048_576, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  ["k3-256k",                   { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, thinkApi: "effort", reasoningEcho: "required", reasoningEffortEnum: ["low", "high", "max"] }],
+  // kimi-k2.6 / kimi-k2.7-code / kimi-k2.7-code-highspeed（三行同形 · 2026-09-24 bench 名单批）：temperature **仅受理 1** = **校验级**
+  // （400 原文「invalid temperature: only 1 is allowed for this model」· roster-expand §2.2）；尺寸位 = **未取证**
+  // （沿现盘兜底值——建行不改变生效值）；**不得声明 tempRange**（KD-31 例外档不变量：入档值 = 实发值）；枚举 / 视觉面 = **未探**。
+  ["kimi-k2.6",                { context: 128_000, maxOutput: 32_000 }],
+  ["kimi-k2.7-code",           { context: 128_000, maxOutput: 32_000 }],
+  ["kimi-k2.7-code-highspeed", { context: 128_000, maxOutput: 32_000 }],
   // GLM series
   // GLM-5.3: thinking always-on (no "disabled"); effort converges to low/high/max — NOT the
-  //          7-level glm-5.2 enum (verified vs docs.bigmodel.cn GLM-5.3 page, 2026-08)
-  ["glm-5.3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5.3-flash",     { context: 1_000_000, maxOutput: 128_000, thinking: true, multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  //          7-level glm-5.2 enum (verified vs docs.bigmodel.cn GLM-5.3 page, 2026-08).
+  //          视觉面 **未探**（不声明 multimodal——保守；补探登记见 MODEL-SPECS.md §13.4）。
+  //          该事实 = 机制位 `thinkAlwaysOn`（单源 = `docs/core/design/MODEL-SPECS.md` §16.3）——**官方口径**
+  //          （智谱 GLM-5.3 文档：thinking always-on、无 `disabled`）。
+  ["glm-5.3",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkAlwaysOn: true, thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  // glm-5.3-flash：始终思考——该事实 = 机制位 `thinkAlwaysOn`（单源 = `docs/core/design/MODEL-SPECS.md` §16.3）
+  //          ——**族沿用**（族据 = `glm-5.3` 行）；其余字段见其下行注与规格面。
+  ["glm-5.3-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  thinkAlwaysOn: true, multimodal: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
   // glm-5.3-flashx（GLM-5.3-FlashX，2026-09 上线）——独立行（不再蹭 flash 前缀行）：maxOutput 131_072 =
   // **校验级**（400「max_tokens…限制数值范围[1,131072]」）；effort 枚举 ["low","high","max"] = **校验级**
   // （400 原文点名：low/high/max 受理，其余拒绝）。multimodal = **实测**（8×8 纯红 PNG → 答「红色」）。
   // thinking = **实测**且**始终思考**（`thinking:{type:"disabled"}` → 400；裸请求默认开）——off 动作在
-  // UI 侧为 no-op（服务端无关闭路径）。其余 = **族沿用** flash 行（context 网络口径未探边；
-  // cacheMode / thinkApi / reasoningEcho / tempRange / noUsageStream = 前缀命中路径日常在用）。
-  // flash 行 128_000 系其自身口径，不得对齐本行（AC-3 零回归）。
-  ["glm-5.3-flashx",    { context: 1_000_000, maxOutput: 131_072, thinking: true, multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5.2",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-5",             { context: 1_000_000, maxOutput: 128_000, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
-  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
+  // UI 侧为 no-op（服务端无关闭路径）。该事实 = 机制位 `thinkAlwaysOn`（单源 = `docs/core/design/MODEL-SPECS.md`
+  // §16.3）——**实测**（`thinking:{type:"disabled"}` → 400；裸请求默认开）。其余 = **族沿用** flash 行（context 网络口径未探边；
+  // thinkApi / reasoningEcho / tempRange / noUsageStream = 前缀命中路径日常在用）。
+  // flash 行 maxOutput 131_072 = **校验级**（#20 批 §1.2 同源 · 2026-09-25 批 §14.2 #1 收正）——两行独立（改一行不动另一行）。
+  ["glm-5.3-flashx",    { context: 1_000_000, maxOutput: 131_072, thinking: true,  thinkAlwaysOn: true, multimodal: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["low", "high", "max"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-5.2",           { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-5",             { context: 1_000_000, maxOutput: 128_000, thinking: true,  thinkApi: "type", reasoningEcho: "optional", reasoningEffortEnum: ["max", "xhigh", "high", "medium", "low", "minimal", "none"], tempRange: [0, 1], noUsageStream: true }],
+  ["glm-4",             { context: 128_000,   maxOutput: 32_000,  thinking: true,  thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
+  // glm-4.5-air（智谱轻档 · 2026-09-24 bench 名单批）：temperature 0 受理 = **实测**（roster-expand §2.2）；其余 = **族沿用 glm-4 行**
+  // （逐名官方口径未取证）；枚举 / 视觉面 = **未探**（不声明）。medium 直发实测受理（batch-params-judge §1.5）。
+  ["glm-4.5-air", { context: 128_000, maxOutput: 32_000, thinking: true, thinkApi: "type", reasoningEcho: "optional", tempRange: [0, 1], noUsageStream: true }],
   // GPT series
-  ["gpt-5.6-sol",       { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "prompt" }],
-  ["gpt-5.6",           { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "prompt" }],
-  ["gpt-4.1",           { context: 1_000_000, maxOutput: 128_000, thinking: false, cacheMode: "prompt" }],
-  ["gpt-4o",            { context: 128_000,   maxOutput: 16_000,  thinking: false, multimodal: true, cacheMode: "prompt" }],
+  ["gpt-5.6-sol",       { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true }],
+  ["gpt-5.6",           { context: 1_050_000, maxOutput: 128_000, thinking: false, multimodal: true }],
+  ["gpt-4.1",           { context: 1_000_000, maxOutput: 128_000, thinking: false }],
+  ["gpt-4o",            { context: 128_000,   maxOutput: 16_000,  thinking: false, multimodal: true }],
   // Qwen series
-  // qwen3.7-max rejects image parts outright (DashScope 400 "Unexpected item type in content") — text-only
-  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["xhigh", "high"], tempRange: [0, 2] }],
+  // qwen-flash / qwen-vl-max（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧实测）：**尺寸行**——maxOutput 32_768 = **实测**
+  // （服务端直报「Range of max_tokens should be [1, 32768]」；131072 / 262144 均 400）。context 未取到 ⇒ 零口径——取兜底值
+  // 128_000（与建行前生效值同值，§14.3 兜底语义续效）；thinking / 视觉有读数（默认关可开 / 不支持）但不落字段（§14.3 五名口径）。
+  ["qwen-flash",        { context: 128_000, maxOutput: 32_768 }],
+  ["qwen-vl-max",       { context: 128_000, maxOutput: 32_768 }],
+  // qwen3.7-max：**实测拒图像**（DashScope 400 "Unexpected item type in content"——真值彩图探针，2026-09-20）
+  // ⇒ 无视觉 = **不声明**（T-7 语义单一：multimodal 仅 true / undefined；补探已毕，无需再探）
+  ["qwen3.7-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true,  thinkApi: "effort", reasoningEffortEnum: ["xhigh", "high"], tempRange: [0, 2] }],
   // qwen3.7-flash / qwen3.8-flash (2026-09-20): thinking + effort enum + maxOutput + image intake
   // are server-MEASURED (batch 2026-09-20-qwen-flash-specs §1.2 — bare request carries reasoning_tokens
   // 166/27; 400 literal lists the enum; max_tokens range [1, 131072]). 3.7-flash enum has NO "max"
   // (6 levels), 3.8-flash adds it (7); enum order = the server's literal order, first member "none"
   // (D-5) — never re-sort by strength. context = doc/web statement (1M class, not API-measured);
-  // partialMode / cacheMode / thinkApi / tempRange = same-family carry-over (not independently
+  // partialMode / thinkApi / tempRange = same-family carry-over (not independently
   // measured on these rows); reasoningEcho stays undeclared (cross-turn echo unverified — R-4).
-  ["qwen3.7-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
-  ["qwen3.8-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
-  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["xhigh", "medium", "low"], tempRange: [0, 2] }],
+  ["qwen3.7-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  // qwen3.7-plus（2026-09-24 bench 名单批）：context = **官方口径（网络转述）**（2026-09-20 qwen-flash-specs 批 AC-6 已裁允许入表——「100 万长文档」族点名）；
+  // maxOutput / partialMode / thinkApi / tempRange / thinking = **族沿用 qwen3.7-flash 行**；枚举 / 视觉面 = **未探**（不声明）。
+  ["qwen3.7-plus",      { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true,  thinkApi: "effort", tempRange: [0, 2] }],
+  ["qwen3.8-flash",     { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-max",       { context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["xhigh", "medium", "low"], tempRange: [0, 2] }],
   // qwen3.8-omni-flash / qwen3.8-27b — parent 80-token probe (batch §1.8-①): reasoning_content
   // present, reasoning_tokens 49, `effort:"none"` removes it ⇒ thinking MEASURED; enum / maxOutput
   // / image intake measured the same way. omni-flash: context = doc/web statement (1M class, not
-  // API-measured); partialMode / cacheMode / thinkApi / tempRange = same-family carry-over;
+  // API-measured); partialMode / thinkApi / tempRange = same-family carry-over;
   // reasoningEcho undeclared (cross-turn echo unverified — R-4). The server also takes audio, this
   // repo's call path sends images only ⇒ multimodal below = image intake (no `modalities` field, §2.7).
-  ["qwen3.8-omni-flash",{ context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-omni-flash",{ context: 1_000_000, maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
   // 27b: context 262_144 = doc/web statement, the native value (1M is YaRN extrapolation, local
-  // deploy only); partialMode / cacheMode / thinkApi / tempRange same-family carry-over; reasoningEcho
+  // deploy only); partialMode / thinkApi / tempRange same-family carry-over; reasoningEcho
   // undeclared (cross-turn echo unverified — R-4).
-  ["qwen3.8-27b",       { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
+  ["qwen3.8-27b",       { context: 262_144,   maxOutput: 131_072, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], tempRange: [0, 2] }],
   // No generic "qwen" fallback row: unlisted qwen names deliberately resolve to DEFAULT_SPEC
   // (128K / 32K, no vision) — per-name disposition in docs/core/design/MODEL-SPECS.md §2.4.
   // Do not re-add a blanket prefix row (silent inheritance is how missing rows stayed invisible).
@@ -100,41 +132,54 @@ const MODEL_SPECS = [
   // ≠ 3.8 系七档）；maxOutput 65_536 = **校验级**（400「Range of max_tokens should be [1, 65536]」；
   // 65536 → 200）。thinking / multimodal = **实测**（裸请求 rc=143–239 默认开、none/minimal rc=0 可关；
   // 64×64 纯红 PNG 五名受理）。context = **官方口径**（未探边，文档口径非 API 实测）；partialMode /
-  // cacheMode / thinkApi / tempRange = **族沿用**（qwen3.7/3.8 行先例）；reasoningEcho 不声明
+  // thinkApi / tempRange = **族沿用**（qwen3.7/3.8 行先例）；reasoningEcho 不声明
   // （跨轮回声未验——R-4）。日期戳快照名不登——前缀命中母名行（有意继承）。
-  ["qwen3.6-flash",     { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-flash",     { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-plus：同形行（共享证据块见 flash 行上）——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位。
-  ["qwen3.6-plus",      { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-plus",      { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-max-preview：同形行——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位；视觉通道实测 ✓ 但纯红图答「黑色」（抽样波动，如实记）。
-  ["qwen3.6-max-preview", { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-max-preview", { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // qwen3.6-27b：同形行——校验级 maxOutput / 六档枚举、官方口径 context（未探边）、族沿用机制位。
-  ["qwen3.6-27b",        { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-27b",        { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  // qwen3.5-27b（2026-09-24 bench 名单批）：context 262_144 = **官方口径（定价页分档佐证——[128K, 256K) 档在售；规格页未逐名核）**；
+  // maxOutput / 机制位 = **族沿用 qwen3.6-27b 行**；枚举 / 视觉面 = **未探**（不声明）。
+  ["qwen3.5-27b",        { context: 262_144, maxOutput: 65_536, thinking: true, partialMode: true,  thinkApi: "effort", tempRange: [0, 2] }],
   // qwen3.6-35b-a3b：MoE 后缀（a3b = 激活参数量级）——校验级 maxOutput / 六档枚举、官方口径 context、族沿用机制位。
-  ["qwen3.6-35b-a3b",    { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true, cacheMode: "none", thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
+  ["qwen3.6-35b-a3b",    { context: 1_000_000, maxOutput: 65_536, thinking: true, partialMode: true, multimodal: true,  thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh"], tempRange: [0, 2] }],
   // MiniMax series
-  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
+  // effort（`reasoning_effort`）在 MiniMax 族 = **无生效面**（**实测**：effort 被忽略——`none` / `medium` 均 rc 在场；
+  // off 路径 = `thinking:{type:"disabled"}` ✓ rc=0）——medium 照发仅口径统一（batch-params-judge §1.2 / KD-32④）。
+  ["MiniMax-M3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true,  thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
   // MiMo series (Xiaomi — OpenAI-compatible https://api.xiaomimimo.com/v1; deep thinking via
   // thinking.type, default ON). Family echo policy stays conservative ("required" — tool rounds
   // always echo); 2026-09-22 re-probe: value / missing field / empty string all 200 — the
   // 2026-09-20 "must be passed back" 400 was NOT reproduced. v2.5 rows aligned 2026-09-22:
-  // maxOutput 131_072 = **校验级**; cacheMode "auto" = **实测** (2nd same-prefix round cached 18,816).
-  ["mimo-v2.5-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
-  ["mimo-v2.5",         { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // maxOutput 131_072 = **校验级**.
+  ["mimo-v2.5-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  ["mimo-v2.5",         { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // MiMo V2.6 series (2026-09-22 launch — three independent rows, each probed individually).
   // maxOutput 131_072 / tempRange [0, 1.5] = **校验级** (400 "at most 131072 completion tokens" /
   // "temperature must be within [0, 1.5]", per model); thinking = **实测** (bare request carries
   // reasoning_content; thinking.type disabled → rc gone ⇒ thinkApi "type" = measured face) and
-  // multimodal = **实测** (8×8 pure-red PNG, pro / flash answered "Red"); cacheMode "auto" = **实测**
-  // (cached_tokens 18,688). context 1_000_000 = **官方口径** (docs); reasoningEcho = **族沿用**.
-  ["mimo-v2.6-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // multimodal = **实测** (8×8 pure-red PNG, pro / flash answered "Red"); context 1_000_000 = **官方口径**
+  // (docs); reasoningEcho = **族沿用**.
+  // effort（`reasoning_effort`）在三款 V2.6 = **无生效面**（**实测**：effort 被忽略——`none` / `medium` 均 rc 在场；
+  // off 路径 = `thinking:{type:"disabled"}` ✓ rc=0）——medium 照发仅口径统一（batch-params-judge §1.2 / KD-32④）。
+  ["mimo-v2.6-pro",     { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // mimo-v2.6-flash: same row shape as pro — maxOutput 131_072 / tempRange [0, 1.5] = **校验级**,
-  // thinking / multimodal / cacheMode = **实测**, context = **官方口径**, reasoningEcho = **族沿用**.
-  ["mimo-v2.6-flash",   { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // thinking / multimodal = **实测**, context = **官方口径**, reasoningEcho = **族沿用**.
+  ["mimo-v2.6-flash",   { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
   // mimo-v2.6-pro-ultraspeed: same row shape — same grades as flash (**校验级** / **实测** /
   // **官方口径** / **族沿用**); image answer at a 64-token budget was truncated — "Red" at 512.
-  ["mimo-v2.6-pro-ultraspeed", { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
-  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true, cacheMode: "auto", thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
-  ["minimax-m1",        { context: 256_000,   maxOutput: 128_000, thinking: false, cacheMode: "auto", noUsageStream: true }],
+  ["mimo-v2.6-pro-ultraspeed", { context: 1_000_000, maxOutput: 131_072, thinking: true,  multimodal: true,  thinkApi: "type", reasoningEcho: "required", tempRange: [0, 1.5] }],
+  // minimax-m3（小写前缀行——与上方 MiniMax-M3 行同值，同族两写法）：effort 同按**无生效面**处置
+  // （**实测**：effort 被忽略；off 路径 = `thinking:{type:"disabled"}` ✓ rc=0——batch-params-judge §1.2 / KD-32④）。
+  ["minimax-m3",        { context: 1_000_000, maxOutput: 128_000, thinking: true,  multimodal: true,  thinkApi: "type", thinkEnabledValue: "adaptive", tempRange: [0, 2], noUsageStream: true }],
+  ["minimax-m1",        { context: 256_000,   maxOutput: 128_000, thinking: false, noUsageStream: true }],
+  // MiniMax-M2.7（前缀行——覆盖标准档与 -highspeed · 2026-09-24 bench 名单批）：thinking = **实测**（裸请求 rc 在场）；off 路径 **未找到**
+  // （`thinking:{type:"disabled"}` 不生效——如实，不声明 thinkApi）；effort 被忽略 ⇒ **无生效面**（枚举不声明；medium 照发仅口径统一）；
+  // 尺寸 / noUsageStream = **族沿用 minimax-m1 行**；视觉面 = **未探**（不声明）。
+  ["MiniMax-M2.7", { context: 256_000, maxOutput: 128_000, thinking: true, noUsageStream: true }],
   // Grok series (xAI — OpenAI-compatible)
   // grok-4.x: 500K context per xAI Grok 4.6 spec (corrected 2026-08; earlier entries said 1M)
   ["grok-4.6",          { context: 500_000,   maxOutput: 64_000,  thinking: false, multimodal: true, tempRange: [0, 2] }],
@@ -145,15 +190,21 @@ const MODEL_SPECS = [
   ["mistral-large",     { context: 128_000,   maxOutput: 32_000,  thinking: false, multimodal: true, tempRange: [0, 2] }],
   ["codestral",         { context: 256_000,   maxOutput: 32_000,  thinking: false, tempRange: [0, 2] }],
   // Claude series (Anthropic)
-  ["claude-opus-5",     { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-sonnet-5",   { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-opus-4",     { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-sonnet-4",   { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, cacheMode: "none", format: "anthropic" }],
-  ["claude-3.5-haiku",  { context: 200_000,   maxOutput: 8_192,   thinking: false, cacheMode: "none", format: "anthropic" }],
+  ["claude-opus-5",     { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-sonnet-5",   { context: 1_000_000, maxOutput: 128_000, thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-opus-4",     { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-sonnet-4",   { context: 200_000,   maxOutput: 32_000,  thinking: false, multimodal: true, format: "anthropic" }],
+  ["claude-3.5-haiku",  { context: 200_000,   maxOutput: 8_192,   thinking: false, format: "anthropic" }],
+  // claude-fable-5.1（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 1_000_000 / maxOutput 128_000
+  // （多源网络转述级，本机无该渠道——行值 + 行注即标级）；thinking「always-on adaptive」= 弱源读数 ⇒ 零口径不落字段（§14.3）。
+  ["claude-fable-5.1",  { context: 1_000_000, maxOutput: 128_000 }],
   // Gemini series (Google)
-  ["gemini-3-pro",      { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
-  ["gemini-2.5-pro",    { context: 2_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
-  ["gemini-2.5-flash",  { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, cacheMode: "none", format: "google", noUsageStream: true }],
+  ["gemini-3-pro",      { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  ["gemini-2.5-pro",    { context: 2_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  ["gemini-2.5-flash",  { context: 1_000_000, maxOutput: 64_000,  thinking: false, multimodal: true, format: "google", noUsageStream: true }],
+  // gemini-3.1-pro（2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 1_048_576 / maxOutput 65_536（多源
+  // 网络转述级，本机无该渠道——行值 + 行注即标级）；thinking「有·计入输出预算」= 弱源读数 ⇒ 零口径不落字段（§14.3）。
+  ["gemini-3.1-pro",    { context: 1_048_576, maxOutput: 65_536 }],
   // Tencent Hunyuan (TokenHub — 聚合网关 `https://tokenhub.tencentmaas.com/v1`；批 2026-09-20-channel-onboarding)
   // hy3：thinking / thinkApi / effort 枚举 = **本渠道实测**——裸请求 reasoning_content 在场（tok=16），
   // `reasoning_effort:"none"` 即消失（实测唯一有效 off 路径）。枚举 = 七值**受理级探针**（七值全 200 受理；
@@ -163,7 +214,7 @@ const MODEL_SPECS = [
   // **参考实配**（网关对 max_tokens 不硬拒 ⇒ 上限未证）；context 256_000 = **网络口径**（他仓/文档 256K，未本渠道实测）。
   ["hy3",               { context: 256_000, maxOutput: 128_000, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
   // hy3-preview（在场实测：/models + 401 前验活；能力位**未探针**）——**尺寸行**：仅登记尺寸位，取值 =
-  // hy3 **同族沿用**（maxOutput 参考实配 / context 网络口径）。D-11：能力位不跨名沿用 ⇒ thinking / 枚举 / 视觉全不声明。
+  // hy3 **同族沿用**（maxOutput 参考实配 / context 网络口径）。D-11（2026-09-25 收窄现口径）：能力位**无族据**一律不声明 ⇒ 该名无族据 ⇒ thinking / 枚举 / 视觉全不声明。
   ["hy3-preview",       { context: 256_000, maxOutput: 128_000 }],
   // hy4-preview（在场实测：/models + 401；能力位**未探针**）——尺寸行：取值 = hy3 族**同族沿用**
   // （maxOutput 参考实配级取值 / context 网络口径；批次档 §1.2 未给该名尺寸行 ⇒ 上报清单 2，行注即标级）。
@@ -177,8 +228,19 @@ const MODEL_SPECS = [
   // seed-lite = 同族全针（批次档 §1.2 lite 行：思考 tok=161 / 视觉 "Red" / 上限 131 072 / echo 真名）——
   // context 同按 **官方口径** + **实测**下界标注；行独立（同值集 ≠ 同行：改一行不动另一行）。
   ["doubao-seed-2-0-lite-260428", { context: 256_000, maxOutput: 131_072, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"], multimodal: true }],
+  // doubao-seed-2-1 三档（火山方舟 · 2026-09-24 bench 名单批 · 三行同形）：thinking / thinkApi / 枚举 = **实测**
+  // （裸请求 rc 在场；`reasoning_effort:"none"` → rc=0；七值全 200、乱值 400 泛化 ⇒ **受理级**——等级低于 2.0 行的
+  // 「真校验」，同值集不混级）；maxOutput 524_288 = **受理级**（262 144 / 524 288 均受理；**上限未证**）；
+  // context = **族沿用 2.0 行**（官方口径未逐名取证）；**视觉面未探**（不声明——如实标级，不冒充实测）。
+  ["doubao-seed-2-1-pro-260915",   { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  ["doubao-seed-2-1-turbo-260628", { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  ["doubao-seed-2-1-lite-260915",  { context: 256_000, maxOutput: 524_288, thinking: true, thinkApi: "effort", reasoningEffortEnum: ["none", "minimal", "low", "medium", "high", "xhigh", "max"] }],
+  // step-3.7-flash（StepFun · 2026-09-25 批 §14.2 #2-6 · §1.7.1 父侧交付）：**尺寸行**——context 262_144（多源 256K =
+  // 网络转述级；渠道未激活——400 "product is not activated"）；maxOutput 未取到 ⇒ 零口径——取兜底值 32_000（与建行前
+  // 生效值同值）；行键 = 裸名（渠道名录形 stepfun/step-3.7-flash——命名空间剥离同命中）。
+  ["step-3.7-flash",    { context: 262_144, maxOutput: 32_000 }],
 ]
-const DEFAULT_SPEC = { context: 128_000, maxOutput: 32_000, cacheMode: "none" }
+const DEFAULT_SPEC = { context: 128_000, maxOutput: 32_000 }
 
 /** Look up spec by model name prefix (case-insensitive), conservative default for unknown models.
  *
@@ -194,9 +256,11 @@ const SORTED_SPECS = [...MODEL_SPECS].sort((a, b) => b[0].length - a[0].length)
 
 /** Single table lookup shared by specForModel / specMatch — prefix match (case-insensitive)
  *  with vendor-namespace stripping. Returns null on a miss (DEFAULT_SPEC is the caller-side
- *  fallback and is deliberately NOT returned here — `matched` needs the miss itself). */
+ *  fallback and is deliberately NOT returned here — `matched` needs the miss itself).
+ *  非串 model（外端脏载 / 手工档）归一等价空串 ⇒ 未命中 ⇒ 调用侧 DEFAULT_SPEC（#638 查表全性：
+ *  下游无类型门——`.toLowerCase` 只认串，非串归一后本函数恒 total，同 providerSpec 文档句）。 */
 function lookupSpec(model) {
-  const m = (model ?? "").toLowerCase()
+  const m = (typeof model === "string" ? model : "").toLowerCase()
   for (const [prefix, spec] of SORTED_SPECS) {
     if (m.startsWith(prefix.toLowerCase())) return spec
   }
@@ -215,7 +279,8 @@ function lookupSpec(model) {
  *  instead of silently degrading to the 128K default (IK5VGJ). The dedupe set is shared by
  *  specForModel / specMatch (PROVIDER.md §16 M5). */
 function warnUnknownModel(model) {
-  const m = (model ?? "").toLowerCase()
+  // 非串 model（脏载）归一空串 ⇒ 不警告（未登记 ≠ 拼错——不刷屏；#638 与 lookupSpec 同归一）
+  const m = (typeof model === "string" ? model : "").toLowerCase()
   if (m && !warnedModels.has(m)) {
     warnedModels.add(m)
     console.warn(`[config] model "${model}" not found in MODEL_SPECS — using default spec (128K context, 32K output). Check the model ID or add an alias.`)

package/notify-policy.mjs

@@ -0,0 +1,51 @@
+/**
+ * notify-policy.mjs — 完成提示面**策略**（失焦门 + 单档 + 载荷成形 + 词键）——「桌面处理流 · VSC 对齐」批 R6 上提产物。
+ *
+ * 上提源 = 桌面现状超集 `thincoder-desktop/src/main/notify.mjs`（零宿主依赖 —— **纯搬，零语义改**）；
+ * 对位锚（迁移前坐标）= VSC `thincoder-vscode/src/extension/notify.mjs:9-18`（失焦门同判据）+
+ * `panel-callbacks.mjs:236`（`!autoTurn` 门径）。平台落子（Electron `Notification` vs
+ * VSC `showInformationMessage`）= 宿主能力面（真端差登记 ①）⇒ **装配留端**（桌面 =
+ * `thincoder-desktop/src/main/main.mjs`；VSC = `notify.mjs` 宿主包装——已随 parity-b4 迁移落形）；本档只承策略面。
+ *
+ * `notify`（平台落子）· `focused`（焦态判据）· `reveal`（点击面）· `localeOf`（语言取值缝——VSC
+ * wrapper 携 `vscode.env.language`，`normalizeLocale` 归一 `zh-CN → zh`；缺省 = `agent.config?.locale`，
+ * 桌面径零改）四件注入 ⇒ 平 node 直测。词键一键自此持本档（值 = 原 VSC 逐字 `notify.done`——该键已随
+ * parity-b4 迁移自 `thincoder-vscode/locales/{zh,en}.json:222` 删）。触发单档（`!autoTurn` 判据在调用面 —— 单回合
+ * 执行面；VSC 同判据）：
+ *   用户回合完成（VSC 逐字同判据 = `panel-callbacks.mjs:236` 的 `!autoTurn` 门）
+ * `title` = 会话标题（`agent.title`——槽装载面在场；空 ∕ 不可得 ⇒ **零携**）；点击 = 聚焦窗口 ∕
+ * 打开 thincoder 视图（reveal 语义 —— 宿主面；不切会话）。
+ * 消费面 = 桌面 `thincoder-desktop/src/main/notify.mjs`（端胶水 re-export）；VSC `notify.mjs`（宿主包装——已随 parity-b4 迁移落形）。
+ */
+import { normalizeLocale } from "./i18n.mjs"
+
+/** 词键值（zh ∕ en 常量对：`notify.done` = 原 VSC `locales/{zh,en}.json:222` 逐字——该键已随 parity-b4 迁移自 locales 删）。 */
+export const NOTIFY_TEXTS = Object.freeze({
+  zh: "ThinCoder：本轮完成",
+  en: "ThinCoder: turn complete",
+})
+
+/** 零动作面（`notify` 注入缺位 = 提示面缺位 ⇒ 零连带 —— 回合链不受累、不抛；沿「禁假造」口径不落假通知）。 */
+const SILENT = Object.freeze({ turnDone: () => false })
+
+/** 提示面工厂：`notify(payload)` = 平台落子（载荷 `{ title?, body, reveal? }`）· `focused()` = 窗口聚焦判据
+ *  （真 ⇒ 门闭合 · 缺省 = 恒聚焦 ⇒ 零通知）· `reveal()` = 点击面（宿主语义）· `localeOf(agent)` =
+ *  语言取值缝（缺省 = `agent?.config?.locale`）。返回单档 `turnDone`（返回值 = 是否落子，测试读数面）。 */
+export function createNotifier({ notify = null, focused = null, reveal = null, localeOf = null } = {}) {
+  if (typeof notify !== "function") return SILENT
+  const isFocused = typeof focused === "function" ? focused : () => true
+  const langOf = (agent) => (normalizeLocale(typeof localeOf === "function" ? localeOf(agent) : agent?.config?.locale) === "zh" ? "zh" : "en")
+  const fire = ({ agent = null } = {}) => {
+    if (isFocused()) return false // 失焦门：聚焦 = no-op（never noise —— VSC `thincoder-vscode/src/extension/notify.mjs:9-18` 同判据）
+    const payload = { body: NOTIFY_TEXTS[langOf(agent)] }
+    const title = typeof agent?.title === "string" && agent.title.trim() !== "" ? agent.title : null
+    if (title !== null) payload.title = title // 会话标题在场才携（不可得 ⇒ 零携）
+    if (typeof reveal === "function") payload.reveal = reveal
+    try { notify(payload) } catch (error) { console.error("[notify] dispatch failed:", error) } // 平台抛 ⇒ 零连带
+    return true
+  }
+  return {
+    /** 用户回合完成（`!autoTurn` 判据在调用面 —— 单回合执行面成功径）。 */
+    turnDone(vars) { return fire(vars) },
+  }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@thincoder/core",
-  "version": "0.9.4",
+  "version": "0.10.1",
   "description": "ThinCoder shared core — shared mechanism modules + shared prompt content (CLI ↔ VS Code extension).",
   "keywords": [
     "ai",