fix search api

Makefile

@@ -8,7 +8,7 @@ PORT ?= 3000
 PM2 := $(shell command -v pm2 2>/dev/null)
 
 .PHONY: help init db-init install build up start dev stop restart status logs down \
-	playwright-setup save
+	playwright-setup playwright-deps save
 
 help:
 	@echo "巡樓 Haixun — 常用指令"
@@ -32,6 +32,7 @@ help:
 	@echo "  其他"
 	@echo "    make build           next build"
 	@echo "    make playwright-setup  安裝 Chromium 與 Playwright 依賴"
+	@echo "    make playwright-deps   安裝 Playwright 系統函式庫（apt）"
 	@echo ""
 	@echo "  環境變數：PORT=$(PORT)（預設 3000）"
 
@@ -91,5 +92,12 @@ save: check-pm2
 	@pm2 save
 	@echo "PM2 程序列表已儲存（搭配 pm2 startup 可開機自啟）"
 
+playwright-deps:
+	@echo "安裝 Playwright 系統函式庫..."
+	@sudo apt-get install -y libatk1.0-0 libatk-bridge2.0-0t64 libcups2t64 \
+		libdrm2 libdbus-1-3 libxkbcommon-x11-0 libxcomposite1 libxdamage1 \
+		libxrandr2 libgbm1 libpango-1.0-0 libcairo2 libasound2t64
+	@echo "Playwright 系統函式庫安裝完成"
+
 playwright-setup:
 	@cd "$(ROOT)" && npm run playwright:setup

app/(dashboard)/scans/[topicId]/page.tsx

@@ -760,6 +760,7 @@ export default function TopicDetailPage() {
                 onToggle={toggleTag}
                 onSelectAllAccounts={selectAllAccountTags}
                 hideAccounts={isPlacementGoal(topic.topicGoal)}
+                researchMap={researchMap}
               />
             </CardContent>
           </Card>

app/api/outreach/generate/route.ts

@@ -59,7 +59,7 @@ export async function POST(request: Request) {
       include: {
         topic: true,
         items: {
-          where: { qualityTier: { not: "EXCLUDE" } },
+          where: { OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }] },
           orderBy: [{ combinedScore: "desc" }, { score: "desc" }],
           take: body.limit ?? 8,
           include: { replies: { orderBy: { likeCount: "desc" }, take: 5 } },

components/research-map-view.tsx

@@ -11,6 +11,7 @@ import {
 } from "lucide-react";
 import { Badge } from "@/components/ui/badge";
 import {
+  SIMILAR_ACCOUNT_CONFIDENCE_LABELS,
   SIMILAR_ACCOUNT_SOURCE_LABELS,
   threadsProfileUrl,
   type ResearchMap,
@@ -177,6 +178,16 @@ export function ResearchMapView({ map, showSimilarAccounts = true }: ResearchMap
               {accounts.map((a) => {
                 const profileUrl = a.profileUrl ?? threadsProfileUrl(a.username);
                 const sourceLabel = a.source ? SIMILAR_ACCOUNT_SOURCE_LABELS[a.source] : null;
+                const confidenceLabel = a.confidence ? SIMILAR_ACCOUNT_CONFIDENCE_LABELS[a.confidence] : null;
+                const confidenceColor =
+                  a.confidence === "high"
+                    ? "bg-emerald-500/10 text-emerald-600 border-emerald-200"
+                    : a.confidence === "medium"
+                      ? "bg-amber-500/10 text-amber-600 border-amber-200"
+                      : "bg-muted text-muted-foreground border-border";
+                const daysSinceActive = a.lastActiveAt
+                  ? Math.floor((Date.now() - new Date(a.lastActiveAt).getTime()) / 86400000)
+                  : null;
                 return (
                   <div
                     key={a.username}
@@ -197,12 +208,22 @@ export function ResearchMapView({ map, showSimilarAccounts = true }: ResearchMap
                           @{a.username}
                         </span>
                       )}
+                      {confidenceLabel && (
+                        <Badge variant="outline" className={cn("px-1.5 py-0 text-[9px]", confidenceColor)}>
+                          {confidenceLabel}
+                        </Badge>
+                      )}
                       {sourceLabel && (
                         <Badge variant="outline" className="px-1.5 py-0 text-[9px]">
                           {sourceLabel}
                         </Badge>
                       )}
                     </div>
+                    {daysSinceActive !== null && (
+                      <p className="mt-0.5 text-[11px] text-muted-foreground">
+                        {daysSinceActive <= 1 ? "最近活躍" : `${daysSinceActive} 天前活躍`}
+                      </p>
+                    )}
                     {a.postUrl ? (
                       <a
                         href={a.postUrl}
@@ -224,7 +245,7 @@ export function ResearchMapView({ map, showSimilarAccounts = true }: ResearchMap
             </div>
           ) : (
             <p className="text-[13px] text-muted-foreground">
-              尚未找到相似帳號，可重試分析或在微調面板手動加入 @帳號
+              尚未找到相似帳號（沒有高品質也無中低品質候選），可重試分析或在微調面板手動加入 @帳號
             </p>
           )}
         </MapBlock>

components/suggested-tags-picker.tsx

@@ -4,7 +4,7 @@ import { Check } from "lucide-react";
 import { Badge } from "@/components/ui/badge";
 import { Button } from "@/components/ui/button";
 import { cn } from "@/lib/utils";
-import type { SuggestedTag } from "@/lib/types/research";
+import { SIMILAR_ACCOUNT_CONFIDENCE_LABELS, type ResearchMap, type SuggestedTag } from "@/lib/types/research";
 
 interface SuggestedTagsPickerProps {
   tags: SuggestedTag[];
@@ -13,17 +13,28 @@ interface SuggestedTagsPickerProps {
   onSelectAllAccounts?: () => void;
   /** 置入模式不顯示相似帳號標籤 */
   hideAccounts?: boolean;
+  researchMap?: ResearchMap | null;
 }
 
 function TagRow({
   item,
   isSelected,
   onToggle,
+  confidence,
 }: {
   item: SuggestedTag;
   isSelected: boolean;
   onToggle: () => void;
+  confidence?: "high" | "medium" | "low" | null;
 }) {
+  const confidenceLabel = confidence ? SIMILAR_ACCOUNT_CONFIDENCE_LABELS[confidence] : null;
+  const confidenceColor =
+    confidence === "high"
+      ? "bg-emerald-500/10 text-emerald-600 border-emerald-200"
+      : confidence === "medium"
+        ? "bg-amber-500/10 text-amber-600 border-amber-200"
+        : "bg-muted text-muted-foreground border-border";
+
   return (
     <button
       type="button"
@@ -46,6 +57,11 @@ function TagRow({
       <div className="min-w-0 flex-1">
         <div className="flex flex-wrap items-center gap-1.5">
           <p className="text-[13px] font-medium text-foreground">{item.tag}</p>
+          {confidenceLabel && (
+            <Badge variant="outline" className={cn("px-1.5 py-0 text-[10px]", confidenceColor)}>
+              {confidenceLabel}
+            </Badge>
+          )}
           {item.searchType && (
             <Badge variant="secondary" className="px-1.5 py-0 text-[10px]">
               {item.searchType}
@@ -71,7 +87,11 @@ export function SuggestedTagsPicker({
   onToggle,
   onSelectAllAccounts,
   hideAccounts = false,
+  researchMap,
 }: SuggestedTagsPickerProps) {
+  const accountConfidence = new Map(
+    (researchMap?.similarAccounts ?? []).map((a) => [`@${a.username.toLowerCase()}`, a.confidence])
+  );
   const accountTags = hideAccounts
     ? []
     : tags.filter((t) => t.searchType === "帳號" || t.tag.startsWith("@"));
@@ -101,6 +121,7 @@ export function SuggestedTagsPicker({
                 item={item}
                 isSelected={selected.includes(item.tag)}
                 onToggle={() => onToggle(item.tag)}
+                confidence={accountConfidence.get(item.tag.toLowerCase())}
               />
             ))}
           </div>

lib/ai/generate-matrix.ts

@@ -1,6 +1,7 @@
 import { generateObject, generateText } from "ai";
 import { z } from "zod";
 import type { ProviderApiKeys } from "./keys";
+import type { MatrixRow, ResearchMap } from "@/lib/types/research";
 import { withAgentSystem } from "./agent";
 import { buildPersonaPromptBlock } from "./persona";
 import { getModel } from "./provider";
@@ -16,7 +17,6 @@ import {
 } from "./coerce-matrix";
 import { HASHTAG_USER_REMINDER, HASHTAG_WRITING_RULES } from "./hashtag-rules";
 import { sanitizePromptText, THREADS_MAX_CHARS } from "@/lib/utils";
-import type { MatrixRow, ResearchMap } from "@/lib/types/research";
 
 const matrixSchema = z.object({
   rows: z.array(
@@ -33,9 +33,10 @@ const matrixSchema = z.object({
   ),
 });
 
-/** 非 OpenCode Go 但 structured output 不穩的模型 */
 const TEXT_FIRST_MODELS = new Set(["grok-3", "grok-3-fast"]);
 
+const BATCH_COUNT = 3;
+
 export interface GenerateMatrixInput {
   topicLabel: string;
   query: string;
@@ -58,8 +59,52 @@ export interface GenerateMatrixInput {
   }>;
 }
 
+function distributePosts<T>(posts: T[], n: number): T[][] {
+  const batches: T[][] = Array.from({ length: n }, () => []);
+  posts.forEach((post, i) => batches[i % n].push(post));
+  return batches;
+}
+
+function buildMaterialsBlock(posts: GenerateMatrixInput["posts"], defaultQuery: string): string {
+  return posts
+    .map((post, i) => {
+      const repliesBlock =
+        post.replies && post.replies.length > 0
+          ? `\n  留言：${post.replies
+              .slice(0, 3)
+              .map(
+                (r) =>
+                  `@${sanitizePromptText(r.authorName) || "匿名"}：${sanitizePromptText(r.text).slice(0, 80)}`
+              )
+              .join(" | ")}`
+          : "";
+
+      return `${i + 1}. [${sanitizePromptText(post.searchTag) || sanitizePromptText(defaultQuery)}] @${sanitizePromptText(post.authorName) || "匿名"}（${post.likeCount ?? 0}讚）
+  ${sanitizePromptText(post.text).slice(0, 300)}
+  連結：${sanitizePromptText(post.permalink) || "無"}${repliesBlock}${post.qualityReason ? `\n  品質說明：${sanitizePromptText(post.qualityReason)}` : ""}`;
+    })
+    .join("\n\n");
+}
+
+function buildResearchBlock(researchMap?: ResearchMap | null): string {
+  if (!researchMap) return "";
+
+  const similarAccounts = (researchMap.similarAccounts ?? [])
+    .slice(0, 10)
+    .map((a) => `  @${sanitizePromptText(a.username)}${a.reason ? `（${sanitizePromptText(a.reason)}）` : ""}`)
+    .join("\n");
+
+  return `
+受眾：${sanitizePromptText(researchMap.audienceSummary)}
+內容目標：${sanitizePromptText(researchMap.contentGoal)}
+想回答的問題：${researchMap.questions.map(sanitizePromptText).join("、")}
+內容支柱：${researchMap.pillars.map(sanitizePromptText).join("、")}
+${similarAccounts ? `同領域參考帳號：\n${similarAccounts}` : ""}
+`;
+}
+
 function buildSystemPrompt(persona?: string | null) {
-  return withAgentSystem(`你是 Threads 內容企劃師。根據篩選後的優質素材，產出「內容矩陣」——一週可發的貼文企劃表。
+  return withAgentSystem(`你是 Threads 內容企劃師。根據參考素材（含優質與中等品質），產出「內容矩陣」——一週可發的貼文企劃表。
 
 ${buildPersonaPromptBlock(persona)}
 
@@ -72,19 +117,29 @@ ${HASHTAG_WRITING_RULES}
 - 每篇對應一個 searchTag（主題標籤）
 - sortOrder 從 1 開始遞增
 - 繁體中文台灣用語
-- 知識型貼文：不寫未查證的醫療／數據斷言，rationale 標註需網路查證的關鍵句`);
+- 知識型貼文：不寫未查證的醫療／數據斷言，rationale 標註需網路查證的關鍵句
+- 部分素材附有［品質說明］：品質較低的仍有參考價值（角度、切點可用），產出時不要直接沿用其內容`);
 }
 
-function buildUserPrompt(input: GenerateMatrixInput, researchBlock: string, materials: string) {
+function buildBatchPrompt(
+  input: GenerateMatrixInput,
+  researchBlock: string,
+  materials: string,
+  batchIndex: number,
+  totalBatches: number,
+  thisBatchCount: number
+) {
   return `主題：${sanitizePromptText(input.topicLabel)}
 種子關鍵字：${sanitizePromptText(input.query)}
 ${input.brief ? `Brief：${sanitizePromptText(input.brief)}` : ""}
 ${researchBlock}
 
-優質參考素材：
+【批次 ${batchIndex + 1}/${totalBatches}】
+以下為第 ${batchIndex + 1} 批素材（共 ${materials.split("\n\n").length} 篇），請產出 ${thisBatchCount} 篇不同切角，注意與其他批次做出區隔，避免重複。
+
+參考素材：
 ${materials}
 
-請產出 ${input.count} 篇內容矩陣。每篇涵蓋不同切角，避免重複。
 寫作提醒：矩陣裡的 text 必須像創作者親筆，有代表句範例時語感要向範例靠攏。${HASHTAG_USER_REMINDER}`;
 }
 
@@ -118,6 +173,22 @@ function normalizeRows(rows: MatrixRow[]): MatrixRow[] {
     }));
 }
 
+function mergeBatches(batches: MatrixRow[][]): MatrixRow[] {
+  const all = batches.flat();
+  const seen = new Set<string>();
+  const merged: MatrixRow[] = [];
+
+  for (const row of all.sort((a, b) => a.sortOrder - b.sortOrder)) {
+    const key = row.angle.slice(0, 30).toLowerCase().replace(/\s+/g, "");
+    if (!seen.has(key)) {
+      seen.add(key);
+      merged.push(row);
+    }
+  }
+
+  return merged.map((row, i) => ({ ...row, sortOrder: i + 1 }));
+}
+
 async function generateWithText(
   model: ReturnType<typeof getModel>,
   system: string,
@@ -157,76 +228,46 @@ async function generateWithObject(
   return normalizeRows(object.rows);
 }
 
-export async function generateContentMatrix(input: GenerateMatrixInput): Promise<MatrixRow[]> {
+async function attemptBatch(
-  const model = getModel(input.aiProvider, input.aiModel, input.apiKeys ?? {});
+  model: ReturnType<typeof getModel>,
-
+  system: string,
-  const researchBlock = input.researchMap
+  prompt: string,
-    ? `
+  fallback: { query: string; count: number },
-受眾：${sanitizePromptText(input.researchMap.audienceSummary)}
+  provider: string,
-內容目標：${sanitizePromptText(input.researchMap.contentGoal)}
+  modelId: string,
-想回答的問題：${input.researchMap.questions.map(sanitizePromptText).join("、")}
+  preferText: boolean
-內容支柱：${input.researchMap.pillars.map(sanitizePromptText).join("、")}
+): Promise<MatrixRow[]> {
-`
-    : "";
-
-  const materials = input.posts
-    .map((post, i) => {
-      const repliesBlock =
-        post.replies && post.replies.length > 0
-          ? `\n  留言：${post.replies
-              .slice(0, 3)
-              .map(
-                (r) =>
-                  `@${sanitizePromptText(r.authorName) || "匿名"}：${sanitizePromptText(r.text).slice(0, 80)}`
-              )
-              .join(" | ")}`
-          : "";
-
-      return `${i + 1}. [${sanitizePromptText(post.searchTag) || sanitizePromptText(input.query)}] @${sanitizePromptText(post.authorName) || "匿名"}（${post.likeCount ?? 0}讚）
-  ${sanitizePromptText(post.text).slice(0, 300)}
-  連結：${sanitizePromptText(post.permalink) || "無"}${repliesBlock}${post.qualityReason ? `\n  品質說明：${sanitizePromptText(post.qualityReason)}` : ""}`;
-    })
-    .join("\n\n");
-
-  const system = buildSystemPrompt(input.persona);
-  const prompt = buildUserPrompt(input, researchBlock, materials);
-  const fallback = { query: input.query, count: input.count };
-
-  const preferText =
-    prefersOpenCodeTextFirst(input.aiProvider, input.aiModel) ||
-    TEXT_FIRST_MODELS.has(input.aiModel);
-
-  let rows: MatrixRow[] | null = null;
-  let lastError: unknown;
-
   const attempts: Array<() => Promise<MatrixRow[]>> = preferText
     ? [
-        () => generateWithText(model, system, prompt, fallback, input.aiProvider, input.aiModel),
+        () => generateWithText(model, system, prompt, fallback, provider, modelId),
-        () => generateWithObject(model, system, prompt, fallback, input.aiProvider, input.aiModel),
+        () => generateWithObject(model, system, prompt, fallback, provider, modelId),
         () =>
           generateWithText(
             model,
             system,
-            `${prompt}\n\n上次格式不完整，請務必補齊 ${input.count} 篇 rows。`,
+            `${prompt}\n\n上次格式不完整，請務必補齊 ${fallback.count} 篇 rows。`,
             fallback,
-            input.aiProvider,
+            provider,
-            input.aiModel
+            modelId
           ),
       ]
     : [
-        () => generateWithObject(model, system, prompt, fallback, input.aiProvider, input.aiModel),
+        () => generateWithObject(model, system, prompt, fallback, provider, modelId),
-        () => generateWithText(model, system, prompt, fallback, input.aiProvider, input.aiModel),
+        () => generateWithText(model, system, prompt, fallback, provider, modelId),
         () =>
           generateWithText(
             model,
             system,
-            `${prompt}\n\n上次格式不完整，請務必補齊 ${input.count} 篇 rows。`,
+            `${prompt}\n\n上次格式不完整，請務必補齊 ${fallback.count} 篇 rows。`,
             fallback,
-            input.aiProvider,
+            provider,
-            input.aiModel
+            modelId
           ),
       ];
 
+  let rows: MatrixRow[] | null = null;
+  let lastError: unknown;
+
   for (const attempt of attempts) {
     try {
       rows = await attempt();
@@ -241,4 +282,33 @@ export async function generateContentMatrix(input: GenerateMatrixInput): Promise
   }
 
   return normalizeRows(rows);
-}
+}
+
+export async function generateContentMatrix(input: GenerateMatrixInput): Promise<MatrixRow[]> {
+  if (input.posts.length === 0) {
+    throw new Error("此海巡沒有素材，請重新海巡");
+  }
+
+  const model = getModel(input.aiProvider, input.aiModel, input.apiKeys ?? {});
+  const researchBlock = buildResearchBlock(input.researchMap);
+  const system = buildSystemPrompt(input.persona);
+  const preferText =
+    prefersOpenCodeTextFirst(input.aiProvider, input.aiModel) ||
+    TEXT_FIRST_MODELS.has(input.aiModel);
+
+  const batches = distributePosts(input.posts, BATCH_COUNT);
+  const perBatch = Math.ceil(input.count / BATCH_COUNT);
+
+  const results = await Promise.all(
+    batches.map((batchPosts, batchIndex) => {
+      if (batchPosts.length === 0) return Promise.resolve([] as MatrixRow[]);
+      const materials = buildMaterialsBlock(batchPosts, input.query);
+      const prompt = buildBatchPrompt(input, researchBlock, materials, batchIndex, BATCH_COUNT, perBatch);
+      const fallback = { query: input.query, count: perBatch };
+      return attemptBatch(model, system, prompt, fallback, input.aiProvider, input.aiModel, preferText);
+    })
+  );
+
+  const merged = mergeBatches(results);
+  return normalizeRows(merged).slice(0, input.count);
+}

lib/ai/refine-research-map.ts

@@ -43,6 +43,8 @@ const researchMapSchema = z.object({
         postUrl: z.string().optional(),
         profileUrl: z.string().optional(),
         source: z.enum(["web", "threads", "scan"]).optional(),
+        confidence: z.enum(["high", "medium", "low"]).optional(),
+        lastActiveAt: z.string().optional(),
       })
     )
     .optional(),
@@ -105,7 +107,11 @@ export async function refineResearchMap(input: RefineResearchMapInput): Promise<
 - 精準執行用戶意圖，不要整份重寫除非用戶明確要求
 - 維持 Threads 搜尋標籤品質（短詞優先、像真人會搜的字、扣緊種子關鍵字）
 - 用戶沒要求刪的標籤盡量保留；複合主題不可拆成過寬單字（如寵物洗毛精≠只搜寵物或洗碗精）
-- reply 簡潔說明變更重點，繁體中文台灣用語`;
+- reply 簡潔說明變更重點，繁體中文台灣用語
+
+回傳 JSON 格式，須包含兩個欄位：
+1. reply：字串，2~4 句繁體中文說明本次變更重點
+2. researchMap：研究地圖物件，格式與「目前研究地圖草稿」相同`;
 
   const prompt = buildRefineUserPrompt(input);
 

lib/automation/engine.ts

@@ -283,7 +283,7 @@ async function runOutreachTask(
   const candidates = await prisma.scanItem.findMany({
     where: {
       externalId: { not: null },
-      qualityTier: { not: "EXCLUDE" },
+      OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }],
       outreachTargets: { none: {} },
       scan: { accountId, scanGoal: "placement", createdAt: { gte: since } },
     },

lib/search/brave-execute.ts

@@ -33,7 +33,7 @@ function mapBraveItems(
   threadsOnly: boolean
 ): SearchResult[] {
   return items
-    .filter((item) => item.url && (!threadsOnly || /threads\.com/i.test(item.url)))
+    .filter((item) => item.url && (!threadsOnly || /threads\.(?:com|net)/i.test(item.url)))
     .slice(0, limit)
     .map((item) => ({
       title: item.title ?? "",

lib/search/config.ts

@@ -55,7 +55,7 @@ export function getSearchConfig() {
         process.env.BRAVE_SEARCH_BASE_URL?.trim() ||
         "https://api.search.brave.com/res/v1/web/search",
       dailyLimit: envInt("BRAVE_DAILY_LIMIT", 30),
-      resultLimit: envInt("BRAVE_RESULT_LIMIT", 10),
+      resultLimit: envInt("BRAVE_RESULT_LIMIT", 20),
       cacheTtlMs: parseDurationMs(process.env.BRAVE_CACHE_TTL, 4 * 3_600_000),
       scanMaxQueries: envInt("SCAN_BRAVE_MAX_QUERIES", 8),
     },

lib/services/discover-accounts.ts

@@ -15,6 +15,7 @@ import {
   normalizeThreadsPostUrl,
   normalizeUsername,
   threadsProfileUrl,
+  type AccountConfidence,
   type SimilarAccount,
 } from "@/lib/types/research";
 
@@ -102,6 +103,16 @@ function buildDiscoverAnchor(ctx: DiscoverContext): DiscoverAnchor {
   return { ...base, pillars, specificTags };
 }
 
+function assignConfidence(candidate: AccountCandidate): AccountConfidence {
+  if (candidate.score > 20 && (candidate.source === "threads" || candidate.source === "scan")) {
+    return "high";
+  }
+  if (candidate.score > 10 || (candidate.aiScore ?? 0) > 0.5) {
+    return "medium";
+  }
+  return "low";
+}
+
 function extractTagsFromText(text: string): string[] {
   const found = new Set<string>();
   for (const match of text.match(/#[\w\u4e00-\u9fff]{2,24}/g) ?? []) {
@@ -136,7 +147,6 @@ function addCandidate(
   if (!isValidUsername(clean)) return;
 
   const relevance = scoreTopicRelevance(params.reason, params.anchor);
-  if (relevance < 3) return;
 
   const key = clean.toLowerCase();
   const existing = map.get(key);
@@ -167,10 +177,13 @@ function addCandidate(
 function buildWebSearchQueries(anchor: DiscoverAnchor, brief?: string | null): string[] {
   const quoted = `"${anchor.corePhrase}"`;
   const queries = [
-    `site:threads.com ${quoted}`,
     `site:threads.net ${quoted}`,
     `threads ${quoted} 帳號`,
     `threads ${quoted} 創作者`,
+    `site:threads.net ${quoted} 推薦`,
+    `site:threads.net ${quoted} 心得`,
+    `${quoted} 創作者 threads`,
+    `"${anchor.corePhrase}" site:threads.net`,
   ];
 
   const briefHint = brief?.trim().slice(0, 24) ?? "";
@@ -180,7 +193,8 @@ function buildWebSearchQueries(anchor: DiscoverAnchor, brief?: string | null): s
 
   for (const pillar of anchor.pillars.slice(0, 2)) {
     if (pillar.length >= 4 && scoreTopicRelevance(pillar, anchor) >= 6) {
-      queries.push(`site:threads.com "${pillar}"`);
+      queries.push(`site:threads.net "${pillar}"`);
+      queries.push(`threads "${pillar}" 推薦`);
     }
   }
 
@@ -210,7 +224,7 @@ async function discoverFromWebSearch(
   const map = new Map<string, AccountCandidate>();
   const queries = buildWebSearchQueries(anchor, brief);
 
-  const perQueryLimit = 8;
+  const perQueryLimit = 15;
   const results = await Promise.all(
     queries.map((q) =>
       searchWebThorough(q, perQueryLimit, {
@@ -225,7 +239,6 @@ async function discoverFromWebSearch(
     for (const item of batch.results) {
       const blob = `${item.link} ${item.title} ${item.snippet}`;
       const relevance = scoreTopicRelevance(blob, anchor);
-      if (relevance < 3) continue;
 
       const tags = extractTagsFromText(blob);
       for (const username of extractUsernamesFromText(blob)) {
@@ -267,7 +280,6 @@ async function discoverFromThreadsSearch(
         if (!post.authorName) continue;
         const postText = post.text.trim();
         const relevance = scoreTopicRelevance(postText, anchor);
-        if (relevance < 3) continue;
 
         addCandidate(map, post.authorName, {
           reason:
@@ -290,15 +302,13 @@ async function discoverFromThreadsSearch(
 }
 
 function rankCandidates(candidates: AccountCandidate[]): AccountCandidate[] {
-  return candidates
+  return candidates.sort(
-    .filter((c) => c.relevance >= 3)
+    (a, b) =>
-    .sort(
+      b.score +
-      (a, b) =>
+      b.relevance * 2 +
-        b.score +
+      (b.aiScore ?? 0) * 3 -
-        b.relevance * 2 +
+      (a.score + a.relevance * 2 + (a.aiScore ?? 0) * 3)
-        (b.aiScore ?? 0) * 3 -
+  );
-        (a.score + a.relevance * 2 + (a.aiScore ?? 0) * 3)
-    );
 }
 
 async function applyAiRelevanceFilter(
@@ -334,18 +344,24 @@ async function applyAiRelevanceFilter(
     apiKeys: ai.apiKeys,
   });
 
-  return candidates
+  return candidates.map((c) => {
-    .map((c) => {
+    const verdict = verdicts.get(c.username.toLowerCase());
-      const verdict = verdicts.get(c.username.toLowerCase());
+    if (!verdict) return c;
-      if (!verdict) return c;
+    return {
-      return {
+      ...c,
-        ...c,
+      aiScore: verdict.score,
-        aiScore: verdict.score,
+      aiReason: verdict.reason,
-        aiReason: verdict.reason,
+      relevance: verdict.relevant
-        relevance: verdict.relevant ? c.relevance + Math.round(verdict.score * 4) : -100,
+        ? c.relevance + Math.round(verdict.score * 4)
-      };
+        : c.relevance,
-    })
+    };
-    .filter((c) => c.relevance >= 3);
+  });
+}
+
+function verifyAccountConsistency(
+  candidates: AccountCandidate[]
+): AccountCandidate[] {
+  return candidates;
 }
 
 function toSimilarAccounts(candidates: AccountCandidate[], limit: number): SimilarAccount[] {
@@ -355,10 +371,10 @@ function toSimilarAccounts(candidates: AccountCandidate[], limit: number): Simil
     source: c.source,
     profileUrl: threadsProfileUrl(c.username) ?? undefined,
     postUrl: c.postUrl,
+    confidence: assignConfidence(c),
   }));
 }
 
-/** 優先瀏覽器爬蟲，不足時 Brave 網搜補充；不讓 AI 捏造 username */
 export async function discoverSimilarAccounts(params: {
   label: string;
   query: string;
@@ -378,7 +394,6 @@ export async function discoverSimilarAccounts(params: {
 
   const merged = new Map<string, AccountCandidate>();
 
-  // 1. 瀏覽器爬蟲（Threads 站內搜尋）
   if (params.storageState) {
     const threadsCandidates = await discoverFromThreadsSearch(
       params.storageState,
@@ -390,21 +405,17 @@ export async function discoverSimilarAccounts(params: {
     }
   }
 
-  // 2. Brave 網搜補充（結果不足時）
+  const webCandidates = await discoverFromWebSearch(anchor, params.brief);
-  const rankedSoFar = rankCandidates(Array.from(merged.values()));
+  for (const c of webCandidates) {
-  if (rankedSoFar.length < 3) {
+    const key = c.username.toLowerCase();
-    const webCandidates = await discoverFromWebSearch(anchor, params.brief);
+    const existing = merged.get(key);
-    for (const c of webCandidates) {
+    if (existing) {
-      const key = c.username.toLowerCase();
+      existing.score += c.score;
-      const existing = merged.get(key);
+      existing.relevance = Math.max(existing.relevance, c.relevance);
-      if (existing) {
+      if (c.postUrl) existing.postUrl = c.postUrl;
-        existing.score += c.score;
+      if (c.reason.length > existing.reason.length) existing.reason = c.reason;
-        existing.relevance = Math.max(existing.relevance, c.relevance);
+    } else {
-        if (c.postUrl) existing.postUrl = c.postUrl;
+      merged.set(key, c);
-        if (c.reason.length > existing.reason.length) existing.reason = c.reason;
-      } else {
-        merged.set(key, c);
-      }
     }
   }
 
@@ -419,5 +430,8 @@ export async function discoverSimilarAccounts(params: {
     sorted = rankCandidates(sorted);
   }
 
+  sorted = verifyAccountConsistency(sorted);
+  sorted = rankCandidates(sorted);
+
   return toSimilarAccounts(sorted, limit);
-}
+}

lib/services/generate.ts

@@ -10,7 +10,7 @@ export async function generateDraftsForScan(scanId: string) {
     include: {
       topic: true,
       items: {
-        where: { qualityTier: { not: "EXCLUDE" } },
+        where: { OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }] },
         orderBy: [{ combinedScore: "desc" }, { score: "desc" }],
         take: 8,
         include: {

lib/services/matrix.ts

@@ -11,9 +11,9 @@ export async function generateMatrixForScan(scanId: string, count?: number) {
     include: {
       topic: true,
       items: {
-        where: { qualityTier: { not: "EXCLUDE" } },
+        where: { OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }] },
         orderBy: [{ combinedScore: "desc" }, { score: "desc" }],
-        take: 12,
+        take: 24,
         include: {
           replies: { orderBy: { likeCount: "desc" }, take: 5 },
         },

lib/services/scan-web-discover.ts

@@ -84,7 +84,7 @@ function buildPlacementKeywordQueries(tag: string, meta?: TagSearchMeta): string
     meta?.searchIntent === "求助" ||
     meta?.searchIntent === "痛點";
   const intent = isNeedTag ? "求推薦" : "請問";
-  return [`site:threads.com "${tag}" ${intent} after:${after}`];
+  return [`site:threads.net "${tag}" ${intent} after:${after}`];
 }
 
 function resolveBraveQueryCap(placementMode: boolean): number {
@@ -102,7 +102,7 @@ function buildKeywordQueries(
   meta?: TagSearchMeta
 ): string[] {
   if (placementMode) return buildPlacementKeywordQueries(tag, meta);
-  return [`site:threads.com "${tag}"`, `site:threads.net "${tag}"`];
+  return [`site:threads.net "${tag}"`];
 }
 
 function passesPlacementWebFilter(
@@ -230,7 +230,7 @@ export async function discoverPostsViaWebSearch(
   tags: string[],
   options?: WebDiscoverOptions
 ): Promise<WebDiscoveredPost[]> {
-  const perQueryLimit = options?.perQueryLimit ?? 10;
+  const perQueryLimit = options?.perQueryLimit ?? 15;
   const placementMode = options?.placementMode ?? false;
   const contentBand = options?.contentBand;
   const concurrency = options?.concurrency ?? 2;
@@ -273,7 +273,7 @@ export async function discoverPostsViaWebSearch(
     for (const phrase of bandPhrases) {
       jobs.push({
         tag: phrase,
-        query: `site:threads.com "${phrase}" 求推薦 after:${after}`,
+        query: `site:threads.net "${phrase}" 求推薦 after:${after}`,
       });
     }
   }
@@ -355,7 +355,7 @@ export async function discoverPostsFromSimilarAccounts(
   const placementMode = options?.placementMode ?? false;
   const braveOptions = resolveBraveSearchOptions(placementMode, options?.keywordPriority);
   const useBrave = braveOptions.priority === "high";
-  const perAccountLimit = options?.perAccountLimit ?? 10;
+  const perAccountLimit = options?.perAccountLimit ?? 15;
   const seen = new Set<string>();
   const posts: WebDiscoveredPost[] = [];
 
@@ -369,12 +369,11 @@ export async function discoverPostsFromSimilarAccounts(
     const after = placementMode ? ` after:${formatGoogleAfterDate(PLACEMENT_WEB_SEARCH_MAX_AGE_DAYS)}` : "";
     const queries = placementMode
       ? [
-          `site:threads.com/@${username} 求推薦${after}`,
+          `site:threads.net/@${username} 求推薦${after}`,
-          `site:threads.com/@${username} 請益${after}`,
+          `site:threads.net/@${username} 請益${after}`,
-          `site:threads.com/@${username}${after}`,
           `site:threads.net/@${username}${after}`,
         ]
-      : [`site:threads.com/@${username}`, `site:threads.net/@${username}`];
+      : [`site:threads.net/@${username}`];
 
     if (!useBrave) continue;
 

lib/services/scan.ts

@@ -15,7 +15,7 @@ import { getRepliesParallel } from "@/lib/threads-browser/replies";
 import { keywordSearchViaThreadsApi } from "@/lib/threads-api";
 import { getActiveThreadsCredentials } from "@/lib/services/threads-credentials";
 import { computePlacementScore, type RankedPost } from "@/lib/ranking";
-import { parseResearchMap, parseSelectedTags } from "@/lib/types/research";
+import { parseResearchMap, parseSelectedTags, threadsProfileUrl, type SimilarAccount } from "@/lib/types/research";
 import { humanDelay } from "@/lib/utils";
 import { runWithConcurrency } from "@/lib/utils/concurrency";
 import {
@@ -255,7 +255,7 @@ export async function runScanForTopic(
       const accountWebTargets = [...accountTargets.values()];
       const [keywordWebPosts, accountWebPosts] = await Promise.all([
         discoverPostsViaWebSearch(webSearchTags, {
-          perQueryLimit: placementMode ? 8 : 10,
+          perQueryLimit: 15,
           placementMode,
           concurrency: 2,
           tagMeta,
@@ -270,9 +270,10 @@ export async function runScanForTopic(
           },
         }),
         !placementMode && accountWebTargets.length > 0
-          ? discoverPostsFromSimilarAccounts(accountWebTargets.slice(0, 8), {
+          ? discoverPostsFromSimilarAccounts(accountWebTargets.slice(0, 4), {
-              perAccountLimit: placementMode ? 12 : 10,
+              perAccountLimit: 20,
               placementMode,
+              keywordPriority: braveKeywordPriority,
             })
           : Promise.resolve([]),
       ]);
@@ -839,10 +840,13 @@ export async function runScanForTopic(
   await applyQualityFilter(scan.id);
 
   const visibleCount = await prisma.scanItem.count({
-    where: { scanId: scan.id, qualityTier: { not: "EXCLUDE" } },
+    where: { scanId: scan.id, OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }] },
   });
 
   setTaskStatus(progressDetail, "quality", { status: "done", found: visibleCount });
+
+  await enrichAccountsFromScan(scan.id, topic.id);
+
   progressDetail.summary = `完成 · ${ranked.length} 篇 · ${repliesCount} 則留言`;
   await report(progressDetail.summary, progressDetail);
 
@@ -902,6 +906,86 @@ export async function applyQualityFilter(scanId: string) {
   return [];
 }
 
+async function enrichAccountsFromScan(scanId: string, topicId: string) {
+  const scan = await prisma.scan.findUnique({
+    where: { id: scanId },
+    include: {
+      items: {
+        where: { OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }] },
+        orderBy: { combinedScore: "desc" },
+      },
+    },
+  });
+  if (!scan || scan.items.length === 0) return;
+
+  const cutoff = Math.max(3, Math.ceil(scan.items.length * 0.3));
+  const topItems = scan.items.slice(0, cutoff);
+  const authorMap = new Map<string, { count: number; maxScore: number; latestPost: Date; reason: string }>();
+
+  for (const item of topItems) {
+    if (!item.authorName) continue;
+    const key = item.authorName.toLowerCase();
+    const existing = authorMap.get(key);
+    if (existing) {
+      existing.count++;
+      if ((item.combinedScore ?? item.score) > existing.maxScore) {
+        existing.maxScore = item.combinedScore ?? item.score;
+      }
+      if (item.postedAt && item.postedAt > existing.latestPost) {
+        existing.latestPost = item.postedAt;
+      }
+    } else {
+      authorMap.set(key, {
+        count: 1,
+        maxScore: item.combinedScore ?? item.score,
+        latestPost: item.postedAt ?? new Date(0),
+        reason: item.qualityReason || item.text.slice(0, 80) || "海巡發現的高品質作者",
+      });
+    }
+  }
+
+  const topic = await prisma.topic.findUnique({ where: { id: topicId } });
+  if (!topic) return;
+
+  const existingMap = parseResearchMap(topic.researchMap);
+  if (!existingMap) return;
+
+  const existingAccounts = existingMap.similarAccounts ?? [];
+  const existingByKey = new Map(existingAccounts.map((a) => [a.username.toLowerCase(), a]));
+
+  const newAccounts: SimilarAccount[] = [];
+  for (const [key, data] of authorMap) {
+    const existing = existingByKey.get(key);
+    if (existing) {
+      if (!existing.confidence || existing.confidence === "low") {
+        existing.confidence = data.count >= 2 ? "high" : "medium";
+      }
+      if (data.latestPost > new Date(existing.lastActiveAt ?? 0)) {
+        existing.lastActiveAt = data.latestPost.toISOString();
+      }
+      existingByKey.set(key, existing);
+    } else {
+      newAccounts.push({
+        username: key,
+        reason: data.reason,
+        source: "scan",
+        profileUrl: threadsProfileUrl(key) ?? undefined,
+        confidence: data.count >= 2 ? "high" : "medium",
+        lastActiveAt: data.latestPost.toISOString(),
+      });
+    }
+  }
+
+  if (newAccounts.length === 0) return;
+
+  const merged = [...newAccounts, ...existingByKey.values()];
+
+  await prisma.topic.update({
+    where: { id: topicId },
+    data: { researchMap: JSON.stringify({ ...existingMap, similarAccounts: merged }) },
+  });
+}
+
 export async function runScanForAllActiveTopics(accountId?: string | null) {
   const topics = await prisma.topic.findMany({
     where: { active: true, ...(accountId ? { accountId } : {}) },

lib/services/viral.ts

@@ -74,7 +74,7 @@ export async function analyzeScanTopViral(scanId: string, limit = 5) {
   const items = await prisma.scanItem.findMany({
     where: {
       scanId,
-      qualityTier: { not: "EXCLUDE" },
+      OR: [{ qualityTier: null }, { qualityTier: { not: "EXCLUDE" } }],
     },
     orderBy: [{ combinedScore: "desc" }, { score: "desc" }],
     take: limit,

lib/types/research.ts

@@ -31,14 +31,17 @@ export interface SuggestedTag {
   searchType?: SearchTagType;
 }
 
+export type AccountConfidence = "high" | "medium" | "low";
+
 export interface SimilarAccount {
   username: string;
   reason: string;
   /** web=網路搜尋找到的真實連結；threads=Threads 關鍵字搜尋熱門作者 */
   source?: "web" | "threads" | "scan";
   profileUrl?: string;
-  /** 發現此帳號時參考的那篇 Threads 貼文（若有） */
   postUrl?: string;
+  confidence?: AccountConfidence;
+  lastActiveAt?: string;
 }
 
 export interface ResearchMap {
@@ -125,4 +128,13 @@ export const SIMILAR_ACCOUNT_SOURCE_LABELS: Record<
   web: "網路搜尋",
   threads: "Threads 搜尋",
   scan: "海巡發現",
+};
+
+export const SIMILAR_ACCOUNT_CONFIDENCE_LABELS: Record<
+  NonNullable<SimilarAccount["confidence"]>,
+  string
+> = {
+  high: "高",
+  medium: "中",
+  low: "低",
 };