Add renovate.json

.gitea/workflows/main.yaml

@@ -7,80 +7,28 @@ on:
   workflow_dispatch: {}
 
 env:
-  REGISTRY_HOST: git.ivanch.me
+  IMAGE: git.ivanch.me/ivanch/recommender
-  REGISTRY_USERNAME: ivanch
-  IMAGE: ${{ env.REGISTRY_HOST }}/ivanch/recommender
-  KUBE_CONFIG: ${{ secrets.KUBE_CONFIG }}
 
 jobs:
-  build_recommender:
+  build:
-    name: Build Recommender Image
+    name: Build Image
-    runs-on: ubuntu-22.04
-
-    steps:
-      - name: Check out repository
-        uses: actions/checkout@v2
-
-      - name: Log in to Container Registry
-        run: |
-          echo "${{ secrets.REGISTRY_PASSWORD }}" \
-            | docker login "${{ env.REGISTRY_HOST }}" \
-                -u "${{ env.REGISTRY_USERNAME }}" \
-                --password-stdin
-
-      - name: Set up Docker Buildx
-        uses: docker/setup-buildx-action@v3
-
-      - name: Build and Push Multi-Arch Image
-        uses: docker/build-push-action@v6
-        with:
-          push: true
-          context: .
-          platforms: linux/amd64,linux/arm64
-          tags: |
-            ${{ env.IMAGE }}:latest
-
-  deploy_recommender:
-    name: Deploy Recommender (internal)
     runs-on: ubuntu-amd64
-    needs: build_recommender
     steps:
-      - name: Check KUBE_CONFIG validity
+      - uses: actions/checkout@v4
-        run: |
+      - name: Build and Push Multi-Arch Image
-          if [ -z "${KUBE_CONFIG}" ] || [ "${KUBE_CONFIG}" = "" ] || [ "${KUBE_CONFIG// }" = "" ]; then
+        uses: https://git.ivanch.me/ivanch/pipeline-actions/build-and-push@main
-            echo "KUBE_CONFIG is not set or is empty."
+        with:
-            exit 1
+          image: ${{ env.IMAGE }}
-          fi
+          registry_password: ${{ secrets.REGISTRY_PASSWORD }}
+          ssh_key: ${{ secrets.SSH_KEY_DOCKERBUILD }}
 
-      - name: Check out repository
+  deploy:
-        uses: actions/checkout@v2
+    name: Deploy (internal)
-
+    runs-on: ubuntu-amd64
-      - name: Download and install dependencies
+    needs: build
-        run: |
+    steps:
-          apt-get update -y
+      - uses: https://git.ivanch.me/ivanch/pipeline-actions/deploy-restart@main
-          apt-get install -y curl
+        with:
-          curl -LO "https://dl.k8s.io/release/$(curl -L -s https://dl.k8s.io/release/stable.txt)/bin/linux/amd64/kubectl"
+          kube_config: ${{ secrets.KUBE_CONFIG }}
-          install -m 0755 kubectl /usr/local/bin/kubectl
+          deployment_name: recommender
-          kubectl version --client
+          namespace: media
-
-      - name: Set up kubeconfig
-        run: |
-          cd deploy
-          echo "$KUBE_CONFIG" > kubeconfig.yaml
-        env:
-          KUBE_CONFIG: ${{ env.KUBE_CONFIG }}
-
-      - name: Check connection to cluster
-        run: |
-          cd deploy
-          kubectl --kubeconfig=kubeconfig.yaml cluster-info
-
-      - name: Apply Recommender deployment
-        run: |
-          cd deploy
-          kubectl --kubeconfig=kubeconfig.yaml apply -f recommender.yaml
-
-      - name: Rollout restart
-        run: |
-          cd deploy
-          kubectl --kubeconfig=kubeconfig.yaml rollout restart deployment recommender -n media

.gitignore

@@ -25,3 +25,10 @@ documents/
 
 # OS
 Thumbs.db
+
+# Deploy
+deploy/terraform/.terraform
+deploy/terraform/*.tfstate
+deploy/terraform/*.tfstate.*
+deploy/terraform/*.tfplan
+deploy/terraform/.terraform.lock.hcl

CONTEXT.md

@@ -0,0 +1,277 @@
+# TV Show Recommendation System (Self-Hosted, Multi-Agent)
+
+## 🎯 Purpose
+
+This document provides **complete context for AI agents and developers** to build and evolve a local, self-hosted TV show recommendation system.
+
+The system is designed to:
+
+* Prioritize **current user intent** over historical behavior
+* Use **multi-agent architecture**
+* Generate, evaluate, and organize recommendations
+* Handle **50–100 candidate series per request**
+
+---
+
+# 🧠 Core Principles
+
+1. **Context over history**
+
+   * Do NOT persist long-term taste profiles
+   * Every request must be evaluated independently
+
+2. **Generation ≠ Evaluation**
+
+   * Candidate generation and ranking must be separated
+
+3. **High recall → aggressive filtering**
+
+   * Generate many candidates (50–100)
+   * Filter and rank later
+
+4. **Structured communication between agents**
+
+   * All agents must exchange structured data (JSON-like)
+
+5. **Deterministic where possible**
+
+   * Interpretation and ranking should be consistent
+
+---
+
+# 🏗️ System Overview
+
+```
+User Input
+   ↓
+[1] Interpreter Agent
+   ↓
+[2] Retrieval Agent
+   ↓
+[3] Ranking Agent
+   ↓
+[4] Curator Agent
+```
+
+---
+
+# 1. 🧾 Interpreter Agent
+
+## Goal
+
+Convert raw user input into structured data.
+
+## Input
+
+Free-form user text describing:
+
+* Liked series
+* Disliked series
+* Preferences (themes, tone, characters)
+* Constraints (things to avoid)
+
+## Output Schema
+
+```json
+{
+  "liked": ["string"],
+  "disliked": ["string"],
+  "themes": ["string"],
+  "character_preferences": ["string"],
+  "tone": ["string"],
+  "avoid": ["string"]
+}
+```
+
+## Requirements
+
+* Normalize terminology (e.g., "spy" → "espionage")
+* Infer implicit preferences
+* Detect contradictions
+* Be deterministic (low temperature)
+
+---
+
+# 2. 🔎 Retrieval Agent
+
+## Goal
+
+Generate a **large and diverse pool (50–100)** of candidate TV series.
+
+## Strategy: LLM Generation
+
+* Generate candidate series from structured input
+* Focus on diversity and coverage (high LLM temperature)
+
+## Output Schema
+
+```json
+[
+  {
+    "title": "string",
+    "metadata": {
+      "themes": [],
+      "tone": [],
+      "tags": []
+    }
+  }
+]
+```
+
+## Constraints
+
+* Do NOT use external APIs
+* Favor recall over precision
+
+---
+
+# 3. ⚖️ Ranking Agent
+
+## Goal
+
+Categorize candidates into 4 confidence levels using relative comparison.
+
+## Categories
+
+* Definitely Like
+* Might Like
+* Questionable
+* Will Not Like
+
+## Method
+
+### Step 1: Pre-filter
+
+* Remove obvious mismatches
+* Enforce "avoid" constraints
+
+### Step 2: Pairwise Comparison
+
+* Compare candidates relative to each other and input
+
+### Step 3: Tagging
+
+* Assign each show to one of the four categories
+
+## Output Schema
+
+```json
+{
+  "definitely_like": ["title"],
+  "might_like": ["title"],
+  "questionable": ["title"],
+  "will_not_like": ["title"]
+}
+```
+
+## Requirements
+
+* Keep logic simple and consistent
+* Be deterministic (low temperature)
+
+---
+
+# 4. 🎯 Curator Agent
+
+## Goal
+
+Produce a clean, user-facing recommendation output.
+
+## Responsibilities
+
+* Group series by category
+* Provide short explanations
+* Ensure readability
+
+## Output Example
+
+```
+Definitely Like:
+- Show A → reason
+
+Might Like:
+- Show B → reason
+```
+
+---
+
+# 🔁 Feedback Loop (Re-Ranking Only)
+
+## Goal
+
+Improve ranking without storing long-term user preferences.
+
+## Inputs
+
+* External summaries, reviews, or insights
+* User-provided feedback from other sources
+
+## Behavior
+
+* Adjust ranking dynamically
+* Re-rank existing candidates
+* Do NOT persist user taste
+
+---
+
+# ⚙️ Agent Configuration
+
+## Temperature Guidelines
+
+* Interpreter: Low
+* Retrieval: Medium/High
+* Ranking: Low
+* Curator: Medium
+
+## Rules
+
+* Agents must be stateless
+* Pass all required context explicitly
+* No hidden memory
+
+---
+
+# 🧱 Project Structure
+
+```
+/agents
+  interpreter
+  retrieval
+  ranking
+  curator
+
+/pipelines
+  recommendation_flow
+```
+
+---
+
+# 🚫 Non-Goals
+
+* No persistent taste model
+* No long-term user profiling
+
+---
+
+# 🧭 Implementation Notes for AI Agents
+
+When extending or modifying this system:
+
+1. Do NOT introduce long-term memory of user preferences
+2. Do NOT merge agent responsibilities
+3. Always preserve structured input/output between agents
+4. Prefer simplicity over overengineering
+5. Ensure ranking remains interpretable and consistent
+
+---
+
+# ✅ Summary
+
+This system is a **stateless, multi-agent recommendation pipeline** focused on:
+
+* Strong alignment with current input
+* High candidate diversity
+* Structured filtering and ranking
+* Scalable handling of large candidate sets
+
+The architecture is intentionally simple, modular, and extensible.

Dockerfile

@@ -7,7 +7,7 @@ COPY package.json package-lock.json ./
 COPY packages/backend/package.json  ./packages/backend/package.json
 COPY packages/frontend/package.json ./packages/frontend/package.json
 
-RUN npm ci --ignore-scripts
+RUN npm ci
 
 # ─── Stage 2: Build the Preact frontend ──────────────────────────────────────
 FROM deps AS build-frontend

README.md

@@ -1,6 +1,6 @@
 # Recommender
 
-A pure TypeScript monolith AI agent application that will recommend TV shows based on a very customized user profile and input.
+A pure TypeScript monolith AI agent application that will recommend TV series based on a very customized user profile and input.
 
 ## Project Structure
 

deploy/recommender.yaml

@@ -1,72 +0,0 @@
-apiVersion: apps/v1
-kind: Deployment
-metadata:
-  name: recommender
-  namespace: media
-spec:
-  replicas: 1
-  selector:
-    matchLabels:
-      app: recommender
-  template:
-    metadata:
-      labels:
-        app: recommender
-    spec:
-      containers:
-        - name: recommender
-          image: git.ivanch.me/ivanch/recommender:latest
-          imagePullPolicy: Always
-          ports:
-            - containerPort: 80
-          env:
-            - name: OPENAI_API_KEY
-              valueFrom:
-                secretKeyRef:
-                  name: recommender-secrets
-                  key: OPENAI_API_KEY
-            - name: DATABASE_URL
-              valueFrom:
-                secretKeyRef:
-                  name: recommender-secrets
-                  key: DATABASE_URL
-          resources:
-            requests:
-              memory: "256Mi"
-              cpu: "500m"
-            limits:
-              memory: "512Mi"
-              cpu: "1"
----
-apiVersion: v1
-kind: Service
-metadata:
-  name: recommender
-  namespace: media
-spec:
-  selector:
-    app: recommender
-  ports:
-    - port: 80
-      targetPort: 80
----
-apiVersion: networking.k8s.io/v1
-kind: Ingress
-metadata:
-  name: recommender
-  namespace: media
-  labels:
-    app.kubernetes.io/name: recommender
-spec:
-  ingressClassName: nginx
-  rules:
-    - host: "recommender.haven"
-      http:
-        paths:
-          - path: "/"
-            pathType: Prefix
-            backend:
-              service:
-                name: recommender
-                port:
-                  number: 80

entrypoint.sh

@@ -4,5 +4,8 @@ set -e
 # Start Nginx in the background
 nginx &
 
+# Run migrations
+node /app/node_modules/.bin/tsx /app/packages/backend/src/migrate.ts
+
 # Start the Node.js backend
 exec node /app/node_modules/.bin/tsx /app/packages/backend/src/index.ts

package-lock.json

@@ -4455,6 +4455,15 @@
       "dev": true,
       "license": "MIT"
     },
+    "node_modules/zod": {
+      "version": "4.3.6",
+      "resolved": "https://registry.npmjs.org/zod/-/zod-4.3.6.tgz",
+      "integrity": "sha512-rftlrkhHZOcjDwkGlnUtZZkvaPHCsDATp4pGpuOOMDaTdDDXF91wuVDJoWoPsKX/3YPQ5fHuF3STjcYyKr+Qhg==",
+      "license": "MIT",
+      "funding": {
+        "url": "https://github.com/sponsors/colinhacks"
+      }
+    },
     "packages/backend": {
       "version": "1.0.0",
       "license": "ISC",
@@ -4463,7 +4472,8 @@
         "drizzle-orm": "^0.45.1",
         "fastify": "^5.8.4",
         "openai": "^6.32.0",
-        "postgres": "^3.4.8"
+        "postgres": "^3.4.8",
+        "zod": "^4.3.6"
       },
       "devDependencies": {
         "@types/node": "^24.12.0",

packages/backend/.env

@@ -3,4 +3,8 @@
 # In production / Docker, supply these as environment variables.
 
 DATABASE_URL=postgres://user:password@localhost:5432/recommender
+
 OPENAI_API_KEY=your-openai-api-key-here
+
+# provider settings
+PROVIDER_URL=https://your-provider.example.com/v1

packages/backend/drizzle/0000_wild_joseph.sql

@@ -1,4 +1,4 @@
-CREATE TABLE "feedback" (
+CREATE TABLE IF NOT EXISTS "feedback" (
 	"id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
 	"tv_show_name" text NOT NULL,
 	"stars" integer NOT NULL,
@@ -6,12 +6,12 @@ CREATE TABLE "feedback" (
 	"created_at" timestamp DEFAULT now() NOT NULL
 );
 --> statement-breakpoint
-CREATE TABLE "recommendations" (
+CREATE TABLE IF NOT EXISTS "recommendations" (
 	"id" uuid PRIMARY KEY DEFAULT gen_random_uuid() NOT NULL,
 	"title" text NOT NULL,
 	"main_prompt" text NOT NULL,
-	"liked_shows" text DEFAULT '' NOT NULL,
+	"liked_series" text DEFAULT '' NOT NULL,
-	"disliked_shows" text DEFAULT '' NOT NULL,
+	"disliked_series" text DEFAULT '' NOT NULL,
 	"themes" text DEFAULT '' NOT NULL,
 	"brainstorm_count" integer DEFAULT 100 NOT NULL,
 	"recommendations" jsonb,
@@ -19,4 +19,4 @@ CREATE TABLE "recommendations" (
 	"created_at" timestamp DEFAULT now() NOT NULL
 );
 --> statement-breakpoint
-CREATE UNIQUE INDEX "feedback_tv_show_name_idx" ON "feedback" USING btree ("tv_show_name");
+CREATE UNIQUE INDEX IF NOT EXISTS "feedback_tv_show_name_idx" ON "feedback" USING btree ("tv_show_name");

packages/backend/drizzle/0001_add_media_type_web_search.sql

@@ -0,0 +1,9 @@
+ALTER TABLE "recommendations" ADD COLUMN "media_type" text DEFAULT 'tv_show' NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "recommendations" ADD COLUMN "use_web_search" boolean DEFAULT false NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "feedback" RENAME COLUMN "tv_show_name" TO "item_name";
+--> statement-breakpoint
+DROP INDEX "feedback_tv_show_name_idx";
+--> statement-breakpoint
+CREATE UNIQUE INDEX "feedback_item_name_idx" ON "feedback" USING btree ("item_name");

packages/backend/drizzle/0002_add_validator_hard_requirements_expansive.sql

@@ -0,0 +1,9 @@
+ALTER TABLE "recommendations" ADD COLUMN "use_validator" boolean DEFAULT false NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "recommendations" ADD COLUMN "hard_requirements" boolean DEFAULT false NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "recommendations" ADD COLUMN "self_expansive" boolean DEFAULT false NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "recommendations" ADD COLUMN "expansive_passes" integer DEFAULT 1 NOT NULL;
+--> statement-breakpoint
+ALTER TABLE "recommendations" ADD COLUMN "expansive_mode" text DEFAULT 'soft' NOT NULL;

packages/backend/drizzle/0003_changing_tv_series.sql

@@ -0,0 +1,2 @@
+ALTER TABLE "recommendations" RENAME COLUMN "liked_shows" TO "liked_series";
+ALTER TABLE "recommendations" RENAME COLUMN "disliked_shows" TO "disliked_series";

packages/backend/drizzle/meta/0000_snapshot.json

@@ -89,15 +89,15 @@
           "primaryKey": false,
           "notNull": true
         },
-        "liked_shows": {
+        "liked_series": {
-          "name": "liked_shows",
+          "name": "liked_series",
           "type": "text",
           "primaryKey": false,
           "notNull": true,
           "default": "''"
         },
-        "disliked_shows": {
+        "disliked_series": {
-          "name": "disliked_shows",
+          "name": "disliked_series",
           "type": "text",
           "primaryKey": false,
           "notNull": true,

packages/backend/drizzle/meta/_journal.json

@@ -8,6 +8,13 @@
       "when": 1774479321371,
       "tag": "0000_wild_joseph",
       "breakpoints": true
+    },
+    {
+      "idx": 1,
+      "version": "7",
+      "when": 1774900000000,
+      "tag": "0001_add_media_type_web_search",
+      "breakpoints": true
     }
   ]
 }

packages/backend/package.json

@@ -5,7 +5,8 @@
   "main": "index.js",
   "scripts": {
     "test": "echo \"Error: no test specified\" && exit 1",
-    "dev": "tsx watch src/index.ts"
+    "dev": "tsx watch src/index.ts",
+    "migrate": "tsx src/migrate.ts"
   },
   "keywords": [],
   "author": "",
@@ -16,7 +17,8 @@
     "drizzle-orm": "^0.45.1",
     "fastify": "^5.8.4",
     "openai": "^6.32.0",
-    "postgres": "^3.4.8"
+    "postgres": "^3.4.8",
+    "zod": "^4.3.6"
   },
   "devDependencies": {
     "@types/node": "^24.12.0",

packages/backend/src/agent.ts

@@ -3,19 +3,37 @@ import * as dotenv from 'dotenv';
 dotenv.config({ path: ['.env.local', '.env'] });
 
 export const openai = new OpenAI({
-  apiKey: process.env.OPENAI_API_KEY,
+  apiKey: process.env.BEARER_TOKEN,
+  baseURL: process.env.PROVIDER_URL,
+  timeout: 600000, // 10 minutes
+  maxRetries: 3,
 });
 
-export async function askAgent(prompt: string) {
+export const defaultModel = process.env.MODEL_NAME ?? 'gpt-5.4';
+export const miniModel = process.env.MODEL_NAME ?? 'gpt-5.4-mini';
+export const serviceOptions = { service_tier: 'flex' as const };
+export const supportsWebSearch = true;
+
+function isJsonParseError(err: unknown): boolean {
+  if (err instanceof SyntaxError) return true;
+  if (!(err instanceof Error)) return false;
+
+  // Some providers wrap JSON parsing failures in plain Error objects.
+  return /not valid json|unexpected token|json/i.test(err.message);
+}
+
+export async function parseWithRetry<T>(fn: () => Promise<T>, retries = 3): Promise<T> {
+  let lastErr: unknown;
+  for (let attempt = 0; attempt <= retries; attempt++) {
     try {
-    const response = await openai.chat.completions.create({
+      return await fn();
-      model: 'gpt-5.4',
-      service_tier: 'flex',
-      messages: [{ role: 'user', content: prompt }],
-    });
-    return response!.choices![0]!.message!.content;
     } catch (err) {
-    console.error('Agent endpoint dummy error:', err instanceof Error ? err.message : err);
+      if (isJsonParseError(err) && attempt < retries) {
-    return 'Agent is in dummy mode or encountered an error.';
+        lastErr = err;
+        continue;
       }
+      throw err;
+    }
+  }
+  throw lastErr;
 }

packages/backend/src/agents/curator.ts

@@ -1,66 +1,80 @@
-import { openai } from '../agent.js';
+import { openai, defaultModel, serviceOptions, supportsWebSearch, parseWithRetry } from '../agent.js';
-import type { InterpreterOutput, RankingOutput, CuratorOutput } from '../types/agents.js';
+import type { InterpreterOutput, RankingOutput, CuratorOutput, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
+
+const CuratorSchema = z.object({
+  series: z.array(z.object({
+    title: z.string(),
+    explanation: z.string(),
+    category: z.enum(["Full Match", "Definitely Like", "Might Like", "Questionable", "Will Not Like"]),
+    genre: z.string(),
+    pros: z.array(z.string()).max(3),
+    cons: z.array(z.string()).max(3)
+  }))
+});
+
+const CHUNK_SIZE = 20;
 
 export async function runCurator(
   ranking: RankingOutput,
   interpreter: InterpreterOutput,
+  mediaType: MediaType = 'tv_show',
+  useWebSearch = false,
 ): Promise<CuratorOutput[]> {
-  const allShows = [
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
+
+  const allSeries = [
+    ...(ranking.full_match ?? []).map((t) => ({ title: t, category: 'Full Match' as const })),
     ...ranking.definitely_like.map((t) => ({ title: t, category: 'Definitely Like' as const })),
     ...ranking.might_like.map((t) => ({ title: t, category: 'Might Like' as const })),
     ...ranking.questionable.map((t) => ({ title: t, category: 'Questionable' as const })),
     ...ranking.will_not_like.map((t) => ({ title: t, category: 'Will Not Like' as const })),
   ];
 
-  if (allShows.length === 0) return [];
+  if (allSeries.length === 0) return [];
 
-  const showList = allShows
+  const canSearch = useWebSearch && supportsWebSearch;
-    .map((s) => `- "${s.title}" (${s.category})`)
+  const preferenceSummary = `User preferences summary:
-    .join('\n');
+Liked: ${interpreter.liked.join(', ') || '(none)'}
+Themes: ${interpreter.themes.join(', ') || '(none)'}
+Tone: ${interpreter.tone.join(', ') || '(none)'}
+Character preferences: ${interpreter.character_preferences.join(', ') || '(none)'}
+Avoid: ${interpreter.avoid.join(', ') || '(none)'}
+Requirements: ${interpreter.requirements.join(', ') || '(none)'}`;
 
-  const response = await openai.chat.completions.create({
+  const instructions = `You are a ${mediaLabel} recommendation curator. For each ${mediaLabel}, write a concise explanation and surface the most useful details for the user.${useWebSearch ? '\n\nUse web search to verify details and enrich explanations with accurate information.' : ''}
-    model: 'gpt-5.4-mini',
-    temperature: 0.5,
-    service_tier: 'flex',
-    response_format: { type: 'json_object' },
-    messages: [
-      {
-        role: 'system',
-        content: `You are a TV show recommendation curator. For each show, write a concise 1-2 sentence explanation of why it was assigned to its category based on the user's preferences.
-
-Your output MUST be valid JSON:
-{
-  "shows": [
-    {
-      "title": string,
-      "explanation": string,
-      "category": "Definitely Like" | "Might Like" | "Questionable" | "Will Not Like"
-    }
-  ]
-}
 
 Rules:
 - Preserve the exact title and category as given
-- Keep explanations concise (1-2 sentences max)
+- explanation: 1-2 sentences explaining why it was assigned to its category, referencing specific user preferences
-- Reference specific user preferences in the explanation
+- genre: 1-3 words capturing the most prominent genre of the title (e.g. "Crime Drama", "Sci-Fi Thriller", "Romantic Comedy")
-- Be honest — explain why "Questionable" or "Will Not Like" shows got that rating`,
+- pros: up to 3 short bullet points about what this title does well relative to the user's taste
-      },
+- cons: up to 3 short bullet points about what the user might not like based on their preferences
-      {
+- Be honest — explain why "Questionable" or "Will Not Like" ${mediaLabel}s got that rating`;
-        role: 'user',
-        content: `User preferences summary:
-Liked: ${JSON.stringify(interpreter.liked)}
-Themes: ${JSON.stringify(interpreter.themes)}
-Tone: ${JSON.stringify(interpreter.tone)}
-Character preferences: ${JSON.stringify(interpreter.character_preferences)}
-Avoid: ${JSON.stringify(interpreter.avoid)}
 
-Shows to describe:
+  const chunks: typeof allSeries[] = [];
+  for (let i = 0; i < allSeries.length; i += CHUNK_SIZE) {
+    chunks.push(allSeries.slice(i, i + CHUNK_SIZE));
+  }
+
+  const results: CuratorOutput[] = [];
+  for (const chunk of chunks) {
+    const showList = chunk.map((s) => `- "${s.title}" (${s.category})`).join('\n');
+    const response = await parseWithRetry(() => openai.responses.parse({
+      model: defaultModel,
+      temperature: 0.5,
+      ...serviceOptions,
+      ...(canSearch ? { tools: [{ type: 'web_search' as const }] } : {}),
+      text: { format: zodTextFormat(CuratorSchema, "series") },
+      instructions,
+      input: `${preferenceSummary}
+
+${mediaLabel}s to describe:
 ${showList}`,
-      },
+    }));
-    ],
+    results.push(...(response.output_parsed?.series ?? []));
-  });
+  }
 
-  const content = response.choices[0]?.message?.content ?? '{"shows":[]}';
+  return results;
-  const result = JSON.parse(content) as { shows: CuratorOutput[] };
-  return result.shows ?? [];
 }

packages/backend/src/agents/interpreter.ts

@@ -1,56 +1,54 @@
-import { openai } from '../agent.js';
+import { openai, defaultModel, serviceOptions, parseWithRetry } from '../agent.js';
-import type { InterpreterOutput } from '../types/agents.js';
+import type { InterpreterOutput, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
+
+const InterpreterSchema = z.object({
+  liked: z.array(z.string()),
+  disliked: z.array(z.string()),
+  themes: z.array(z.string()),
+  character_preferences: z.array(z.string()),
+  tone: z.array(z.string()),
+  avoid: z.array(z.string()),
+  requirements: z.array(z.string())
+});
 
 interface InterpreterInput {
   main_prompt: string;
-  liked_shows: string;
+  liked_series: string;
-  disliked_shows: string;
+  disliked_series: string;
   themes: string;
+  media_type: MediaType;
   feedback_context?: string;
 }
 
 export async function runInterpreter(input: InterpreterInput): Promise<InterpreterOutput> {
+  const mediaLabel = input.media_type === 'movie' ? 'movie' : 'TV show';
   const feedbackSection = input.feedback_context
     ? `\n\nUser Feedback Context (incorporate into preferences):\n${input.feedback_context}`
     : '';
 
-  const response = await openai.chat.completions.create({
+  const response = await parseWithRetry(() => openai.responses.parse({
-    model: 'gpt-5.4-mini',
+    model: defaultModel,
     temperature: 0.2,
-    service_tier: 'flex',
+    ...serviceOptions,
-    response_format: { type: 'json_object' },
+    text: { format: zodTextFormat(InterpreterSchema, "preferences") },
-    messages: [
+    instructions: `You are a ${mediaLabel} preference interpreter. Transform raw user input into structured, normalized preferences.
-      {
-        role: 'system',
-        content: `You are a TV show preference interpreter. Transform raw user input into structured, normalized preferences.
-
-Your output MUST be valid JSON matching this schema:
-{
-  "liked": string[],        // shows the user likes
-  "disliked": string[],     // shows the user dislikes
-  "themes": string[],       // normalized themes (e.g. "spy" -> "espionage")
-  "character_preferences": string[], // character types they prefer
-  "tone": string[],         // tone descriptors (e.g. "serious", "grounded", "dark")
-  "avoid": string[]         // things to explicitly avoid
-}
 
 Rules:
 - Extract implicit preferences from the main prompt
 - Normalize terminology (e.g. "spy" → "espionage", "cop show" → "police procedural")
 - Detect and resolve contradictions (prefer explicit over implicit)
 - Do NOT assume anything not stated or clearly implied
-- Be specific and concrete, not vague`,
+- Be specific and concrete, not vague
-      },
+- For "requirements": capture explicit hard requirements the user stated that recommendations must satisfy — things like "must be from the 2000s onward", "must have subtitles", "must feature a female lead". Leave empty if no such constraints were stated.`,
-      {
+    input: `Main prompt: ${input.main_prompt}
-        role: 'user',
+Liked ${mediaLabel}s: ${input.liked_series || '(none)'}
-        content: `Main prompt: ${input.main_prompt}
+Disliked ${mediaLabel}s: ${input.disliked_series || '(none)'}
-Liked shows: ${input.liked_shows || '(none)'}
-Disliked shows: ${input.disliked_shows || '(none)'}
 Themes and requirements: ${input.themes || '(none)'}${feedbackSection}`,
-      },
+  }));
-    ],
-  });
 
-  const content = response.choices[0]?.message?.content ?? '{}';
+  return (response.output_parsed as InterpreterOutput) ?? {
-  return JSON.parse(content) as InterpreterOutput;
+    liked: [], disliked: [], themes: [], character_preferences: [], tone: [], avoid: [], requirements: []
+  };
 }

packages/backend/src/agents/ranking.ts

@@ -1,10 +1,24 @@
-import { openai } from '../agent.js';
+import { openai, defaultModel, serviceOptions, parseWithRetry } from '../agent.js';
-import type { InterpreterOutput, RetrievalOutput, RankingOutput } from '../types/agents.js';
+import type { InterpreterOutput, RetrievalOutput, RankingOutput, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
+
+const RankingSchema = z.object({
+  full_match: z.array(z.string()),
+  definitely_like: z.array(z.string()),
+  might_like: z.array(z.string()),
+  questionable: z.array(z.string()),
+  will_not_like: z.array(z.string())
+});
 
 export async function runRanking(
   interpreter: InterpreterOutput,
   retrieval: RetrievalOutput,
+  mediaType: MediaType = 'tv_show',
+  hardRequirements = false,
 ): Promise<RankingOutput> {
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
+
   // Phase 1: Pre-filter — remove avoidance violations
   const avoidList = interpreter.avoid.map((a) => a.toLowerCase());
   const filtered = retrieval.candidates.filter((c) => {
@@ -19,7 +33,8 @@ export async function runRanking(
     chunks.push(filtered.slice(i, i + CHUNK_SIZE));
   }
 
-  const allBuckets: RankingOutput = {
+  const allTags: RankingOutput = {
+    full_match: [],
     definitely_like: [],
     might_like: [],
     questionable: [],
@@ -29,55 +44,41 @@ export async function runRanking(
   for (const chunk of chunks) {
     const chunkTitles = chunk.map((c) => `- ${c.title}: ${c.reason}`).join('\n');
 
-    const response = await openai.chat.completions.create({
+    const response = await parseWithRetry(() => openai.responses.parse({
-      model: 'gpt-5.4',
+      model: defaultModel,
       temperature: 0.2,
-      service_tier: 'flex',
+      ...serviceOptions,
-      response_format: { type: 'json_object' },
+      text: { format: zodTextFormat(RankingSchema, "ranking") },
-      messages: [
+      instructions: `You are a ${mediaLabel} ranking critic. Assign each ${mediaLabel} to exactly one of five confidence tags based on how well it matches the user's preferences.
-        {
-          role: 'system',
-          content: `You are a TV show ranking critic. Assign each show to exactly one of four confidence buckets based on how well it matches the user's preferences.
 
-Buckets:
+Tags:
-- "definitely_like": Near-perfect match to all preferences
+- "full_match": 100% match — perfectly satisfies every stated preference, requirement, and avoidance criteria with no compromises
+- "definitely_like": Near-perfect match to all preferences with only minor caveats
 - "might_like": Strong match to most preferences
 - "questionable": Partial alignment, some aspects don't match
 - "will_not_like": Likely mismatch, conflicts with preferences or avoidance criteria
 
-Your output MUST be valid JSON:
+Every ${mediaLabel} in the input must appear in exactly one tag. Use the title exactly as given.${hardRequirements ? '\n\nHARD REQUIREMENTS MODE: Any candidate that does not satisfy every stated requirement must be placed in "will_not_like", regardless of other qualities.' : ''}`,
-{
+      input: `User preferences:
-  "definitely_like": string[],
+Liked ${mediaLabel}s: ${interpreter.liked.join(', ') || '(none)'}
-  "might_like": string[],
+Themes: ${interpreter.themes.join(', ') || '(none)'}
-  "questionable": string[],
+Character preferences: ${interpreter.character_preferences.join(', ') || '(none)'}
-  "will_not_like": string[]
+Tone: ${interpreter.tone.join(', ') || '(none)'}
-}
+Avoid: ${interpreter.avoid.join(', ') || '(none)'}
+Requirements: ${interpreter.requirements.join(', ') || '(none)'}
 
-Every show in the input must appear in exactly one bucket. Use the title exactly as given.`,
+Rank these ${mediaLabel}s:
-        },
-        {
-          role: 'user',
-          content: `User preferences:
-Liked shows: ${JSON.stringify(interpreter.liked)}
-Themes: ${JSON.stringify(interpreter.themes)}
-Character preferences: ${JSON.stringify(interpreter.character_preferences)}
-Tone: ${JSON.stringify(interpreter.tone)}
-Avoid: ${JSON.stringify(interpreter.avoid)}
-
-Rank these shows:
 ${chunkTitles}`,
-        },
+    }));
-      ],
-    });
 
-    const content = response.choices[0]?.message?.content ?? '{}';
+    const chunkResult = (response.output_parsed as Partial<RankingOutput>) ?? {};
-    const chunkResult = JSON.parse(content) as Partial<RankingOutput>;
 
-    allBuckets.definitely_like.push(...(chunkResult.definitely_like ?? []));
+    allTags.full_match.push(...(chunkResult.full_match ?? []));
-    allBuckets.might_like.push(...(chunkResult.might_like ?? []));
+    allTags.definitely_like.push(...(chunkResult.definitely_like ?? []));
-    allBuckets.questionable.push(...(chunkResult.questionable ?? []));
+    allTags.might_like.push(...(chunkResult.might_like ?? []));
-    allBuckets.will_not_like.push(...(chunkResult.will_not_like ?? []));
+    allTags.questionable.push(...(chunkResult.questionable ?? []));
+    allTags.will_not_like.push(...(chunkResult.will_not_like ?? []));
   }
 
-  return allBuckets;
+  return allTags;
 }

packages/backend/src/agents/retrieval.ts

@@ -1,47 +1,54 @@
-import { openai } from '../agent.js';
+import { openai, defaultModel, serviceOptions, supportsWebSearch, parseWithRetry } from '../agent.js';
-import type { InterpreterOutput, RetrievalOutput } from '../types/agents.js';
+import type { InterpreterOutput, RetrievalOutput, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
 
-export async function runRetrieval(input: InterpreterOutput, brainstormCount = 100): Promise<RetrievalOutput> {
+const RetrievalSchema = z.object({
-  const response = await openai.chat.completions.create({
+  candidates: z.array(z.object({
-    model: 'gpt-5.4',
+    title: z.string(),
+    reason: z.string()
+  }))
+});
+
+export async function runRetrieval(
+  input: InterpreterOutput,
+  brainstormCount = 100,
+  mediaType: MediaType = 'tv_show',
+  useWebSearch = false,
+  hardRequirements = false,
+  previousFullMatches: string[] = [],
+): Promise<RetrievalOutput> {
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
+  const mediaLabelPlural = mediaType === 'movie' ? 'movies' : 'TV series';
+
+  const canSearch = useWebSearch && supportsWebSearch;
+  const response = await parseWithRetry(() => openai.responses.parse({
+    model: defaultModel,
     temperature: 0.9,
-    service_tier: 'flex',
+    ...serviceOptions,
-    response_format: { type: 'json_object' },
+    ...(canSearch ? { tools: [{ type: 'web_search' as const }] } : {}),
-    messages: [
+    text: { format: zodTextFormat(RetrievalSchema, "candidates") },
-      {
+    instructions: `You are a ${mediaLabel} candidate generator. Your goal is to brainstorm a LARGE, DIVERSE pool of ${brainstormCount} ${mediaLabel} candidates that match the user's structured preferences.${useWebSearch ? '\n\nUse web search to find recent and accurate titles, including newer releases.' : ''}
-        role: 'system',
-        content: `You are a TV show candidate generator. Your goal is to brainstorm a LARGE, DIVERSE pool of ${brainstormCount} TV show candidates that match the user's structured preferences.
-
-Your output MUST be valid JSON matching this schema:
-{
-  "candidates": [
-    { "title": string, "reason": string }
-  ]
-}
 
 Rules:
-- Include both well-known and obscure shows
+- Include both well-known and obscure ${mediaLabelPlural}
 - Prioritize RECALL over precision — it's better to include too many than too few
-- Each "reason" should briefly explain why the show matches the preferences
+- Each "reason" should briefly explain why the ${mediaLabel} matches the preferences
 - Avoid duplicates
-- Include shows from different decades, countries, and networks
+- Include ${mediaLabelPlural} from different decades, countries${mediaType === 'tv_show' ? ', and networks' : ', and directors'}
-- Aim for ${brainstormCount} candidates minimum`,
+- The "title" field must contain ONLY the exact title name — no years, descriptions, network names, episode counts, or parenthetical notes. ✗ Bad: "Breaking Bad (2008–2013, AMC)" ✓ Good: "Breaking Bad"
-      },
+- Aim for ${brainstormCount} candidates minimum${previousFullMatches.length > 0 ? '\n- Do NOT suggest titles already in the Previous Full Matches list — generate NEW candidates inspired by what made those successful' : ''}${hardRequirements ? '\n\nIMPORTANT: Strictly follow ALL requirements. Exclude any candidate that does not meet every stated requirement.' : ''}`,
-      {
+    input: `Structured preferences:
-        role: 'user',
+Liked ${mediaLabelPlural}: ${input.liked.join(', ') || '(none)'}
-        content: `Structured preferences:
+Disliked ${mediaLabelPlural}: ${input.disliked.join(', ') || '(none)'}
-Liked shows: ${JSON.stringify(input.liked)}
+Themes: ${input.themes.join(', ') || '(none)'}
-Disliked shows: ${JSON.stringify(input.disliked)}
+Character preferences: ${input.character_preferences.join(', ') || '(none)'}
-Themes: ${JSON.stringify(input.themes)}
+Tone: ${input.tone.join(', ') || '(none)'}
-Character preferences: ${JSON.stringify(input.character_preferences)}
+Avoid: ${input.avoid.join(', ') || '(none)'}
-Tone: ${JSON.stringify(input.tone)}
+Requirements: ${input.requirements.join(', ') || '(none)'}${previousFullMatches.length > 0 ? `\n\nPrevious Full Match titles (DO NOT repeat these; use them as inspiration for NEW candidates with similar qualities): ${previousFullMatches.join(', ')}` : ''}
-Avoid: ${JSON.stringify(input.avoid)}
 
-Generate a large, diverse pool of TV show candidates.`,
+Generate a large, diverse pool of ${mediaLabel} candidates.`,
-      },
+  }));
-    ],
-  });
 
-  const content = response.choices[0]?.message?.content ?? '{"candidates":[]}';
+  return (response.output_parsed as RetrievalOutput) ?? { candidates: [] };
-  return JSON.parse(content) as RetrievalOutput;
 }

packages/backend/src/agents/retrievalContinuous.ts

@@ -0,0 +1,116 @@
+import { openai, defaultModel, serviceOptions, supportsWebSearch, parseWithRetry } from '../agent.js';
+import type { InterpreterOutput, RetrievalCandidate, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
+
+const RetrievalBatchSchema = z.object({
+  candidates: z.array(z.object({
+    title: z.string(),
+    type: z.enum(['movie', 'tv']),
+    year: z.string(),
+    reason: z.string()
+  }))
+});
+
+export interface RetrievalBatchOutput {
+  candidates: Array<{
+    title: string;
+    type: 'movie' | 'tv';
+    year: string;
+    reason: string;
+  }>;
+}
+
+const RetrievalCandidateSchema = z.object({
+  title: z.string(),
+  reason: z.string()
+});
+
+export interface RetrievalOutput {
+  candidates: RetrievalCandidate[];
+}
+
+function buildSystemPrompt(
+  interpreterOutput: InterpreterOutput,
+  mediaType: MediaType,
+  useWebSearch: boolean,
+  hardRequirements: boolean,
+  seenTitles: string[]
+): string {
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
+  const mediaLabelPlural = mediaType === 'movie' ? 'movies' : 'TV series';
+
+  let prompt = `You are a ${mediaLabel} recommendation specialist. Your task is to recommend titles that match the user's taste profile.
+
+USER TASTE PROFILE:
+- Liked ${mediaLabelPlural}: ${interpreterOutput.liked.join(', ') || '(none)'}
+- Disliked ${mediaLabelPlural}: ${interpreterOutput.disliked.join(', ') || '(none)'}
+- Themes: ${interpreterOutput.themes.join(', ') || '(none)'}
+- Tone: ${interpreterOutput.tone.join(', ') || '(none)'}
+- Requirements: ${interpreterOutput.requirements.join(', ') || '(none)'}
+- Avoid: ${interpreterOutput.avoid.join(', ') || '(none)'}
+
+RESPONSE FORMAT:
+Output exactly 10 recommendations in the following JSON format. No additional text.
+
+[
+  { "title": "Title Name", "type": "movie"|"tv", "year": "2024", "reason": "1-sentence why it matches" },
+  ...
+]
+
+REQUIREMENTS:
+- Output exactly 10 titles, no more, no less
+- Each title must include: title, type (movie/tv), year, and a concise reason
+- Prioritize variety across themes and decades
+- Do not include titles the user already mentioned as liked/disliked
+${seenTitles.length > 0 ? `- Do NOT repeat the following titles that have already been suggested: ${seenTitles.join(', ')}` : ''}
+${useWebSearch ? '\n- Use web search to find recent and accurate titles, including newer releases' : ''}
+${hardRequirements ? '\n- Strictly follow ALL requirements. Exclude any candidate that does not meet every stated requirement.' : ''}
+- If asked for "more 10" or "more recommendations", provide 10 new recommendations following the same format`;
+
+  return prompt;
+}
+
+export async function runRetrievalBatch(
+  interpreterOutput: InterpreterOutput,
+  mediaType: MediaType,
+  useWebSearch: boolean,
+  hardRequirements: boolean,
+  seenTitles: string[],
+  previousResponseId: string | null
+): Promise<{ candidates: RetrievalCandidate[]; responseId: string }> {
+  const canSearch = useWebSearch && supportsWebSearch;
+  const systemPrompt = buildSystemPrompt(interpreterOutput, mediaType, useWebSearch, hardRequirements, seenTitles);
+
+  let input: string;
+
+  if (previousResponseId === null) {
+    input = 'Please recommend 10 titles that match my taste profile. Output exactly 10 recommendations in the specified JSON format.';
+  } else {
+    input = 'Give me 10 more recommendations following the same JSON format as before.';
+  }
+
+  const response = await parseWithRetry(() => openai.responses.parse({
+    model: defaultModel,
+    ...serviceOptions,
+    ...(canSearch ? { tools: [{ type: 'web_search' as const }] } : {}),
+    ...(previousResponseId ? { previous_response_id: previousResponseId } : {}),
+    text: { format: zodTextFormat(RetrievalBatchSchema, "candidates") },
+    instructions: systemPrompt,
+    input: input,
+  }));
+
+  const parsed = response.output_parsed as RetrievalBatchOutput | undefined;
+  const responseId = response.id;
+
+  if (!parsed || !parsed.candidates) {
+    return { candidates: [], responseId };
+  }
+
+  const candidates: RetrievalCandidate[] = parsed.candidates.map((c) => ({
+    title: c.title,
+    reason: c.reason,
+  }));
+
+  return { candidates, responseId };
+}

packages/backend/src/agents/titleGenerator.ts

@@ -1,28 +1,22 @@
-import { openai } from '../agent.js';
+import { openai, miniModel, serviceOptions } from '../agent.js';
-import type { InterpreterOutput } from '../types/agents.js';
+import type { InterpreterOutput, MediaType } from '../types/agents.js';
 
-export async function generateTitle(interpreter: InterpreterOutput): Promise<string> {
+export async function generateTitle(interpreter: InterpreterOutput, mediaType: MediaType = 'tv_show'): Promise<string> {
-  const response = await openai.chat.completions.create({
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
-    model: 'gpt-4o-mini',
+
+  const response = await openai.responses.create({
+    model: miniModel,
     temperature: 0.7,
-    service_tier: 'flex',
+    ...serviceOptions,
-    messages: [
+    instructions: `Generate a concise 5-8 word title for a ${mediaLabel} recommendation session.
-      {
-        role: 'system',
-        content: `Generate a concise 5-8 word title for a TV show recommendation session.
 Capture the essence of the user's taste — genre, tone, key themes.
 Respond with ONLY the title. No quotes, no trailing punctuation.
 Examples: "Dark Crime Dramas With Moral Ambiguity", "Cozy British Mysteries With Quirky Detectives"`,
-      },
+    input: `Liked: ${JSON.stringify(interpreter.liked)}
-      {
-        role: 'user',
-        content: `Liked: ${JSON.stringify(interpreter.liked)}
 Themes: ${JSON.stringify(interpreter.themes)}
 Tone: ${JSON.stringify(interpreter.tone)}
 Character preferences: ${JSON.stringify(interpreter.character_preferences)}`,
-      },
-    ],
   });
 
-  return (response.choices[0]?.message?.content ?? '').trim() || 'My Recommendation Session';
+  return (response.output_text ?? '').trim() || 'My Recommendation Session';
 }

packages/backend/src/agents/validator.ts

@@ -0,0 +1,67 @@
+import { openai, defaultModel, serviceOptions, supportsWebSearch, parseWithRetry } from '../agent.js';
+import type { RetrievalCandidate, ValidatorOutput, MediaType } from '../types/agents.js';
+import { z } from 'zod';
+import { zodTextFormat } from 'openai/helpers/zod';
+
+const ValidatorSchema = z.object({
+  candidates: z.array(z.object({
+    title: z.string(),
+    reason: z.string(),
+    isTrash: z.boolean(),
+  })),
+});
+
+const CHUNK_SIZE = 30;
+
+function splitIntoChunks<T>(items: T[], size: number): T[][] {
+  const chunks: T[][] = [];
+  for (let i = 0; i < items.length; i += size) {
+    chunks.push(items.slice(i, i + size));
+  }
+  return chunks;
+}
+
+async function runValidatorChunk(
+  candidates: RetrievalCandidate[],
+  mediaLabel: string,
+): Promise<ValidatorOutput> {
+  const list = candidates.map((c) => `- ${c.title}: ${c.reason}`).join('\n');
+
+  const response = await parseWithRetry(() => openai.responses.parse({
+    model: defaultModel,
+    temperature: 0.1,
+    ...serviceOptions,
+    ...(supportsWebSearch ? { tools: [{ type: 'web_search' as const }] } : {}),
+    text: { format: zodTextFormat(ValidatorSchema, 'validation') },
+    instructions: `You are a ${mediaLabel} metadata validator. For each candidate in the list, use web search to verify:
+1. The title actually exists as a real, produced ${mediaLabel} (not a made-up or hallucinated title)
+2. Correct the "reason" field with accurate metadata (actual genres, tone, year) if it contains errors
+
+Set isTrash: true for entries that:
+- Do not exist as a real ${mediaLabel}
+- Are clearly hallucinated or fictional titles
+- Are so incorrect that no real match can be identified
+
+Set isTrash: false for real, verifiable ${mediaLabel}s, even if minor metadata corrections are needed.
+Return every candidate — do not drop any entries from the output.`,
+    input: `Validate these ${mediaLabel} candidates:\n${list}`,
+  }));
+
+  return (response.output_parsed as ValidatorOutput) ?? { candidates: [] };
+}
+
+export async function runValidator(
+  candidates: RetrievalCandidate[],
+  mediaType: MediaType = 'tv_show',
+): Promise<ValidatorOutput> {
+  const mediaLabel = mediaType === 'movie' ? 'movie' : 'TV show';
+  const chunks = splitIntoChunks(candidates, CHUNK_SIZE);
+
+  const chunkResults = await Promise.all(
+    chunks.map((chunk) => runValidatorChunk(chunk, mediaLabel))
+  );
+
+  return {
+    candidates: chunkResults.flatMap((r) => r.candidates),
+  };
+}

packages/backend/src/db/schema.ts

@@ -1,14 +1,21 @@
-import { pgTable, uuid, text, jsonb, timestamp, integer, uniqueIndex } from 'drizzle-orm/pg-core';
+import { pgTable, uuid, text, jsonb, timestamp, integer, uniqueIndex, boolean } from 'drizzle-orm/pg-core';
 import type { CuratorOutput } from '../types/agents.js';
 
 export const recommendations = pgTable('recommendations', {
   id: uuid('id').defaultRandom().primaryKey(),
   title: text('title').notNull(),
   main_prompt: text('main_prompt').notNull(),
-  liked_shows: text('liked_shows').notNull().default(''),
+  liked_series: text('liked_series').notNull().default(''),
-  disliked_shows: text('disliked_shows').notNull().default(''),
+  disliked_series: text('disliked_series').notNull().default(''),
   themes: text('themes').notNull().default(''),
   brainstorm_count: integer('brainstorm_count').notNull().default(100),
+  media_type: text('media_type').notNull().default('tv_show'),
+  use_web_search: boolean('use_web_search').notNull().default(false),
+  use_validator: boolean('use_validator').notNull().default(false),
+  hard_requirements: boolean('hard_requirements').notNull().default(false),
+  self_expansive: boolean('self_expansive').notNull().default(false),
+  expansive_passes: integer('expansive_passes').notNull().default(1),
+  expansive_mode: text('expansive_mode').notNull().default('soft'),
   recommendations: jsonb('recommendations').$type<CuratorOutput[]>(),
   status: text('status').notNull().default('pending'),
   created_at: timestamp('created_at').defaultNow().notNull(),
@@ -18,10 +25,10 @@ export const feedback = pgTable(
   'feedback',
   {
     id: uuid('id').defaultRandom().primaryKey(),
-    tv_show_name: text('tv_show_name').notNull(),
+    item_name: text('item_name').notNull(),
     stars: integer('stars').notNull(),
     feedback: text('feedback').notNull().default(''),
     created_at: timestamp('created_at').defaultNow().notNull(),
   },
-  (table) => [uniqueIndex('feedback_tv_show_name_idx').on(table.tv_show_name)],
+  (table) => [uniqueIndex('feedback_item_name_idx').on(table.item_name)],
 );

packages/backend/src/migrate.ts

@@ -0,0 +1,43 @@
+import { drizzle } from 'drizzle-orm/postgres-js';
+import { migrate } from 'drizzle-orm/postgres-js/migrator';
+import postgres from 'postgres';
+import * as dotenv from 'dotenv';
+import path from 'path';
+import { fileURLToPath } from 'url';
+import fs from 'fs/promises';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+dotenv.config({ path: ['.env.local', '.env'] });
+
+const connectionString = process.env.DATABASE_URL;
+
+if (!connectionString) {
+  console.error('DATABASE_URL is not set');
+  process.exit(1);
+}
+
+// Using max: 1 connection since it's only for migration
+const migrationClient = postgres(connectionString, { max: 1 });
+const db = drizzle(migrationClient);
+
+const runMigrations = async () => {
+  console.log('Running database migrations...');
+  try {
+    const folder = path.join(__dirname, '../drizzle');
+    // print all migrations
+    const migrations = await fs.readdir(folder);
+    console.log('Migrations:', JSON.stringify(migrations));
+
+    await migrate(db, { migrationsFolder: folder });
+    console.log('Migrations completed successfully.');
+  } catch (err) {
+    console.error('Error running migrations:', err);
+    process.exit(1);
+  } finally {
+    await migrationClient.end();
+  }
+};
+
+runMigrations();

packages/backend/src/pipelineStreams.ts

@@ -0,0 +1,93 @@
+import type { SSEEvent } from './types/agents.js';
+
+type Listener = (event: SSEEvent) => void;
+
+interface PipelineSession {
+  history: SSEEvent[];
+  listeners: Set<Listener>;
+  terminal: boolean;
+  cleanupTimer: NodeJS.Timeout | null;
+}
+
+const sessions = new Map<string, PipelineSession>();
+const CLEANUP_DELAY_MS = 10 * 60 * 1000;
+
+function getOrCreateSession(recId: string): PipelineSession {
+  const existing = sessions.get(recId);
+  if (existing) {
+    if (existing.cleanupTimer) {
+      clearTimeout(existing.cleanupTimer);
+      existing.cleanupTimer = null;
+    }
+    return existing;
+  }
+
+  const session: PipelineSession = {
+    history: [],
+    listeners: new Set(),
+    terminal: false,
+    cleanupTimer: null,
+  };
+  sessions.set(recId, session);
+  return session;
+}
+
+function scheduleCleanup(recId: string, session: PipelineSession): void {
+  if (session.cleanupTimer) {
+    clearTimeout(session.cleanupTimer);
+  }
+
+  session.cleanupTimer = setTimeout(() => {
+    const current = sessions.get(recId);
+    if (current === session && current.listeners.size === 0) {
+      sessions.delete(recId);
+    }
+  }, CLEANUP_DELAY_MS);
+}
+
+export function resetPipelineSession(recId: string): void {
+  const existing = sessions.get(recId);
+  if (existing?.cleanupTimer) {
+    clearTimeout(existing.cleanupTimer);
+  }
+  sessions.delete(recId);
+}
+
+export function publishPipelineEvent(recId: string, event: SSEEvent): void {
+  const session = getOrCreateSession(recId);
+  session.history.push(event);
+
+  for (const listener of session.listeners) {
+    listener(event);
+  }
+
+  if (event.stage === 'complete' || event.status === 'error') {
+    session.terminal = true;
+    if (session.listeners.size === 0) {
+      scheduleCleanup(recId, session);
+    }
+  }
+}
+
+export function getPipelineHistory(recId: string): SSEEvent[] {
+  return sessions.get(recId)?.history ?? [];
+}
+
+export function hasPipelineSession(recId: string): boolean {
+  return sessions.has(recId);
+}
+
+export function subscribeToPipelineEvents(recId: string, listener: Listener): () => void {
+  const session = getOrCreateSession(recId);
+  session.listeners.add(listener);
+
+  return () => {
+    const current = sessions.get(recId);
+    if (!current) return;
+
+    current.listeners.delete(listener);
+    if (current.terminal && current.listeners.size === 0) {
+      scheduleCleanup(recId, current);
+    }
+  };
+}

packages/backend/src/pipelines/continuous.ts

@@ -0,0 +1,300 @@
+import { eq } from 'drizzle-orm';
+import { db } from '../db.js';
+import { recommendations } from '../db/schema.js';
+import { runInterpreter } from '../agents/interpreter.js';
+import { runRetrievalBatch } from '../agents/retrievalContinuous.js';
+import { runValidator } from '../agents/validator.js';
+import { runRanking } from '../agents/ranking.js';
+import { runCurator } from '../agents/curator.js';
+import type { CuratorOutput, InterpreterOutput, MediaType, RankingOutput, RetrievalCandidate, SSEEvent } from '../types/agents.js';
+import { generateTitle } from '../agents/titleGenerator.js';
+
+function log(msg: string, data?: unknown) {
+  const ts = new Date().toISOString();
+  if (data !== undefined) {
+    console.log(`[continuous-pipeline] [${ts}] ${msg}`, data);
+  } else {
+    console.log(`[continuous-pipeline] [${ts}] ${msg}`);
+  }
+}
+
+function deduplicateCandidates(candidates: RetrievalCandidate[], seenTitles: Set<string>): RetrievalCandidate[] {
+  return candidates.filter((c) => {
+    const key = c.title.toLowerCase();
+    if (seenTitles.has(key)) return false;
+    seenTitles.add(key);
+    return true;
+  });
+}
+
+function splitIntoBuckets<T>(items: T[], n: number): T[][] {
+  const size = Math.ceil(items.length / n);
+  return Array.from({ length: n }, (_, i) => items.slice(i * size, (i + 1) * size))
+    .filter((b) => b.length > 0);
+}
+
+function mergeCuratorOutputs(a: CuratorOutput[], b: CuratorOutput[]): CuratorOutput[] {
+  const seen = new Set(a.map((x) => x.title.toLowerCase()));
+  return [...a, ...b.filter((x) => !seen.has(x.title.toLowerCase()))];
+}
+
+interface ContinuousPipelineInput {
+  likedSeries: string;
+  dislikedSeries?: string;
+  themes?: string;
+  requirements?: string;
+  avoid?: string;
+  totalCount: number;
+  useWebSearch: boolean;
+  validateResults: boolean;
+  mediaType: MediaType;
+}
+
+export async function runContinuousPipeline(
+  recId: string,
+  input: ContinuousPipelineInput,
+  sseWrite: (event: SSEEvent) => void,
+): Promise<CuratorOutput[]> {
+  const startTime = Date.now();
+  const {
+    likedSeries,
+    dislikedSeries = '',
+    themes = '',
+    requirements = '',
+    avoid = '',
+    totalCount,
+    useWebSearch,
+    validateResults,
+    mediaType,
+  } = input;
+
+  const totalBatches = Math.ceil(totalCount / 10);
+  const allSeenTitles = new Set<string>();
+  let previousResponseId: string | null = null;
+  let interpreterOutput: InterpreterOutput | null = null;
+  const accumulatedCandidates: RetrievalCandidate[] = [];
+
+  log(`Starting continuous pipeline: ${totalCount} titles (${totalBatches} batches), mediaType=${mediaType}, webSearch=${useWebSearch}, validate=${validateResults}`);
+
+  try {
+    // --- Interpreter (runs once) ---
+    log('Interpreter: start');
+    sseWrite({ stage: 'interpreter', status: 'start' });
+    const t0 = Date.now();
+
+    interpreterOutput = await runInterpreter({
+      main_prompt: themes || 'recommend series based on user preferences',
+      liked_series: likedSeries,
+      disliked_series: dislikedSeries,
+      themes: themes,
+      media_type: mediaType,
+    });
+
+    log(`Interpreter: done (${Date.now() - t0}ms)`, {
+      liked: interpreterOutput.liked,
+      themes: interpreterOutput.themes,
+    });
+    sseWrite({ stage: 'interpreter', status: 'done', data: interpreterOutput });
+
+    // --- Retrieval (batched, chained) ---
+    log(`Retrieval: start (${totalBatches} batches)`);
+    sseWrite({ stage: 'retrieval', status: 'start' });
+
+    const hardRequirements = requirements.length > 0;
+
+    for (let batchIndex = 0; batchIndex < totalBatches; batchIndex++) {
+      log(`Retrieval batch ${batchIndex + 1}/${totalBatches}: start`);
+      sseWrite({
+        stage: 'retrieval',
+        status: 'start',
+        data: {
+          batch: batchIndex + 1,
+          totalBatches,
+          totalCandidates: accumulatedCandidates.length,
+        },
+      });
+
+      const seenTitlesArray = Array.from(allSeenTitles);
+      const result = await runRetrievalBatch(
+        interpreterOutput,
+        mediaType,
+        useWebSearch,
+        hardRequirements,
+        seenTitlesArray,
+        previousResponseId
+      );
+
+      previousResponseId = result.responseId;
+
+      // Deduplicate against previously seen titles
+      const deduped = deduplicateCandidates(result.candidates, allSeenTitles);
+      accumulatedCandidates.push(...deduped);
+
+      log(`Retrieval batch ${batchIndex + 1}/${totalBatches}: done, ${deduped.length} new candidates (total: ${accumulatedCandidates.length})`);
+    }
+
+    log(`Retrieval: complete, ${accumulatedCandidates.length} total candidates`);
+    sseWrite({
+      stage: 'retrieval',
+      status: 'done',
+      data: {
+        totalBatches,
+        totalCandidates: accumulatedCandidates.length,
+      },
+    });
+
+    // --- Validator (optional, per batch already done during retrieval) ---
+    // Note: In continuous mode, we could run validator after each batch
+    // For now, we'll run it once after all batches if validateResults is true
+    let candidatesForRanking = accumulatedCandidates;
+
+    if (validateResults && candidatesForRanking.length > 0) {
+      log(`Validator: start (${candidatesForRanking.length} candidates)`);
+      sseWrite({ stage: 'validator', status: 'start' });
+
+      const tV = Date.now();
+      const validatorOutput = await runValidator(candidatesForRanking, mediaType);
+      const verified = validatorOutput.candidates.filter((c) => !c.isTrash);
+      const trashCount = validatorOutput.candidates.length - verified.length;
+
+      candidatesForRanking = verified.map(({ title, reason }) => ({ title, reason }));
+      allSeenTitles.clear();
+      candidatesForRanking.forEach((c) => allSeenTitles.add(c.title.toLowerCase()));
+
+      log(`Validator: done (${Date.now() - tV}ms) — removed ${trashCount} trash entries`);
+      sseWrite({ stage: 'validator', status: 'done', data: { removed: trashCount, remaining: candidatesForRanking.length } });
+    } else {
+      sseWrite({ stage: 'validator', status: 'done', data: { skipped: !validateResults } });
+    }
+
+    // --- Ranking (bucketed) ---
+    log(`Ranking: start (${candidatesForRanking.length} candidates)`);
+    sseWrite({ stage: 'ranking', status: 'start' });
+    const t2 = Date.now();
+
+    const rankBucketCount = Math.ceil(candidatesForRanking.length / 15) || 1;
+    const candidateBuckets = splitIntoBuckets(candidatesForRanking, rankBucketCount);
+
+    const rankingBuckets = await Promise.all(
+      candidateBuckets.map((bucket) =>
+        runRanking(interpreterOutput!, { candidates: bucket }, mediaType, hardRequirements)
+      )
+    );
+
+    const dedupTitles = (titles: string[]) => [...new Map(titles.map((t) => [t.toLowerCase(), t])).values()];
+    const rankingOutput: RankingOutput = {
+      full_match: dedupTitles(rankingBuckets.flatMap((r) => r.full_match)),
+      definitely_like: dedupTitles(rankingBuckets.flatMap((r) => r.definitely_like)),
+      might_like: dedupTitles(rankingBuckets.flatMap((r) => r.might_like)),
+      questionable: dedupTitles(rankingBuckets.flatMap((r) => r.questionable)),
+      will_not_like: dedupTitles(rankingBuckets.flatMap((r) => r.will_not_like)),
+    };
+
+    log(`Ranking: done (${Date.now() - t2}ms)`, {
+      full_match: rankingOutput.full_match.length,
+      definitely_like: rankingOutput.definitely_like.length,
+      might_like: rankingOutput.might_like.length,
+    });
+    sseWrite({ stage: 'ranking', status: 'done', data: rankingOutput });
+
+    // print all ranked titles for debuggings
+    log('Ranked titles:', {
+      full_match: rankingOutput.full_match,
+      definitely_like: rankingOutput.definitely_like,
+      might_like: rankingOutput.might_like,
+      questionable: rankingOutput.questionable,
+      will_not_like: rankingOutput.will_not_like,
+    });
+
+    // --- Curator (bucketed) ---
+    log(`Curator: start`);
+    sseWrite({ stage: 'curator', status: 'start' });
+    const t3 = Date.now();
+
+    type CategorizedItem = { title: string; category: keyof RankingOutput };
+    const categorizedItems: CategorizedItem[] = [
+      ...rankingOutput.full_match.map((t) => ({ title: t, category: 'full_match' as const })),
+      ...rankingOutput.definitely_like.map((t) => ({ title: t, category: 'definitely_like' as const })),
+      ...rankingOutput.might_like.map((t) => ({ title: t, category: 'might_like' as const })),
+      ...rankingOutput.questionable.map((t) => ({ title: t, category: 'questionable' as const })),
+      ...rankingOutput.will_not_like.map((t) => ({ title: t, category: 'will_not_like' as const })),
+    ];
+
+    const curatorBucketCount = Math.ceil(categorizedItems.length / 15) || 1;
+    const curatorItemBuckets = splitIntoBuckets(categorizedItems, curatorBucketCount);
+    const curatorBucketRankings: RankingOutput[] = curatorItemBuckets.map((bucket) => ({
+      full_match: bucket.filter((i) => i.category === 'full_match').map((i) => i.title),
+      definitely_like: bucket.filter((i) => i.category === 'definitely_like').map((i) => i.title),
+      might_like: bucket.filter((i) => i.category === 'might_like').map((i) => i.title),
+      questionable: bucket.filter((i) => i.category === 'questionable').map((i) => i.title),
+      will_not_like: bucket.filter((i) => i.category === 'will_not_like').map((i) => i.title),
+    }));
+
+    const curatorBucketOutputs = await Promise.all(
+      curatorBucketRankings.map((ranking) =>
+        runCurator(ranking, interpreterOutput!, mediaType, useWebSearch)
+      )
+    );
+
+    const curatorOutput = curatorBucketOutputs.reduce((acc, bucket) => mergeCuratorOutputs(acc, bucket), [] as CuratorOutput[]);
+    log(`Curator: done (${Date.now() - t3}ms) — ${curatorOutput.length} items curated`);
+    sseWrite({ stage: 'curator', status: 'done', data: curatorOutput });
+
+    // Generate AI title
+    let aiTitle = 'Continuous Recommendations';
+    try {
+      log('Title generation: start');
+      aiTitle = await generateTitle(interpreterOutput, mediaType);
+      log(`Title generation: done — "${aiTitle}"`);
+    } catch (err) {
+      log(`Title generation failed: ${String(err)}`);
+    }
+
+    // Sort by category order
+    const CATEGORY_ORDER: Record<string, number> = {
+      'Full Match': 0,
+      'Definitely Like': 1,
+      'Might Like': 2,
+      'Questionable': 3,
+      'Will Not Like': 4,
+    };
+    curatorOutput.sort((a, b) => (CATEGORY_ORDER[a.category] ?? 99) - (CATEGORY_ORDER[b.category] ?? 99));
+
+    // Save to database (update existing record)
+    log('Saving results to DB');
+    await db
+      .update(recommendations)
+      .set({
+        title: aiTitle,
+        main_prompt: themes || 'Continuous recommendations',
+        liked_series: likedSeries,
+        disliked_series: dislikedSeries,
+        themes: themes,
+        brainstorm_count: totalCount,
+        media_type: mediaType,
+        use_web_search: useWebSearch,
+        use_validator: validateResults,
+        status: 'done',
+        recommendations: curatorOutput,
+      })
+      .where(eq(recommendations.id, recId));
+
+    sseWrite({ stage: 'complete', status: 'done', data: { id: recId, title: aiTitle, batchCount: totalBatches } });
+
+    log(`Pipeline complete (total: ${Date.now() - startTime}ms), saved as ${recId}`);
+    return curatorOutput;
+
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    log(`Pipeline error: ${message}`);
+
+    // Update status to error
+    await db
+      .update(recommendations)
+      .set({ status: 'error' })
+      .where(eq(recommendations.id, recId));
+
+    sseWrite({ stage: 'curator', status: 'error', data: { message } });
+    return [];
+  }
+}

packages/backend/src/pipelines/recommendation.ts

@@ -3,20 +3,47 @@ import { db } from '../db.js';
 import { recommendations } from '../db/schema.js';
 import { runInterpreter } from '../agents/interpreter.js';
 import { runRetrieval } from '../agents/retrieval.js';
+import { runValidator } from '../agents/validator.js';
 import { runRanking } from '../agents/ranking.js';
 import { runCurator } from '../agents/curator.js';
-import type { CuratorOutput, SSEEvent } from '../types/agents.js';
+import type { CuratorOutput, InterpreterOutput, MediaType, RankingOutput, RetrievalCandidate, SSEEvent } from '../types/agents.js';
 import { generateTitle } from '../agents/titleGenerator.js';
 
 /* -- Agent pipeline --
 [1] Interpreter -> gets user input, transforms into structured data
-[2] Retrieval -> gets shows from OpenAI (high temperature)
+[2] Retrieval -> gets candidates from OpenAI (high temperature)
-[3] Ranking -> ranks shows based on user input
+[2.5] Validator (optional) -> verifies candidates exist, removes trash
-[4] Curator -> curates shows based on user input
+[3] Ranking -> ranks candidates based on user input
+[4] Curator -> curates candidates based on user input
 */
 
 type RecommendationRecord = typeof recommendations.$inferSelect;
 
+function getBucketCount(count: number): number {
+  return Math.ceil(count / 15);
+}
+
+function deduplicateCandidates(candidates: RetrievalCandidate[], seenTitles?: Set<string>): RetrievalCandidate[] {
+  const seen = seenTitles ?? new Set<string>();
+  return candidates.filter((c) => {
+    const key = c.title.toLowerCase();
+    if (seen.has(key)) return false;
+    seen.add(key);
+    return true;
+  });
+}
+
+function splitIntoBuckets<T>(items: T[], n: number): T[][] {
+  const size = Math.ceil(items.length / n);
+  return Array.from({ length: n }, (_, i) => items.slice(i * size, (i + 1) * size))
+    .filter((b) => b.length > 0);
+}
+
+function mergeCuratorOutputs(a: CuratorOutput[], b: CuratorOutput[]): CuratorOutput[] {
+  const seen = new Set(a.map((x) => x.title.toLowerCase()));
+  return [...a, ...b.filter((x) => !seen.has(x.title.toLowerCase()))];
+}
+
 function log(recId: string, msg: string, data?: unknown) {
   const ts = new Date().toISOString();
   if (data !== undefined) {
@@ -26,6 +53,124 @@ function log(recId: string, msg: string, data?: unknown) {
   }
 }
 
+interface SubPipelineCtx {
+  recId: string;
+  interpreterOutput: InterpreterOutput;
+  mediaType: MediaType;
+  useWebSearch: boolean;
+  useValidator: boolean;
+  useHardRequirements: boolean;
+  brainstormCount: number;
+  previousFullMatches: string[];
+  allSeenTitles: Set<string>;
+  stagePrefix: string;
+  sseWrite: (event: SSEEvent) => void;
+}
+
+async function runSubPipeline(ctx: SubPipelineCtx): Promise<CuratorOutput[]> {
+  const {
+    recId, interpreterOutput, mediaType, useWebSearch, useValidator,
+    useHardRequirements, brainstormCount, previousFullMatches,
+    allSeenTitles, stagePrefix, sseWrite,
+  } = ctx;
+
+  const p = (stage: string) => (stagePrefix + stage) as SSEEvent['stage'];
+
+  // --- Retrieval (bucketed) ---
+  log(recId, `${stagePrefix}Retrieval: start`);
+  sseWrite({ stage: p('retrieval'), status: 'start' });
+  const t1 = Date.now();
+  const retrievalBucketCount = getBucketCount(brainstormCount);
+  const perBucketCount = Math.ceil(brainstormCount / retrievalBucketCount);
+  const retrievalBuckets = await Promise.all(
+    Array.from({ length: retrievalBucketCount }, () =>
+      runRetrieval(interpreterOutput, perBucketCount, mediaType, useWebSearch, useHardRequirements, previousFullMatches)
+    )
+  );
+  const allCandidates = retrievalBuckets.flatMap((r) => r.candidates);
+  const dedupedCandidates = deduplicateCandidates(allCandidates, allSeenTitles);
+  log(recId, `${stagePrefix}Retrieval: done (${Date.now() - t1}ms) — ${dedupedCandidates.length} candidates (${retrievalBucketCount} buckets, ${allCandidates.length} before dedup)`, {
+    titles: dedupedCandidates.map((c) => c.title),
+  });
+  sseWrite({ stage: p('retrieval'), status: 'done', data: { candidates: dedupedCandidates } });
+
+  // --- Validator (optional) ---
+  let candidatesForRanking = dedupedCandidates;
+  if (useValidator) {
+    log(recId, `${stagePrefix}Validator: start`);
+    sseWrite({ stage: p('validator'), status: 'start' });
+    const tV = Date.now();
+    const validatorOutput = await runValidator(dedupedCandidates, mediaType);
+    const verified = validatorOutput.candidates.filter((c) => !c.isTrash);
+    const trashCount = validatorOutput.candidates.length - verified.length;
+    candidatesForRanking = verified.map(({ title, reason }) => ({ title, reason }));
+    log(recId, `${stagePrefix}Validator: done (${Date.now() - tV}ms) — removed ${trashCount} trash entries`);
+    sseWrite({ stage: p('validator'), status: 'done', data: { removed: trashCount } });
+  } else {
+    sseWrite({ stage: p('validator'), status: 'done', data: { skipped: true } });
+  }
+
+  // --- Ranking (bucketed) ---
+  log(recId, `${stagePrefix}Ranking: start`);
+  sseWrite({ stage: p('ranking'), status: 'start' });
+  const t2 = Date.now();
+  const rankBucketCount = getBucketCount(candidatesForRanking.length);
+  const candidateBuckets = splitIntoBuckets(candidatesForRanking, rankBucketCount);
+  const rankingBuckets = await Promise.all(
+    candidateBuckets.map((bucket) =>
+      runRanking(interpreterOutput, { candidates: bucket }, mediaType, useHardRequirements)
+    )
+  );
+  const dedupTitles = (titles: string[]) => [...new Map(titles.map((t) => [t.toLowerCase(), t])).values()];
+  const rankingOutput: RankingOutput = {
+    full_match: dedupTitles(rankingBuckets.flatMap((r) => r.full_match)),
+    definitely_like: dedupTitles(rankingBuckets.flatMap((r) => r.definitely_like)),
+    might_like: dedupTitles(rankingBuckets.flatMap((r) => r.might_like)),
+    questionable: dedupTitles(rankingBuckets.flatMap((r) => r.questionable)),
+    will_not_like: dedupTitles(rankingBuckets.flatMap((r) => r.will_not_like)),
+  };
+  log(recId, `${stagePrefix}Ranking: done (${Date.now() - t2}ms) — ${rankBucketCount} buckets`, {
+    full_match: rankingOutput.full_match.length,
+    definitely_like: rankingOutput.definitely_like.length,
+    might_like: rankingOutput.might_like.length,
+    questionable: rankingOutput.questionable.length,
+    will_not_like: rankingOutput.will_not_like.length,
+  });
+  sseWrite({ stage: p('ranking'), status: 'done', data: rankingOutput });
+
+  // --- Curator (bucketed) ---
+  log(recId, `${stagePrefix}Curator: start`);
+  sseWrite({ stage: p('curator'), status: 'start' });
+  const t3 = Date.now();
+  type CategorizedItem = { title: string; category: keyof RankingOutput };
+  const categorizedItems: CategorizedItem[] = [
+    ...rankingOutput.full_match.map((t) => ({ title: t, category: 'full_match' as const })),
+    ...rankingOutput.definitely_like.map((t) => ({ title: t, category: 'definitely_like' as const })),
+    ...rankingOutput.might_like.map((t) => ({ title: t, category: 'might_like' as const })),
+    ...rankingOutput.questionable.map((t) => ({ title: t, category: 'questionable' as const })),
+    ...rankingOutput.will_not_like.map((t) => ({ title: t, category: 'will_not_like' as const })),
+  ];
+  const curatorBucketCount = getBucketCount(categorizedItems.length);
+  const curatorItemBuckets = splitIntoBuckets(categorizedItems, curatorBucketCount);
+  const curatorBucketRankings: RankingOutput[] = curatorItemBuckets.map((bucket) => ({
+    full_match: bucket.filter((i) => i.category === 'full_match').map((i) => i.title),
+    definitely_like: bucket.filter((i) => i.category === 'definitely_like').map((i) => i.title),
+    might_like: bucket.filter((i) => i.category === 'might_like').map((i) => i.title),
+    questionable: bucket.filter((i) => i.category === 'questionable').map((i) => i.title),
+    will_not_like: bucket.filter((i) => i.category === 'will_not_like').map((i) => i.title),
+  }));
+  const curatorBucketOutputs = await Promise.all(
+    curatorBucketRankings.map((ranking) =>
+      runCurator(ranking, interpreterOutput, mediaType, useWebSearch)
+    )
+  );
+  const curatorOutput = curatorBucketOutputs.reduce((acc, bucket) => mergeCuratorOutputs(acc, bucket), [] as CuratorOutput[]);
+  log(recId, `${stagePrefix}Curator: done (${Date.now() - t3}ms) — ${curatorOutput.length} items curated (${curatorBucketCount} buckets)`);
+  sseWrite({ stage: p('curator'), status: 'done', data: curatorOutput });
+
+  return curatorOutput;
+}
+
 export async function runPipeline(
   rec: RecommendationRecord,
   sseWrite: (event: SSEEvent) => void,
@@ -33,8 +178,13 @@ export async function runPipeline(
 ): Promise<CuratorOutput[]> {
   let currentStage: SSEEvent['stage'] = 'interpreter';
   const startTime = Date.now();
+  const mediaType = (rec.media_type ?? 'tv_show') as MediaType;
+  const useWebSearch = rec.use_web_search ?? false;
+  const useValidator = rec.use_validator ?? false;
+  const useHardRequirements = rec.hard_requirements ?? false;
+  const selfExpansive = rec.self_expansive ?? false;
 
-  log(rec.id, `Starting pipeline for "${rec.title}"${feedbackContext ? ' (with feedback context)' : ''}`);
+  log(rec.id, `Starting pipeline for "${rec.title}" [${mediaType}${useWebSearch ? ', web_search' : ''}${useValidator ? ', validator' : ''}${useHardRequirements ? ', hard_req' : ''}${selfExpansive ? `, expansive×${rec.expansive_passes}(${rec.expansive_mode})` : ''}]${feedbackContext ? ' (with feedback context)' : ''}`);
 
   try {
     // Set status to running
@@ -51,9 +201,10 @@ export async function runPipeline(
     const t0 = Date.now();
     const interpreterOutput = await runInterpreter({
       main_prompt: rec.main_prompt,
-      liked_shows: rec.liked_shows,
+      liked_series: rec.liked_series,
-      disliked_shows: rec.disliked_shows,
+      disliked_series: rec.disliked_series,
       themes: rec.themes,
+      media_type: mediaType,
       ...(feedbackContext !== undefined ? { feedback_context: feedbackContext } : {}),
     });
     log(rec.id, `Interpreter: done (${Date.now() - t0}ms)`, {
@@ -65,61 +216,101 @@ export async function runPipeline(
     });
     sseWrite({ stage: 'interpreter', status: 'done', data: interpreterOutput });
 
-    // --- Retrieval ---
+    // --- Pass 1: Retrieval → [Validator?] → Ranking → Curator ---
     currentStage = 'retrieval';
-    log(rec.id, 'Retrieval: start');
+    const allSeenTitles = new Set<string>();
-    sseWrite({ stage: 'retrieval', status: 'start' });
+    const pass1Output = await runSubPipeline({
-    const t1 = Date.now();
+      recId: rec.id,
-    const retrievalOutput = await runRetrieval(interpreterOutput, rec.brainstorm_count);
+      interpreterOutput,
-    log(rec.id, `Retrieval: done (${Date.now() - t1}ms) — ${retrievalOutput.candidates.length} candidates`, {
+      mediaType,
-      titles: retrievalOutput.candidates.map((c) => c.title),
+      useWebSearch,
+      useValidator,
+      useHardRequirements,
+      brainstormCount: rec.brainstorm_count,
+      previousFullMatches: [],
+      allSeenTitles,
+      stagePrefix: '',
+      sseWrite: (event) => {
+        currentStage = event.stage;
+        sseWrite(event);
+      },
     });
-    sseWrite({ stage: 'retrieval', status: 'done', data: retrievalOutput });
 
-    // --- Ranking ---
+    let mergedOutput = pass1Output;
-    currentStage = 'ranking';
+
-    log(rec.id, 'Ranking: start');
+    // --- Self Expansive: extra passes ---
-    sseWrite({ stage: 'ranking', status: 'start' });
+    if (selfExpansive && rec.expansive_passes > 0) {
-    const t2 = Date.now();
+      const allFullMatches = pass1Output
-    const rankingOutput = await runRanking(interpreterOutput, retrievalOutput);
+        .filter((c) => c.category === 'Full Match')
-    log(rec.id, `Ranking: done (${Date.now() - t2}ms)`, {
+        .map((c) => c.title);
-      definitely_like: rankingOutput.definitely_like.length,
+
-      might_like: rankingOutput.might_like.length,
+      for (let i = 0; i < rec.expansive_passes; i++) {
-      questionable: rankingOutput.questionable.length,
+        const passNum = i + 2;
-      will_not_like: rankingOutput.will_not_like.length,
+        const passCount = rec.expansive_mode === 'extreme' ? rec.brainstorm_count : 60;
+        const passPrefix = `pass${passNum}:` as const;
+
+        log(rec.id, `Self Expansive Pass ${passNum}: start (${passCount} candidates, ${allFullMatches.length} full matches as context)`);
+        currentStage = `${passPrefix}retrieval` as SSEEvent['stage'];
+
+        const passOutput = await runSubPipeline({
+          recId: rec.id,
+          interpreterOutput,
+          mediaType,
+          useWebSearch,
+          useValidator,
+          useHardRequirements,
+          brainstormCount: passCount,
+          previousFullMatches: [...allFullMatches],
+          allSeenTitles,
+          stagePrefix: passPrefix,
+          sseWrite: (event) => {
+            currentStage = event.stage;
+            sseWrite(event);
+          },
         });
-    sseWrite({ stage: 'ranking', status: 'done', data: rankingOutput });
 
-    // --- Curator ---
+        mergedOutput = mergeCuratorOutputs(mergedOutput, passOutput);
-    currentStage = 'curator';
+
-    log(rec.id, 'Curator: start');
+        const newFullMatches = passOutput
-    sseWrite({ stage: 'curator', status: 'start' });
+          .filter((c) => c.category === 'Full Match')
-    const t3 = Date.now();
+          .map((c) => c.title);
-    const curatorOutput = await runCurator(rankingOutput, interpreterOutput);
+        allFullMatches.push(...newFullMatches);
-    log(rec.id, `Curator: done (${Date.now() - t3}ms) — ${curatorOutput.length} shows curated`);
+
-    sseWrite({ stage: 'curator', status: 'done', data: curatorOutput });
+        log(rec.id, `Self Expansive Pass ${passNum}: done — ${passOutput.length} new items, ${mergedOutput.length} total`);
+      }
+    }
 
     // Generate AI title
     let aiTitle: string = rec.title;
     try {
       log(rec.id, 'Title generation: start');
-      aiTitle = await generateTitle(interpreterOutput);
+      aiTitle = await generateTitle(interpreterOutput, mediaType);
       log(rec.id, `Title generation: done — "${aiTitle}"`);
     } catch (err) {
       log(rec.id, `Title generation failed, keeping initial title: ${String(err)}`);
     }
 
+    // Sort by category order before saving
+    const CATEGORY_ORDER: Record<string, number> = {
+      'Full Match': 0,
+      'Definitely Like': 1,
+      'Might Like': 2,
+      'Questionable': 3,
+      'Will Not Like': 4,
+    };
+    mergedOutput.sort((a, b) => (CATEGORY_ORDER[a.category] ?? 99) - (CATEGORY_ORDER[b.category] ?? 99));
+
     // Save results to DB
     log(rec.id, 'Saving results to DB');
     await db
       .update(recommendations)
-      .set({ recommendations: curatorOutput, status: 'done', title: aiTitle })
+      .set({ recommendations: mergedOutput, status: 'done', title: aiTitle })
       .where(eq(recommendations.id, rec.id));
 
-    sseWrite({ stage: 'complete', status: 'done' });
+    sseWrite({ stage: 'complete', status: 'done', data: { title: aiTitle } });
 
     log(rec.id, `Pipeline complete (total: ${Date.now() - startTime}ms)`);
-    return curatorOutput;
+    return mergedOutput;
   } catch (err) {
     const message = err instanceof Error ? err.message : String(err);
     log(rec.id, `Pipeline error at stage "${currentStage}": ${message}`);

packages/backend/src/routes/feedback.ts

@@ -1,13 +1,12 @@
 import type { FastifyInstance } from 'fastify';
-import { eq } from 'drizzle-orm';
 import { db } from '../db.js';
 import { feedback } from '../db/schema.js';
 
 export default async function feedbackRoute(fastify: FastifyInstance) {
-  // POST /feedback — upsert by tv_show_name
+  // POST /feedback — upsert by item_name
   fastify.post('/feedback', async (request, reply) => {
     const body = request.body as {
-      tv_show_name: string;
+      item_name: string;
       stars: number;
       feedback?: string;
     };
@@ -15,12 +14,12 @@ export default async function feedbackRoute(fastify: FastifyInstance) {
     await db
       .insert(feedback)
       .values({
-        tv_show_name: body.tv_show_name,
+        item_name: body.item_name,
         stars: body.stars,
         feedback: body.feedback ?? '',
       })
       .onConflictDoUpdate({
-        target: feedback.tv_show_name,
+        target: feedback.item_name,
         set: {
           stars: body.stars,
           feedback: body.feedback ?? '',

packages/backend/src/routes/recommendations.ts

@@ -3,17 +3,43 @@ import { eq, desc } from 'drizzle-orm';
 import { db } from '../db.js';
 import { recommendations, feedback } from '../db/schema.js';
 import { runPipeline } from '../pipelines/recommendation.js';
-import type { SSEEvent } from '../types/agents.js';
+import { runContinuousPipeline } from '../pipelines/continuous.js';
+import type { MediaType, SSEEvent } from '../types/agents.js';
+import { supportsWebSearch } from '../agent.js';
+import {
+  getPipelineHistory,
+  hasPipelineSession,
+  publishPipelineEvent,
+  resetPipelineSession,
+  subscribeToPipelineEvents,
+} from '../pipelineStreams.js';
+
+const UUID_RE = /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+
+function isUuid(id: string): boolean {
+  return UUID_RE.test(id);
+}
+
+function writeSSE(raw: typeof import('node:http').ServerResponse.prototype, event: SSEEvent): void {
+  raw.write(`data: ${JSON.stringify(event)}\n\n`);
+}
 
 export default async function recommendationsRoute(fastify: FastifyInstance) {
   // POST /recommendations — create record, return { id }
   fastify.post('/recommendations', async (request, reply) => {
     const body = request.body as {
       main_prompt: string;
-      liked_shows?: string;
+      liked_series?: string;
-      disliked_shows?: string;
+      disliked_series?: string;
       themes?: string;
       brainstorm_count?: number;
+      media_type?: string;
+      use_web_search?: boolean;
+      use_validator?: boolean;
+      hard_requirements?: boolean;
+      self_expansive?: boolean;
+      expansive_passes?: number;
+      expansive_mode?: string;
     };
 
     const title = (body.main_prompt ?? '')
@@ -24,16 +50,31 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
 
     const rawCount = Number(body.brainstorm_count ?? 100);
     const brainstorm_count = Number.isFinite(rawCount) ? Math.min(200, Math.max(50, rawCount)) : 100;
+    const media_type: MediaType = body.media_type === 'movie' ? 'movie' : 'tv_show';
+    const use_web_search = body.use_web_search === true;
+    const use_validator = body.use_validator === true && supportsWebSearch;
+    const hard_requirements = body.hard_requirements === true;
+    const self_expansive = body.self_expansive === true;
+    const rawPasses = Number(body.expansive_passes ?? 2);
+    const expansive_passes = Number.isFinite(rawPasses) ? Math.min(5, Math.max(1, rawPasses)) : 2;
+    const expansive_mode = body.expansive_mode === 'extreme' ? 'extreme' : 'soft';
 
     const [rec] = await db
       .insert(recommendations)
       .values({
         title: title || 'Untitled',
         main_prompt: body.main_prompt ?? '',
-        liked_shows: body.liked_shows ?? '',
+        liked_series: body.liked_series ?? '',
-        disliked_shows: body.disliked_shows ?? '',
+        disliked_series: body.disliked_series ?? '',
         themes: body.themes ?? '',
         brainstorm_count,
+        media_type,
+        use_web_search,
+        use_validator,
+        hard_requirements,
+        self_expansive,
+        expansive_passes,
+        expansive_mode,
         status: 'pending',
       })
       .returning({ id: recommendations.id });
@@ -41,6 +82,94 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
     return reply.code(201).send({ id: rec?.id });
   });
 
+  // POST /recommendations/continuous — create record and run continuous pipeline with SSE
+  fastify.post('/recommendations/continuous', async (request, reply) => {
+    const body = request.body as {
+      liked_series: string;
+      disliked_series?: string;
+      themes?: string;
+      requirements?: string;
+      avoid?: string;
+      total_count: number;
+      media_type?: string;
+      use_web_search?: boolean;
+      validate_results?: boolean;
+    };
+
+    const mediaType: MediaType = body.media_type === 'movie' ? 'movie' : 'tv_show';
+    const useWebSearch = body.use_web_search === true;
+    const validateResults = body.validate_results === true && supportsWebSearch;
+    const totalCount = Number.isFinite(body.total_count) ? Math.min(100, Math.max(10, body.total_count)) : 30;
+
+    // Create record first with pending status
+    const title = body.themes?.trim().split(/\s+/).slice(0, 5).join(' ') || 'Continuous Recommendations';
+    const [rec] = await db
+      .insert(recommendations)
+      .values({
+        title,
+        main_prompt: body.themes ?? 'Continuous recommendations',
+        liked_series: body.liked_series,
+        disliked_series: body.disliked_series ?? '',
+        themes: body.themes ?? '',
+        brainstorm_count: totalCount,
+        media_type: mediaType,
+        use_web_search: useWebSearch,
+        use_validator: validateResults,
+        status: 'pending',
+      })
+      .returning({ id: recommendations.id });
+
+    if (!rec) {
+      return reply.code(500).send({ error: 'Failed to create recommendation record' });
+    }
+
+    const recId = rec.id;
+    resetPipelineSession(recId);
+
+    // Set SSE headers and hijack
+    reply.raw.setHeader('Content-Type', 'text/event-stream');
+    reply.raw.setHeader('Cache-Control', 'no-cache');
+    reply.raw.setHeader('Connection', 'keep-alive');
+    reply.raw.setHeader('Access-Control-Allow-Origin', '*');
+    reply.raw.flushHeaders();
+    reply.hijack();
+
+    // Immediately send the record ID so frontend can navigate to it
+    reply.raw.write(`data: ${JSON.stringify({ type: 'created', id: recId })}\n\n`);
+
+    const sseWrite = (event: SSEEvent) => {
+      publishPipelineEvent(recId, event);
+      try {
+        writeSSE(reply.raw, event);
+      } catch {
+        // Client disconnected
+      }
+    };
+
+    try {
+      // Update status to running
+      await db
+        .update(recommendations)
+        .set({ status: 'running' })
+        .where(eq(recommendations.id, recId));
+
+      // Run the continuous pipeline (it will now update the existing record)
+      await runContinuousPipeline(recId, {
+        likedSeries: body.liked_series,
+        dislikedSeries: body.disliked_series ?? '',
+        themes: body.themes ?? '',
+        requirements: body.requirements ?? '',
+        avoid: body.avoid ?? '',
+        totalCount,
+        useWebSearch,
+        validateResults,
+        mediaType,
+      }, sseWrite);
+    } finally {
+      reply.raw.end();
+    }
+  });
+
   // GET /recommendations — list all
   fastify.get('/recommendations', async (_request, reply) => {
     const rows = await db
@@ -48,6 +177,7 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
         id: recommendations.id,
         title: recommendations.title,
         status: recommendations.status,
+        media_type: recommendations.media_type,
         created_at: recommendations.created_at,
       })
       .from(recommendations)
@@ -58,6 +188,7 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
   // GET /recommendations/:id — full record
   fastify.get('/recommendations/:id', async (request, reply) => {
     const { id } = request.params as { id: string };
+    if (!isUuid(id)) return reply.code(404).send({ error: 'Not found' });
     const [rec] = await db
       .select()
       .from(recommendations)
@@ -68,9 +199,9 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
   });
 
   // GET /recommendations/:id/stream — SSE pipeline stream
-  // Always fetches all current feedback and injects if present (supports rerank flow)
   fastify.get('/recommendations/:id/stream', async (request, reply) => {
     const { id } = request.params as { id: string };
+    if (!isUuid(id)) return reply.code(404).send({ error: 'Not found' });
     const [rec] = await db
       .select()
       .from(recommendations)
@@ -78,19 +209,7 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
 
     if (!rec) return reply.code(404).send({ error: 'Not found' });
 
-    // Load all feedback to potentially inject as context
+    // Set SSE headers and hijack before any branching
-    const feedbackRows = await db.select().from(feedback);
-    const feedbackContext =
-      feedbackRows.length > 0
-        ? feedbackRows
-            .map(
-              (f) =>
-                `Show: "${f.tv_show_name}" — Rating: ${f.stars}/3 stars${f.feedback ? ` — Comment: ${f.feedback}` : ''}`,
-            )
-            .join('\n')
-        : undefined;
-
-    // Set SSE headers and hijack
     reply.raw.setHeader('Content-Type', 'text/event-stream');
     reply.raw.setHeader('Cache-Control', 'no-cache');
     reply.raw.setHeader('Connection', 'keep-alive');
@@ -98,20 +217,130 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
     reply.raw.flushHeaders();
     reply.hijack();
 
+    // Resilient write — swallows errors so a disconnected client never crashes
+    // an in-flight pipeline that is still running server-side.
     const sseWrite = (event: SSEEvent) => {
-      reply.raw.write(`data: ${JSON.stringify(event)}\n\n`);
+      try {
+        writeSSE(reply.raw, event);
+      } catch {
+        // Client disconnected — pipeline continues, writes are silently dropped
+      }
     };
 
     try {
-      await runPipeline(rec, sseWrite, feedbackContext);
+      // Already finished — send a synthetic completion event and close immediately.
+      if (rec.status === 'done') {
+        sseWrite({ stage: 'complete', status: 'done', data: { title: rec.title } });
+        return;
+      }
+
+      // Already errored — send a synthetic error event and close immediately.
+      if (rec.status === 'error') {
+        sseWrite({ stage: 'curator', status: 'error', data: { message: 'Pipeline failed' } });
+        return;
+      }
+
+      // Already running — the pipeline is executing on a previous connection.
+      // Poll the DB until it reaches a terminal state, then report the result.
+      // This prevents starting a duplicate pipeline run on page reload.
+      if (rec.status === 'running') {
+        if (hasPipelineSession(id)) {
+          const history = getPipelineHistory(id);
+          for (const event of history) {
+            sseWrite(event);
+          }
+
+          const lastEvent = history[history.length - 1];
+          if (lastEvent && (lastEvent.stage === 'complete' || lastEvent.status === 'error')) {
+            return;
+          }
+
+          await new Promise<void>((resolve) => {
+            const unsubscribe = subscribeToPipelineEvents(id, (event) => {
+              sseWrite(event);
+              if (event.stage === 'complete' || event.status === 'error') {
+                unsubscribe();
+                resolve();
+              }
+            });
+
+            request.raw.on('close', () => {
+              unsubscribe();
+              resolve();
+            });
+          });
+          return;
+        }
+
+        const POLL_INTERVAL_MS = 2000;
+        const TIMEOUT_MS = 20 * 60 * 1000; // 20 minutes hard ceiling
+        const start = Date.now();
+        while (Date.now() - start < TIMEOUT_MS) {
+          await new Promise((resolve) => setTimeout(resolve, POLL_INTERVAL_MS));
+          const [current] = await db
+            .select({ status: recommendations.status, title: recommendations.title })
+            .from(recommendations)
+            .where(eq(recommendations.id, id));
+          if (!current || current.status === 'done') {
+            sseWrite({ stage: 'complete', status: 'done', data: { title: current?.title ?? rec.title } });
+            return;
+          }
+          if (current.status === 'error') {
+            sseWrite({ stage: 'curator', status: 'error', data: { message: 'Pipeline failed' } });
+            return;
+          }
+          // Still running — keep polling
+        }
+        // Timed out waiting — report as error
+        sseWrite({ stage: 'curator', status: 'error', data: { message: 'Pipeline timed out' } });
+        return;
+      }
+
+      // status === 'pending' — start the pipeline normally.
+      resetPipelineSession(id);
+
+      // Load all feedback to potentially inject as context
+      const feedbackRows = await db.select().from(feedback);
+      const mediaLabel = rec.media_type === 'movie' ? 'Movie' : 'Show';
+      const feedbackContext =
+        feedbackRows.length > 0
+          ? feedbackRows
+            .map(
+              (f) =>
+                `${mediaLabel}: "${f.item_name}" — Rating: ${f.stars}/3 stars${f.feedback ? ` — Comment: ${f.feedback}` : ''}`,
+            )
+            .join('\n')
+          : undefined;
+
+      await runPipeline(rec, (event) => {
+        publishPipelineEvent(id, event);
+        sseWrite(event);
+      }, feedbackContext);
     } finally {
       reply.raw.end();
     }
   });
 
+  // DELETE /recommendations/:id — delete a recommendation
+  fastify.delete('/recommendations/:id', async (request, reply) => {
+    const { id } = request.params as { id: string };
+    if (!isUuid(id)) return reply.code(404).send({ error: 'Not found' });
+    const [rec] = await db
+      .select({ id: recommendations.id })
+      .from(recommendations)
+      .where(eq(recommendations.id, id));
+
+    if (!rec) return reply.code(404).send({ error: 'Not found' });
+
+    await db.delete(recommendations).where(eq(recommendations.id, id));
+
+    return reply.send({ ok: true });
+  });
+
   // POST /recommendations/:id/rerank — reset status so client can re-open SSE stream
   fastify.post('/recommendations/:id/rerank', async (request, reply) => {
     const { id } = request.params as { id: string };
+    if (!isUuid(id)) return reply.code(404).send({ error: 'Not found' });
     const [rec] = await db
       .select({ id: recommendations.id })
       .from(recommendations)
@@ -123,6 +352,7 @@ export default async function recommendationsRoute(fastify: FastifyInstance) {
       .update(recommendations)
       .set({ status: 'pending' })
       .where(eq(recommendations.id, id));
+    resetPipelineSession(id);
 
     return reply.send({ ok: true });
   });

packages/backend/src/types/agents.ts

@@ -1,3 +1,5 @@
+export type MediaType = 'tv_show' | 'movie';
+
 export interface InterpreterOutput {
   liked: string[];
   disliked: string[];
@@ -5,6 +7,7 @@ export interface InterpreterOutput {
   character_preferences: string[];
   tone: string[];
   avoid: string[];
+  requirements: string[];
 }
 
 export interface RetrievalCandidate {
@@ -16,22 +19,47 @@ export interface RetrievalOutput {
   candidates: RetrievalCandidate[];
 }
 
+export interface ValidatorCandidate {
+  title: string;
+  reason: string;
+  isTrash: boolean;
+}
+
+export interface ValidatorOutput {
+  candidates: ValidatorCandidate[];
+}
+
 export interface RankingOutput {
+  full_match: string[];
   definitely_like: string[];
   might_like: string[];
   questionable: string[];
   will_not_like: string[];
 }
 
-export type CuratorCategory = 'Definitely Like' | 'Might Like' | 'Questionable' | 'Will Not Like';
+export type CuratorCategory = 'Full Match' | 'Definitely Like' | 'Might Like' | 'Questionable' | 'Will Not Like';
 
 export interface CuratorOutput {
   title: string;
   explanation: string;
   category: CuratorCategory;
+  genre: string;
+  pros: string[];
+  cons: string[];
 }
 
-export type PipelineStage = 'interpreter' | 'retrieval' | 'ranking' | 'curator' | 'complete';
+export type PipelineStage =
+  | 'interpreter'
+  | 'retrieval'
+  | 'validator'
+  | 'ranking'
+  | 'curator'
+  | 'complete'
+  | `pass${number}:retrieval`
+  | `pass${number}:validator`
+  | `pass${number}:ranking`
+  | `pass${number}:curator`;
+
 export type SSEStatus = 'start' | 'done' | 'error';
 
 export interface SSEEvent {
@@ -39,3 +67,28 @@ export interface SSEEvent {
   status: SSEStatus;
   data?: unknown;
 }
+
+export interface ContinuousSession {
+  sessionId: string;
+  mediaType: MediaType;
+  interpreterOutput: InterpreterOutput;
+  accumulatedCandidates: RetrievalCandidate[];
+  previousResponseId: string | null;
+  batchCount: number;
+  totalCount: number;
+  useWebSearch: boolean;
+  validateResults: boolean;
+  allSeenTitles: Set<string>;
+}
+
+export interface ContinuousStartRequest {
+  mediaType: 'tv_show' | 'movie';
+  likedSeries: string;
+  dislikedSeries?: string;
+  themes?: string;
+  requirements?: string;
+  avoid?: string;
+  totalCount: number;
+  useWebSearch: boolean;
+  validateResults: boolean;
+}

packages/frontend/index.html

@@ -5,6 +5,9 @@
   <meta charset="UTF-8" />
   <link rel="icon" type="image/x-icon" href="/favicon.ico" />
   <meta name="viewport" content="width=device-width, initial-scale=1.0" />
+  <link rel="preconnect" href="https://fonts.googleapis.com" />
+  <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
+  <link href="https://fonts.googleapis.com/css2?family=Sora:wght@300;400;500;600;700&display=swap" rel="stylesheet" />
   <title>Recommender</title>
 </head>
 

packages/frontend/src/api/client.ts

@@ -1,10 +1,12 @@
-import type { Recommendation, RecommendationSummary, FeedbackEntry } from '../types/index.js';
+import type { MediaType, Recommendation, RecommendationSummary, FeedbackEntry } from '../types/index.js';
 
 const BASE = '/api';
 
 async function request<T>(path: string, options?: RequestInit): Promise<T> {
   const res = await fetch(`${BASE}${path}`, {
-    headers: { 'Content-Type': 'application/json' },
+    headers: {
+      ...(options?.body ? { 'Content-Type': 'application/json' } : {}),
+    },
     ...options,
   });
   if (!res.ok) {
@@ -16,10 +18,17 @@ async function request<T>(path: string, options?: RequestInit): Promise<T> {
 
 export function createRecommendation(body: {
   main_prompt: string;
-  liked_shows: string;
+  liked_series: string;
-  disliked_shows: string;
+  disliked_series: string;
   themes: string;
   brainstorm_count?: number;
+  media_type: MediaType;
+  use_web_search?: boolean;
+  use_validator?: boolean;
+  hard_requirements?: boolean;
+  self_expansive?: boolean;
+  expansive_passes?: number;
+  expansive_mode?: 'soft' | 'extreme';
 }): Promise<{ id: string }> {
   return request('/recommendations', {
     method: 'POST',
@@ -40,7 +49,7 @@ export function rerankRecommendation(id: string): Promise<{ ok: boolean }> {
 }
 
 export function submitFeedback(body: {
-  tv_show_name: string;
+  item_name: string;
   stars: number;
   feedback?: string;
 }): Promise<{ ok: boolean }> {
@@ -53,3 +62,85 @@ export function submitFeedback(body: {
 export function getFeedback(): Promise<FeedbackEntry[]> {
   return request('/feedback');
 }
+
+export function deleteRecommendation(id: string): Promise<{ ok: boolean }> {
+  return request(`/recommendations/${id}`, { method: 'DELETE' });
+}
+
+export function createContinuousRecommendation(body: {
+  liked_series: string;
+  disliked_series?: string;
+  themes?: string;
+  requirements?: string;
+  avoid?: string;
+  total_count: number;
+  media_type: MediaType;
+  use_web_search?: boolean;
+  validate_results?: boolean;
+}): Promise<{ id: string }> {
+  return (async () => {
+    const res = await fetch(`${BASE}/recommendations/continuous`, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        Accept: 'text/event-stream',
+      },
+      body: JSON.stringify(body),
+    });
+
+    if (!res.ok) {
+      const text = await res.text();
+      throw new Error(`HTTP ${res.status}: ${text}`);
+    }
+
+    if (!res.body) {
+      throw new Error('Missing response stream for continuous recommendation');
+    }
+
+    const reader = res.body.getReader();
+    const decoder = new TextDecoder();
+    let buffer = '';
+
+    try {
+      while (true) {
+        const { value, done } = await reader.read();
+
+        if (value) {
+          buffer += decoder.decode(value, { stream: !done });
+        }
+
+        let boundary = buffer.indexOf('\n\n');
+        while (boundary !== -1) {
+          const rawEvent = buffer.slice(0, boundary).trim();
+          buffer = buffer.slice(boundary + 2);
+
+          if (rawEvent) {
+            const dataLine = rawEvent
+              .split('\n')
+              .find((line) => line.startsWith('data:'));
+
+            if (dataLine) {
+              const payload = dataLine.slice(5).trim();
+              const event = JSON.parse(payload) as { type?: string; id?: string };
+
+              if (event.type === 'created' && event.id) {
+                await reader.cancel();
+                return { id: event.id };
+              }
+            }
+          }
+
+          boundary = buffer.indexOf('\n\n');
+        }
+
+        if (done) {
+          break;
+        }
+      }
+    } finally {
+      reader.releaseLock();
+    }
+
+    throw new Error('Continuous recommendation stream ended before returning an id');
+  })();
+}

packages/frontend/src/components/Cards.css

@@ -3,20 +3,30 @@
 .cards-grid {
   display: flex;
   flex-direction: column;
-  gap: 12px;
+  gap: var(--sp-3);
-  margin-bottom: 32px;
+  margin-bottom: var(--sp-8);
 }
 
 .card {
-  background: var(--bg-surface);
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.04), transparent 100%),
+    rgba(28, 25, 42, 0.7);
+  backdrop-filter: blur(14px) saturate(150%);
   border: 1px solid var(--border);
-  border-radius: var(--radius);
+  border-radius: var(--radius-md);
-  padding: 16px;
+  padding: var(--sp-4);
-  transition: border-color 0.15s;
+  box-shadow:
+    0 8px 28px rgba(0, 0, 0, 0.18),
+    inset 0 1px 0 rgba(255, 255, 255, 0.04);
+  transition: border-color 0.2s ease, transform 0.2s ease, box-shadow 0.2s ease;
 }
 
 .card:hover {
-  border-color: var(--bg-surface-3);
+  border-color: rgba(232, 168, 124, 0.28);
+  transform: translateY(-1px);
+  box-shadow:
+    0 12px 36px rgba(0, 0, 0, 0.24),
+    inset 0 1px 0 rgba(255, 255, 255, 0.05);
 }
 
 .card-header {
@@ -31,30 +41,53 @@
   font-size: 11px;
   font-weight: 600;
   padding: 2px 8px;
-  border-radius: 100px;
+  border-radius: var(--radius-pill);
   text-transform: uppercase;
-  letter-spacing: 0.04em;
+  letter-spacing: 0.08em;
   flex-shrink: 0;
+  box-shadow: inset 0 0 0 1px rgba(255, 255, 255, 0.04);
 }
 
 .badge-green {
-  background: rgba(34, 197, 94, 0.15);
+  background: rgba(34, 197, 94, 0.14);
   color: #4ade80;
+  box-shadow: 0 0 18px rgba(34, 197, 94, 0.14);
 }
 
 .badge-blue {
   background: rgba(59, 130, 246, 0.15);
   color: #60a5fa;
+  box-shadow: 0 0 18px rgba(59, 130, 246, 0.14);
 }
 
 .badge-yellow {
   background: rgba(234, 179, 8, 0.15);
   color: #facc15;
+  box-shadow: 0 0 18px rgba(234, 179, 8, 0.12);
 }
 
 .badge-red {
   background: rgba(239, 68, 68, 0.15);
   color: #f87171;
+  box-shadow: 0 0 18px rgba(239, 68, 68, 0.12);
+}
+
+.badge-magenta {
+  background: rgba(217, 70, 239, 0.15);
+  color: #e879f9;
+  box-shadow: 0 0 18px rgba(217, 70, 239, 0.14);
+}
+
+.badge-verified {
+  background: rgba(34, 197, 94, 0.15);
+  color: #4ade80;
+}
+
+.card-badges {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 6px;
+  margin-bottom: 10px;
 }
 
 .card-title {
@@ -64,10 +97,44 @@
 }
 
 .card-explanation {
-  font-size: 13px;
+  font-size: 14px;
   color: var(--text-muted);
+  line-height: 1.6;
+  margin-bottom: 10px;
+}
+
+.genre-badge {
+  font-size: 10px;
+  font-weight: 500;
+  background: rgba(245, 240, 232, 0.08);
+  color: var(--text-dim);
+  display: inline-block;
+  text-transform: none;
+  letter-spacing: 0;
+}
+
+.pros-cons-table {
+  display: flex;
+  gap: 16px;
+  margin-bottom: 10px;
+  font-size: 12px;
   line-height: 1.5;
-  margin-bottom: 12px;
+}
+
+.pros-col,
+.cons-col {
+  flex: 1;
+  display: flex;
+  flex-direction: column;
+  gap: 2px;
+}
+
+.pro-item {
+  color: #4ade80;
+}
+
+.con-item {
+  color: #f87171;
 }
 
 .card-feedback {
@@ -125,11 +192,13 @@
   padding: 16px 0 32px;
   display: flex;
   justify-content: center;
+  gap: 12px;
+  flex-wrap: wrap;
 }
 
 .btn-rerank {
   padding: 12px 28px;
-  background: var(--bg-surface-2);
+  background: rgba(255, 255, 255, 0.05);
   color: var(--text);
   border: 1px solid var(--border);
   border-radius: var(--radius);
@@ -140,9 +209,9 @@
 }
 
 .btn-rerank:hover:not(:disabled) {
-  background: var(--bg-surface-3);
+  background: rgba(255, 255, 255, 0.09);
-  border-color: var(--accent);
+  border-color: rgba(232, 168, 124, 0.24);
-  color: var(--accent);
+  color: var(--text);
 }
 
 .btn-rerank:disabled {

packages/frontend/src/components/Modal.css

@@ -1,105 +1,580 @@
-/* ── Modal ──────────────────────────────────────────────── */
-
 .modal-backdrop {
   position: fixed;
   inset: 0;
-  background: rgba(0, 0, 0, 0.6);
+  z-index: 100;
   display: flex;
   align-items: center;
   justify-content: center;
-  z-index: 100;
+  padding: 24px;
-  backdrop-filter: blur(2px);
+  background:
+    radial-gradient(circle at top, rgba(232, 168, 124, 0.12), transparent 32%),
+    radial-gradient(circle at bottom right, rgba(245, 240, 232, 0.08), transparent 28%),
+    rgba(10, 8, 15, 0.78);
+  backdrop-filter: blur(12px);
 }
 
 .modal {
-  background: var(--bg-surface);
+  width: min(900px, 100%);
-  border: 1px solid var(--border);
+  max-height: min(92vh, 980px);
-  border-radius: 12px;
-  width: 650px;
-  max-width: 96vw;
-  max-height: 90vh;
   overflow-y: auto;
-  box-shadow: 0 20px 60px rgba(0, 0, 0, 0.5);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-radius: 28px;
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.03), transparent 18%),
+    linear-gradient(145deg, rgba(28, 25, 42, 0.96), rgba(22, 19, 34, 0.97));
+  box-shadow:
+    0 32px 90px rgba(0, 0, 0, 0.45),
+    inset 0 1px 0 rgba(255, 255, 255, 0.05);
 }
 
+.modal-hero,
 .modal-header {
   display: flex;
-  align-items: center;
+  align-items: flex-start;
   justify-content: space-between;
-  padding: 20px 20px 0;
+  gap: 24px;
+  padding: 28px 30px 0;
 }
 
+.modal-hero h2,
 .modal-header h2 {
-  font-size: 18px;
+  font-size: clamp(1.8rem, 3vw, 2.35rem);
-  font-weight: 700;
+  line-height: 1.05;
-}
+  letter-spacing: -0.03em;
-
-.modal-close {
-  background: none;
-  border: none;
-  font-size: 22px;
-  color: var(--text-muted);
-  cursor: pointer;
-  line-height: 1;
-  padding: 0;
-}
-
-.modal-close:hover {
   color: var(--text);
 }
 
+.modal-hero-copy {
+  max-width: 560px;
+  margin-top: 12px;
+  color: rgba(245, 240, 232, 0.7);
+  font-size: 1rem;
+  line-height: 1.6;
+}
+
+.modal-eyebrow {
+  display: inline-flex;
+  align-items: center;
+  margin-bottom: 14px;
+  padding: 6px 10px;
+  border-radius: 999px;
+  background: rgba(232, 168, 124, 0.14);
+  color: var(--accent);
+  font-size: 0.72rem;
+  font-weight: 700;
+  letter-spacing: 0.12em;
+  text-transform: uppercase;
+}
+
+.modal-close,
+.modal-back {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  width: 42px;
+  height: 42px;
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-radius: 14px;
+  background: rgba(255, 255, 255, 0.03);
+  color: var(--text-muted);
+  cursor: pointer;
+  transition: transform 0.2s ease, border-color 0.2s ease, color 0.2s ease, background 0.2s ease;
+}
+
+.modal-close:hover:not(:disabled),
+.modal-back:hover:not(:disabled) {
+  transform: translateY(-1px);
+  color: var(--text);
+  border-color: rgba(232, 168, 124, 0.34);
+  background: rgba(232, 168, 124, 0.08);
+}
+
+.modal-close:disabled,
+.modal-back:disabled {
+  opacity: 0.45;
+  cursor: not-allowed;
+}
+
+.modal-header-left {
+  display: flex;
+  align-items: flex-start;
+  gap: 16px;
+}
+
+.modal-type-select,
 .modal-form {
-  padding: 20px;
+  padding: 28px 30px 30px;
+}
+
+.modal-type-select {
   display: flex;
   flex-direction: column;
+  gap: 24px;
+}
+
+.modal-section {
+  padding: 22px;
+  border: 1px solid rgba(255, 255, 255, 0.07);
+  border-radius: 24px;
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.02), transparent 100%),
+    rgba(28, 25, 42, 0.58);
+  backdrop-filter: blur(16px) saturate(160%);
+}
+
+.modal-section-header {
+  display: flex;
+  gap: 14px;
+  align-items: flex-start;
+  margin-bottom: 18px;
+}
+
+.modal-section-header h3,
+.settings-card-header h3 {
+  font-size: 1.08rem;
+  color: var(--text);
+}
+
+.modal-section-header p,
+.settings-card-header p {
+  margin-top: 4px;
+  color: var(--text-muted);
+  line-height: 1.5;
+}
+
+.modal-section-step {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  width: 32px;
+  height: 32px;
+  border-radius: 50%;
+  background: linear-gradient(135deg, rgba(232, 168, 124, 0.22), rgba(245, 240, 232, 0.14));
+  color: var(--text);
+  font-weight: 700;
+}
+
+.modal-type-cards,
+.mode-cards,
+.modal-form-grid {
+  display: grid;
   gap: 16px;
 }
 
+.modal-type-cards,
+.mode-cards {
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+}
+
+.type-card,
+.mode-card {
+  position: relative;
+  display: flex;
+  flex-direction: column;
+  align-items: flex-start;
+  gap: 12px;
+  padding: 20px 20px 18px;
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-radius: 22px;
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.03), transparent 100%),
+    rgba(36, 33, 56, 0.68);
+  color: inherit;
+  text-align: left;
+  cursor: pointer;
+  transition: transform 0.2s ease, border-color 0.2s ease, box-shadow 0.2s ease, background 0.2s ease;
+}
+
+.type-card:hover,
+.mode-card:hover {
+  transform: translateY(-2px);
+  border-color: rgba(232, 168, 124, 0.36);
+  box-shadow: 0 18px 32px rgba(2, 8, 23, 0.25);
+}
+
+.type-card--selected,
+.mode-card--active {
+  border-color: rgba(232, 168, 124, 0.55);
+  background:
+    linear-gradient(180deg, rgba(232, 168, 124, 0.12), rgba(245, 240, 232, 0.04)),
+    rgba(28, 25, 42, 0.92);
+  box-shadow: inset 0 0 0 1px rgba(232, 168, 124, 0.16);
+}
+
+.type-card-icon {
+  display: inline-block;
+  font-size: 1.55rem;
+  line-height: 1;
+  margin: 0 2px 0 0;
+  transform: translateY(1px);
+  filter: drop-shadow(0 4px 8px rgba(0, 0, 0, 0.18));
+}
+
+.type-card-label,
+.mode-card-label {
+  color: var(--text);
+  font-size: 1.04rem;
+  font-weight: 700;
+}
+
+.type-card-title-row {
+  display: flex;
+  align-items: center;
+  gap: 8px;
+}
+
+.type-card-desc,
+.mode-card-desc {
+  color: rgba(245, 240, 232, 0.66);
+  line-height: 1.55;
+  font-size: 0.92rem;
+}
+
+.mode-card-badge {
+  display: inline-flex;
+  align-items: center;
+  padding: 4px 10px;
+  border-radius: 999px;
+  background: rgba(232, 168, 124, 0.14);
+  color: var(--accent);
+  font-size: 0.72rem;
+  font-weight: 700;
+  letter-spacing: 0.04em;
+}
+
+.modal-type-footer {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 20px;
+}
+
+.modal-selection-summary,
+.modal-summary-strip {
+  display: flex;
+  align-items: center;
+  gap: 10px;
+  flex-wrap: wrap;
+}
+
+.summary-pill {
+  display: inline-flex;
+  align-items: center;
+  padding: 8px 12px;
+  border-radius: 999px;
+  background: rgba(255, 255, 255, 0.05);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  color: var(--text);
+  font-size: 0.8rem;
+  font-weight: 600;
+}
+
+.summary-pill--accent {
+  background: rgba(232, 168, 124, 0.12);
+  border-color: rgba(232, 168, 124, 0.22);
+  color: var(--accent);
+}
+
+.summary-caption {
+  color: var(--text-muted);
+  font-size: 0.9rem;
+}
+
+.modal-continue {
+  padding-inline: 22px;
+}
+
+.modal-form {
+  display: flex;
+  flex-direction: column;
+  gap: 22px;
+}
+
+.modal-form-grid {
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+}
+
 .form-group {
   display: flex;
   flex-direction: column;
-  gap: 6px;
+  gap: 8px;
-  flex: 1;
+}
+
+.form-group--full {
+  grid-column: 1 / -1;
 }
 
 .form-group label {
-  font-size: 13px;
+  color: var(--text);
-  font-weight: 500;
+  font-size: 0.92rem;
-  color: var(--text-muted);
+  font-weight: 600;
+}
+
+.form-help {
+  color: var(--text-dim);
+  font-size: 0.8rem;
+  line-height: 1.5;
 }
 
 .form-input,
 .form-textarea {
-  background: var(--bg-surface-2);
-  border: 1px solid var(--border);
-  border-radius: var(--radius-sm);
-  color: var(--text);
-  font-size: 14px;
-  padding: 10px 12px;
-  outline: none;
-  transition: border-color 0.15s;
-  font-family: inherit;
   width: 100%;
+  padding: 13px 14px;
+  border: 1px solid rgba(255, 255, 255, 0.09);
+  border-radius: 16px;
+  background: rgba(18, 16, 26, 0.85);
+  color: var(--text);
+  font: inherit;
+  outline: none;
+  transition: border-color 0.2s ease, box-shadow 0.2s ease, background 0.2s ease;
 }
 
-.form-input[type="range"] {
+.form-input::placeholder,
-  padding: 0;
+.form-textarea::placeholder {
-  cursor: pointer;
+  color: var(--text-dim);
 }
 
 .form-input:focus,
 .form-textarea:focus {
-  border-color: var(--accent);
+  border-color: rgba(232, 168, 124, 0.52);
+  box-shadow: 0 0 0 4px rgba(232, 168, 124, 0.12);
+  background: rgba(18, 16, 26, 0.94);
+}
+
+.form-input[type='range'] {
+  padding: 0;
+  border: none;
+  background: rgba(255, 255, 255, 0.1);
+  box-shadow: none;
+  cursor: pointer;
 }
 
 .form-textarea {
   resize: vertical;
-  min-height: 80px;
+  min-height: 120px;
+}
+
+.form-textarea--hero {
+  min-height: 138px;
+}
+
+.settings-card {
+  display: flex;
+  flex-direction: column;
+  gap: 16px;
+  padding: 22px;
+  border: 1px solid rgba(255, 255, 255, 0.07);
+  border-radius: 24px;
+  background:
+    radial-gradient(circle at top right, rgba(232, 168, 124, 0.08), transparent 28%),
+    rgba(28, 25, 42, 0.58);
+}
+
+.slider-card {
+  padding: 16px 18px;
+  border-radius: 18px;
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  background: rgba(18, 16, 26, 0.72);
+}
+
+.slider-card--compact {
+  padding: 14px 16px;
+}
+
+.slider-card-header {
+  display: flex;
+  align-items: flex-start;
+  justify-content: space-between;
+  gap: 18px;
+  margin-bottom: 12px;
+}
+
+.slider-label {
+  display: block;
+  color: var(--text);
+  font-weight: 600;
+}
+
+.slider-copy {
+  display: block;
+  margin-top: 4px;
+  color: var(--text-muted);
+  font-size: 0.82rem;
+  line-height: 1.5;
+}
+
+.slider-value {
+  min-width: 48px;
+  padding: 8px 10px;
+  border-radius: 12px;
+  background: rgba(232, 168, 124, 0.12);
+  color: var(--accent);
+  font-weight: 700;
+  text-align: center;
+}
+
+.toggle-stack {
+  display: flex;
+  flex-direction: column;
+  gap: 12px;
+}
+
+.toggle-card {
+  padding: 15px 16px;
+  border-radius: 18px;
+  border: 1px solid rgba(255, 255, 255, 0.07);
+  background: rgba(28, 25, 42, 0.6);
+}
+
+.toggle-card--disabled {
+  opacity: 0.55;
+}
+
+.toggle-label {
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  gap: 16px;
+}
+
+.toggle-text {
+  display: flex;
+  flex-direction: column;
+  gap: 4px;
+}
+
+.toggle-title {
+  color: var(--text);
+  font-weight: 600;
+}
+
+.toggle-desc {
+  color: var(--text-muted);
+  font-size: 0.84rem;
+  line-height: 1.5;
+}
+
+.toggle-switch {
+  position: relative;
+  flex-shrink: 0;
+  width: 52px;
+  height: 30px;
+  border: 1px solid rgba(255, 255, 255, 0.11);
+  border-radius: 999px;
+  background: rgba(255, 255, 255, 0.06);
+  cursor: pointer;
+  transition: background 0.2s ease, border-color 0.2s ease;
+}
+
+.toggle-switch.on {
+  background: rgba(232, 168, 124, 0.3);
+  border-color: rgba(232, 168, 124, 0.4);
+}
+
+.toggle-switch-disabled {
+  cursor: not-allowed;
+}
+
+.toggle-knob {
+  position: absolute;
+  top: 3px;
+  left: 3px;
+  width: 22px;
+  height: 22px;
+  border-radius: 50%;
+  background: var(--text-muted);
+  transition: transform 0.2s ease;
+}
+
+.toggle-switch.on .toggle-knob {
+  transform: translateX(22px);
+  background: var(--accent);
+  box-shadow: 0 0 8px rgba(232, 168, 124, 0.4);
+}
+
+.expansive-options {
+  display: flex;
+  flex-direction: column;
+  gap: 14px;
+  padding: 16px;
+  border-radius: 18px;
+  border: 1px solid rgba(232, 168, 124, 0.14);
+  background: rgba(232, 168, 124, 0.08);
+}
+
+.segmented-control {
+  display: grid;
+  grid-template-columns: repeat(2, minmax(0, 1fr));
+  gap: 10px;
+}
+
+.segmented-control-btn {
+  padding: 12px 14px;
+  border-radius: 14px;
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  background: rgba(18, 16, 26, 0.66);
+  color: var(--text-muted);
+  cursor: pointer;
+  font: inherit;
+  font-weight: 600;
+  transition: border-color 0.2s ease, color 0.2s ease, background 0.2s ease;
+}
+
+.segmented-control-btn--active {
+  background: rgba(232, 168, 124, 0.12);
+  border-color: rgba(232, 168, 124, 0.34);
+  color: var(--accent);
 }
 
 .modal-actions {
   display: flex;
   justify-content: flex-end;
-  gap: 10px;
+  gap: 12px;
-  padding-top: 4px;
+  padding-top: 2px;
+}
+
+@media (max-width: 860px) {
+  .modal {
+    width: 100%;
+    border-radius: 24px;
+  }
+
+  .modal-hero,
+  .modal-header,
+  .modal-type-select,
+  .modal-form {
+    padding-inline: 22px;
+  }
+}
+
+@media (max-width: 720px) {
+  .modal-backdrop {
+    padding: 12px;
+  }
+
+  .modal-hero,
+  .modal-header {
+    flex-direction: column;
+  }
+
+  .modal-close {
+    align-self: flex-end;
+  }
+
+  .modal-type-cards,
+  .mode-cards,
+  .modal-form-grid {
+    grid-template-columns: 1fr;
+  }
+
+  .modal-type-footer,
+  .modal-actions {
+    flex-direction: column;
+    align-items: stretch;
+  }
+
+  .toggle-label,
+  .slider-card-header {
+    align-items: flex-start;
+  }
 }

packages/frontend/src/components/NewRecommendationModal.tsx

@@ -1,37 +1,160 @@
-import { useState } from 'preact/hooks';
+import { useEffect, useState } from 'preact/hooks';
+import type { MediaType } from '../types/index.js';
 import './Modal.css';
 
+type GenerationMode = 'brainstorm' | 'continuous';
+
 interface NewRecommendationModalProps {
   onClose: () => void;
   onSubmit: (body: {
     main_prompt: string;
-    liked_shows: string;
+    liked_series: string;
-    disliked_shows: string;
+    disliked_series: string;
     themes: string;
+    requirements?: string;
+    avoid?: string;
     brainstorm_count?: number;
+    media_type: MediaType;
+    use_web_search?: boolean;
+    use_validator?: boolean;
+    hard_requirements?: boolean;
+    self_expansive?: boolean;
+    expansive_passes?: number;
+    expansive_mode?: 'soft' | 'extreme';
+    generation_mode?: GenerationMode;
+    total_count?: number;
+    validate_results?: boolean;
   }) => Promise<void>;
 }
 
+const MEDIA_OPTIONS: Array<{
+  type: MediaType;
+  icon: string;
+  label: string;
+  description: string;
+}> = [
+    {
+      type: 'tv_show',
+      icon: '📺',
+      label: 'TV series',
+      description: 'Serialized stories, limited series, and long-form comfort watches.',
+    },
+    {
+      type: 'movie',
+      icon: '🎬',
+      label: 'Movies',
+      description: 'Feature films, prestige cinema, and one-night picks.',
+    },
+  ];
+
+const MODE_OPTIONS: Array<{
+  mode: GenerationMode;
+  label: string;
+  badge: string;
+  description: string;
+}> = [
+    {
+      mode: 'brainstorm',
+      label: 'Brainstorm',
+      badge: 'Best for variety',
+      description: 'Explore a broad pool of options, then rank and curate the strongest fits.',
+    },
+    {
+      mode: 'continuous',
+      label: 'Continuous',
+      badge: 'Best for deep search',
+      description: 'Generate recommendations in chained batches for a steadier, longer-running hunt.',
+    },
+  ];
+
 export function NewRecommendationModal({ onClose, onSubmit }: NewRecommendationModalProps) {
+  const [step, setStep] = useState<'type' | 'mode' | 'form'>('type');
+  const [mediaType, setMediaType] = useState<MediaType>('tv_show');
+  const [generationMode, setGenerationMode] = useState<GenerationMode>('brainstorm');
   const [mainPrompt, setMainPrompt] = useState('');
-  const [likedShows, setLikedShows] = useState('');
+  const [likedSeries, setLikedSeries] = useState('');
-  const [dislikedShows, setDislikedShows] = useState('');
+  const [dislikedSeries, setDislikedSeries] = useState('');
   const [themes, setThemes] = useState('');
+  const [requirements, setRequirements] = useState('');
+  const [avoid, setAvoid] = useState('');
   const [brainstormCount, setBrainstormCount] = useState(100);
+  const [totalCount, setTotalCount] = useState(30);
+  const [useWebSearch, setUseWebSearch] = useState(false);
+  const [useValidator, setUseValidator] = useState(false);
+  const [useHardRequirements, setUseHardRequirements] = useState(false);
+  const [selfExpansive, setSelfExpansive] = useState(false);
+  const [expansivePasses, setExpansivePasses] = useState(2);
+  const [expansiveMode, setExpansiveMode] = useState<'soft' | 'extreme'>('soft');
+  const [validateResults, setValidateResults] = useState(false);
   const [loading, setLoading] = useState(false);
 
+  useEffect(() => {
+    const handleKeyDown = (event: KeyboardEvent) => {
+      if (event.key === 'Escape' && !loading) {
+        onClose();
+      }
+    };
+
+    window.addEventListener('keydown', handleKeyDown);
+    return () => window.removeEventListener('keydown', handleKeyDown);
+  }, [loading, onClose]);
+
+  const mediaLabel = mediaType === 'movie' ? 'Movie' : 'TV Show';
+  const mediaPluralLabel = mediaType === 'movie' ? 'movies' : 'series';
+
+  const handleSelectType = (type: MediaType) => {
+    setMediaType(type);
+    setStep('mode');
+  };
+
+  const handleWebSearchToggle = () => {
+    const next = !useWebSearch;
+    setUseWebSearch(next);
+    if (!next) {
+      setUseValidator(false);
+      setValidateResults(false);
+    }
+  };
+
   const handleSubmit = async (e: Event) => {
     e.preventDefault();
-    if (!mainPrompt.trim()) return;
+    if (generationMode === 'brainstorm' && !mainPrompt.trim()) return;
+    if (!likedSeries.trim()) return;
+
     setLoading(true);
     try {
+      if (generationMode === 'brainstorm') {
         await onSubmit({
           main_prompt: mainPrompt.trim(),
-        liked_shows: likedShows.trim(),
+          liked_series: likedSeries.trim(),
-        disliked_shows: dislikedShows.trim(),
+          disliked_series: dislikedSeries.trim(),
           themes: themes.trim(),
           brainstorm_count: brainstormCount,
+          media_type: mediaType,
+          use_web_search: useWebSearch,
+          use_validator: useValidator,
+          hard_requirements: useHardRequirements,
+          self_expansive: selfExpansive,
+          expansive_passes: selfExpansive ? expansivePasses : 1,
+          expansive_mode: expansiveMode,
+          generation_mode: 'brainstorm',
         });
+      } else {
+        await onSubmit({
+          main_prompt: '',
+          liked_series: likedSeries.trim(),
+          disliked_series: dislikedSeries.trim(),
+          themes: themes.trim(),
+          requirements: requirements.trim(),
+          avoid: avoid.trim(),
+          media_type: mediaType,
+          use_web_search: useWebSearch,
+          validate_results: validateResults,
+          generation_mode: 'continuous',
+          total_count: totalCount,
+        });
+      }
+
       onClose();
     } finally {
       setLoading(false);
@@ -39,71 +162,253 @@ export function NewRecommendationModal({ onClose, onSubmit }: NewRecommendationM
   };
 
   const handleBackdropClick = (e: MouseEvent) => {
-    if ((e.target as HTMLElement).classList.contains('modal-backdrop')) {
+    if ((e.target as HTMLElement).classList.contains('modal-backdrop') && !loading) {
       onClose();
     }
   };
 
+  const selectedMode = MODE_OPTIONS.find((option) => option.mode === generationMode);
+
   return (
     <div class="modal-backdrop" onClick={handleBackdropClick}>
-      <div class="modal">
+      <div class="modal" role="dialog" aria-modal="true" aria-labelledby="recommendation-modal-title">
+        {step === 'type' ? (
+          <>
+            <div class="modal-hero">
+              <div>
+                <div class="modal-eyebrow">Create Recommendation</div>
+                <h2 id="recommendation-modal-title">Shape the kind of discovery session you want.</h2>
+                <p class="modal-hero-copy">
+                  Pick a format, choose how deep the search should go, and we&apos;ll take it from there.
+                </p>
+              </div>
+              <button class="modal-close" onClick={onClose} aria-label="Close" disabled={loading}>×</button>
+            </div>
+
+            <div class="modal-type-select">
+              <section class="modal-section">
+                <div class="modal-section-header">
+                  <span class="modal-section-step">1</span>
+                  <div>
+                    <h3>Choose your format</h3>
+                    <p>Start with the kind of thing you want to watch next.</p>
+                  </div>
+                </div>
+                <div class="modal-type-cards">
+                  {MEDIA_OPTIONS.map((option) => (
+                    <button
+                      key={option.type}
+                      type="button"
+                      class={`type-card${mediaType === option.type ? ' type-card--selected' : ''}`}
+                      onClick={() => handleSelectType(option.type)}
+                    >
+                      <span class="type-card-title-row">
+                        <span class="type-card-icon">{option.icon}</span>
+                        <span class="type-card-label">{option.label}</span>
+                      </span>
+                      <span class="type-card-desc">{option.description}</span>
+                    </button>
+                  ))}
+                </div>
+              </section>
+            </div>
+          </>
+        ) : step === 'mode' ? (
+          <>
             <div class="modal-header">
-          <h2>New Recommendation</h2>
+              <div class="modal-header-left">
-          <button class="modal-close" onClick={onClose} aria-label="Close">×</button>
+                <button class="modal-back" onClick={() => setStep('type')} aria-label="Back" disabled={loading}>
+                  ←
+                </button>
+                <div>
+                  <div class="modal-eyebrow">Step 2 of 3</div>
+                  <h2 id="recommendation-modal-title">Choose the pipeline style.</h2>
+                </div>
+              </div>
+              <button class="modal-close" onClick={onClose} aria-label="Close" disabled={loading}>×</button>
+            </div>
+
+            <div class="modal-type-select">
+              <section class="modal-section">
+                <div class="modal-section-header">
+                  <span class="modal-section-step">2</span>
+                  <div>
+                    <h3>Pick the search mode</h3>
+                    <p>Go broad for variety or deeper for a more persistent search.</p>
+                  </div>
+                </div>
+                <div class="mode-cards">
+                  {MODE_OPTIONS.map((option) => (
+                    <button
+                      key={option.mode}
+                      type="button"
+                      class={`mode-card${generationMode === option.mode ? ' mode-card--active' : ''}`}
+                      onClick={() => {
+                        setGenerationMode(option.mode);
+                        setStep('form');
+                      }}
+                    >
+                      <span class="mode-card-badge">{option.badge}</span>
+                      <span class="mode-card-label">{option.label}</span>
+                      <span class="mode-card-desc">{option.description}</span>
+                    </button>
+                  ))}
+                </div>
+              </section>
+
+              <div class="modal-type-footer">
+                <div class="modal-selection-summary">
+                  <span class="summary-pill">{mediaType === 'movie' ? 'Movies' : 'TV series'}</span>
+                  <span class="summary-pill">{selectedMode?.label}</span>
+                </div>
+              </div>
+            </div>
+          </>
+        ) : (
+          <>
+            <div class="modal-header">
+              <div class="modal-header-left">
+                <button class="modal-back" onClick={() => setStep('mode')} aria-label="Back" disabled={loading}>
+                  ←
+                </button>
+                <div>
+                  <div class="modal-eyebrow">Step 3 of 3</div>
+                  <h2 id="recommendation-modal-title">Tell us what a great match looks like.</h2>
+                </div>
+              </div>
+              <button class="modal-close" onClick={onClose} aria-label="Close" disabled={loading}>×</button>
             </div>
 
             <form class="modal-form" onSubmit={handleSubmit}>
-          <div class="form-group">
+              <div class="modal-summary-strip">
+                <span class="summary-pill">{mediaType === 'movie' ? 'Movies' : 'TV series'}</span>
+                <span class="summary-pill summary-pill--accent">{selectedMode?.label}</span>
+                <span class="summary-caption">
+                  {generationMode === 'brainstorm'
+                    ? 'Fast wide search with ranking and curation.'
+                    : 'Longer chained search that keeps exploring in batches.'}
+                </span>
+              </div>
+
+              {generationMode === 'brainstorm' && (
+                <div class="form-group form-group--full">
                   <label for="main-prompt">What are you looking for?</label>
                   <textarea
                     id="main-prompt"
-              class="form-textarea"
+                    class="form-textarea form-textarea--hero"
-              placeholder="Describe what you want to watch. Be as specific as you like — mood, themes, setting, what you've enjoyed before..."
+                    placeholder="Describe the mood, themes, setting, pacing, or vibe you want right now."
                     value={mainPrompt}
                     onInput={(e) => setMainPrompt((e.target as HTMLTextAreaElement).value)}
                     rows={5}
                     required
                   />
                 </div>
+              )}
 
+              <div class="modal-form-grid">
                 <div class="form-group">
-            <label for="liked-shows">Shows you liked</label>
+                  <label for="liked-series">{mediaLabel}s you liked</label>
                   <input
-              id="liked-shows"
+                    id="liked-series"
                     type="text"
                     class="form-input"
-              placeholder="e.g. Breaking Bad, The Wire"
+                    placeholder={mediaType === 'movie' ? 'e.g. Inception, The Godfather' : 'e.g. Breaking Bad, The Wire'}
-              value={likedShows}
+                    value={likedSeries}
-              onInput={(e) => setLikedShows((e.target as HTMLInputElement).value)}
+                    onInput={(e) => setLikedSeries((e.target as HTMLInputElement).value)}
+                    required
                   />
+                  <span class="form-help">A few strong examples help the pipeline lock onto your taste.</span>
                 </div>
 
                 <div class="form-group">
-            <label for="disliked-shows">Shows you disliked</label>
+                  <label for="disliked-series">{mediaLabel}s you disliked</label>
                   <input
-              id="disliked-shows"
+                    id="disliked-series"
                     type="text"
                     class="form-input"
-              placeholder="e.g. Game of Thrones"
+                    placeholder={mediaType === 'movie' ? 'e.g. Transformers' : 'e.g. Game of Thrones'}
-              value={dislikedShows}
+                    value={dislikedSeries}
-              onInput={(e) => setDislikedShows((e.target as HTMLInputElement).value)}
+                    onInput={(e) => setDislikedSeries((e.target as HTMLInputElement).value)}
                   />
+                  <span class="form-help">Optional, but useful when you want to steer away from common misses.</span>
                 </div>
 
-          <div class="form-group">
+                <div class={`form-group${generationMode === 'continuous' ? '' : ' form-group--full'}`}>
-            <label for="themes">Themes and requirements</label>
+                  <label for="themes">Themes, moods, and preferences</label>
                   <input
                     id="themes"
                     type="text"
                     class="form-input"
-              placeholder="e.g. dramatic setting, historical, sci-fi, etc."
+                    placeholder="e.g. tense, grounded sci-fi, emotional, political intrigue"
                     value={themes}
                     onInput={(e) => setThemes((e.target as HTMLInputElement).value)}
                   />
                 </div>
 
+                {generationMode === 'continuous' && (
+                  <>
                     <div class="form-group">
-            <label for="brainstorm-count">Shows to brainstorm ({brainstormCount})</label>
+                      <label for="requirements">Hard requirements</label>
+                      <input
+                        id="requirements"
+                        type="text"
+                        class="form-input"
+                        placeholder="e.g. 2020+, under 2 hours, streaming on Netflix"
+                        value={requirements}
+                        onInput={(e) => setRequirements((e.target as HTMLInputElement).value)}
+                      />
+                    </div>
+
+                    <div class="form-group">
+                      <label for="avoid">Avoid</label>
+                      <input
+                        id="avoid"
+                        type="text"
+                        class="form-input"
+                        placeholder="e.g. gore, bleak endings, broad comedy"
+                        value={avoid}
+                        onInput={(e) => setAvoid((e.target as HTMLInputElement).value)}
+                      />
+                    </div>
+                  </>
+                )}
+              </div>
+
+              <div class="settings-card">
+                <div class="settings-card-header">
+                  <h3>Tuning</h3>
+                  <p>Adjust depth, validation, and how aggressively the search expands.</p>
+                </div>
+
+                {generationMode === 'continuous' ? (
+                  <div class="slider-card">
+                    <div class="slider-card-header">
+                      <div>
+                        <span class="slider-label">Total recommendations</span>
+                        <span class="slider-copy">Generated in batches of 10 through the continuous pipeline.</span>
+                      </div>
+                      <span class="slider-value">{totalCount}</span>
+                    </div>
+                    <input
+                      id="total-count"
+                      type="range"
+                      class="form-input"
+                      min={10}
+                      max={100}
+                      step={10}
+                      value={totalCount}
+                      onInput={(e) => setTotalCount(Number((e.target as HTMLInputElement).value))}
+                    />
+                  </div>
+                ) : (
+                  <div class="slider-card">
+                    <div class="slider-card-header">
+                      <div>
+                        <span class="slider-label">Brainstorm pool size</span>
+                        <span class="slider-copy">Higher values search wider before ranking and curation.</span>
+                      </div>
+                      <span class="slider-value">{brainstormCount}</span>
+                    </div>
                     <input
                       id="brainstorm-count"
                       type="range"
@@ -115,16 +420,154 @@ export function NewRecommendationModal({ onClose, onSubmit }: NewRecommendationM
                       onInput={(e) => setBrainstormCount(Number((e.target as HTMLInputElement).value))}
                     />
                   </div>
+                )}
+
+                <div class="toggle-stack">
+                  <div class="toggle-card">
+                    <label class="toggle-label" for="web-search">
+                      <div class="toggle-text">
+                        <span class="toggle-title">Use Web Search</span>
+                        <span class="toggle-desc">Pull in live web context for fresher, more grounded {mediaPluralLabel}.</span>
+                      </div>
+                      <button
+                        id="web-search"
+                        type="button"
+                        class={`toggle-switch${useWebSearch ? ' on' : ''}`}
+                        aria-pressed={useWebSearch}
+                        onClick={handleWebSearchToggle}
+                      >
+                        <div class="toggle-knob" />
+                      </button>
+                    </label>
+                  </div>
+
+                  {generationMode === 'brainstorm' ? (
+                    <>
+                      <div class={`toggle-card${!useWebSearch ? ' toggle-card--disabled' : ''}`}>
+                        <label class="toggle-label">
+                          <div class="toggle-text">
+                            <span class="toggle-title">Validator Agent</span>
+                            <span class="toggle-desc">
+                              Verify brainstormed candidates against real metadata before ranking.
+                            </span>
+                          </div>
+                          <button
+                            type="button"
+                            class={`toggle-switch${useValidator ? ' on' : ''}${!useWebSearch ? ' toggle-switch-disabled' : ''}`}
+                            aria-pressed={useValidator}
+                            onClick={() => useWebSearch && setUseValidator((value) => !value)}
+                          >
+                            <div class="toggle-knob" />
+                          </button>
+                        </label>
+                      </div>
+
+                      <div class="toggle-card">
+                        <label class="toggle-label">
+                          <div class="toggle-text">
+                            <span class="toggle-title">Hard Requirements</span>
+                            <span class="toggle-desc">Bias ranking toward strict fit instead of softer exploratory matches.</span>
+                          </div>
+                          <button
+                            type="button"
+                            class={`toggle-switch${useHardRequirements ? ' on' : ''}`}
+                            aria-pressed={useHardRequirements}
+                            onClick={() => setUseHardRequirements((value) => !value)}
+                          >
+                            <div class="toggle-knob" />
+                          </button>
+                        </label>
+                      </div>
+
+                      <div class="toggle-card">
+                        <label class="toggle-label">
+                          <div class="toggle-text">
+                            <span class="toggle-title">Self Expansive Search</span>
+                            <span class="toggle-desc">Run extra passes using earlier full matches as context.</span>
+                          </div>
+                          <button
+                            type="button"
+                            class={`toggle-switch${selfExpansive ? ' on' : ''}`}
+                            aria-pressed={selfExpansive}
+                            onClick={() => setSelfExpansive((value) => !value)}
+                          >
+                            <div class="toggle-knob" />
+                          </button>
+                        </label>
+                      </div>
+                    </>
+                  ) : (
+                    <div class={`toggle-card${!useWebSearch ? ' toggle-card--disabled' : ''}`}>
+                      <label class="toggle-label">
+                        <div class="toggle-text">
+                          <span class="toggle-title">Validate Results</span>
+                          <span class="toggle-desc">Verify continuous batches against real metadata before final ranking.</span>
+                        </div>
+                        <button
+                          type="button"
+                          class={`toggle-switch${validateResults ? ' on' : ''}${!useWebSearch ? ' toggle-switch-disabled' : ''}`}
+                          aria-pressed={validateResults}
+                          onClick={() => useWebSearch && setValidateResults((value) => !value)}
+                        >
+                          <div class="toggle-knob" />
+                        </button>
+                      </label>
+                    </div>
+                  )}
+                </div>
+
+                {generationMode === 'brainstorm' && selfExpansive && (
+                  <div class="expansive-options">
+                    <div class="slider-card slider-card--compact">
+                      <div class="slider-card-header">
+                        <div>
+                          <span class="slider-label">Extra passes</span>
+                          <span class="slider-copy">How many additional exploration rounds to run.</span>
+                        </div>
+                        <span class="slider-value">{expansivePasses}</span>
+                      </div>
+                      <input
+                        type="range"
+                        class="form-input"
+                        min={1}
+                        max={5}
+                        step={1}
+                        value={expansivePasses}
+                        onInput={(e) => setExpansivePasses(Number((e.target as HTMLInputElement).value))}
+                      />
+                    </div>
+
+                    <div class="segmented-control">
+                      <button
+                        type="button"
+                        class={`segmented-control-btn${expansiveMode === 'soft' ? ' segmented-control-btn--active' : ''}`}
+                        onClick={() => setExpansiveMode('soft')}
+                      >
+                        Soft
+                      </button>
+                      <button
+                        type="button"
+                        class={`segmented-control-btn${expansiveMode === 'extreme' ? ' segmented-control-btn--active' : ''}`}
+                        onClick={() => setExpansiveMode('extreme')}
+                      >
+                        Extreme
+                      </button>
+                    </div>
+                  </div>
+                )}
+              </div>
 
               <div class="modal-actions">
                 <button type="button" class="btn-secondary" onClick={onClose} disabled={loading}>
                   Cancel
                 </button>
-            <button type="submit" class="btn-primary" disabled={loading || !mainPrompt.trim()}>
+                <button type="submit" class="btn-primary" disabled={loading}>
-              {loading ? 'Starting…' : 'Get Recommendations'}
+                  {loading ? 'Creating…' : generationMode === 'brainstorm' ? 'Start Recommendation' : 'Start Continuous Search'}
                 </button>
               </div>
             </form>
+          </>
+        )}
       </div>
     </div>
   );

packages/frontend/src/components/PipelineProgress.css

@@ -1,14 +1,14 @@
 /* ── Pipeline Progress ──────────────────────────────────── */
 
 .pipeline-progress {
-  padding: 40px 0;
+  padding: var(--sp-8) 0;
 }
 
 .pipeline-title {
   font-size: 18px;
   font-weight: 600;
   color: var(--text);
-  margin-bottom: 24px;
+  margin-bottom: var(--sp-6);
 }
 
 .pipeline-steps {
@@ -23,15 +23,21 @@
   align-items: center;
   gap: 12px;
   padding: 12px 16px;
-  background: var(--bg-surface);
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.03), transparent 100%),
+    var(--bg-surface);
+  backdrop-filter: blur(14px) saturate(150%);
   border: 1px solid var(--border);
-  border-radius: var(--radius);
+  border-radius: var(--radius-md);
   font-size: 14px;
+  box-shadow: 0 10px 24px rgba(0, 0, 0, 0.16);
 }
 
 .pipeline-step--running {
-  border-color: var(--accent);
+  border-color: rgba(232, 168, 124, 0.3);
-  background: var(--accent-dim);
+  background:
+    linear-gradient(180deg, rgba(232, 168, 124, 0.08), transparent 100%),
+    rgba(28, 25, 42, 0.82);
 }
 
 .pipeline-step--done {
@@ -84,3 +90,30 @@
 .pipeline-step-label {
   font-weight: 500;
 }
+
+.pipeline-retry {
+  margin-top: 20px;
+}
+
+.pipeline-step--skipped {
+  border-color: var(--border);
+  background: var(--bg-surface);
+  opacity: 0.45;
+}
+
+.stage-skipped {
+  color: var(--text-dim);
+}
+
+.pipeline-pass-group--extra {
+  margin-top: 20px;
+}
+
+.pipeline-pass-label {
+  font-size: 11px;
+  font-weight: 700;
+  text-transform: uppercase;
+  letter-spacing: 0.06em;
+  color: var(--text-dim);
+  margin-bottom: 8px;
+}

packages/frontend/src/components/PipelineProgress.tsx

@@ -1,15 +1,20 @@
 import './PipelineProgress.css';
 import type { StageMap, StageStatus } from '../types/index.js';
 
-const STAGES: { key: keyof StageMap; label: string }[] = [
+export interface StageEntry {
-  { key: 'interpreter', label: 'Interpreting Preferences' },
+  key: string;
-  { key: 'retrieval', label: 'Generating Candidates' },
+  label: string;
-  { key: 'ranking', label: 'Ranking Shows' },
+}
-  { key: 'curator', label: 'Curating Results' },
+
-];
+export interface StageGroup {
+  label: string;
+  stages: StageEntry[];
+}
 
 interface PipelineProgressProps {
+  stageGroups: StageGroup[];
   stages: StageMap;
+  onRetry?: () => void;
 }
 
 function StageIcon({ status }: { status: StageStatus }) {
@@ -20,23 +25,40 @@ function StageIcon({ status }: { status: StageStatus }) {
       return <span class="stage-icon stage-error">✗</span>;
     case 'running':
       return <span class="stage-icon stage-running spinner">⟳</span>;
+    case 'skipped':
+      return <span class="stage-icon stage-skipped">—</span>;
     default:
       return <span class="stage-icon stage-pending">○</span>;
   }
 }
 
-export function PipelineProgress({ stages }: PipelineProgressProps) {
+export function PipelineProgress({ stageGroups, stages, onRetry }: PipelineProgressProps) {
+  const hasError = Object.values(stages).some((s) => s === 'error');
+
   return (
     <div class="pipeline-progress">
-      <h3 class="pipeline-title">Generating Recommendations…</h3>
+      <h3 class="pipeline-title">{hasError ? 'Pipeline Failed' : 'Generating Recommendations…'}</h3>
+      {stageGroups.map((group, gi) => (
+        <div key={gi} class={`pipeline-pass-group${gi > 0 ? ' pipeline-pass-group--extra' : ''}`}>
+          {group.label && <div class="pipeline-pass-label">{group.label}</div>}
           <ul class="pipeline-steps">
-        {STAGES.map(({ key, label }) => (
+            {group.stages.map(({ key, label }) => {
-          <li key={key} class={`pipeline-step pipeline-step--${stages[key]}`}>
+              const status: StageStatus = stages[key] ?? 'pending';
-            <StageIcon status={stages[key]} />
+              return (
+                <li key={key} class={`pipeline-step pipeline-step--${status}`}>
+                  <StageIcon status={status} />
                   <span class="pipeline-step-label">{label}</span>
                 </li>
-        ))}
+              );
+            })}
           </ul>
         </div>
+      ))}
+      {hasError && onRetry && (
+        <div class="pipeline-retry">
+          <button class="btn-primary" onClick={onRetry}>Re-run Pipeline</button>
+        </div>
+      )}
+    </div>
   );
 }

packages/frontend/src/components/RecommendationCard.tsx

@@ -4,19 +4,22 @@ import type { CuratorOutput, CuratorCategory } from '../types/index.js';
 
 interface RecommendationCardProps {
   show: CuratorOutput;
+  verified?: boolean;
   existingFeedback?: { stars: number; feedback: string };
-  onFeedback: (tv_show_name: string, stars: number, feedback: string) => Promise<void>;
+  onFeedback: (item_name: string, stars: number, feedback: string) => Promise<void>;
 }
 
 const CATEGORY_COLORS: Record<CuratorCategory, string> = {
+  'Full Match': 'badge-magenta',
   'Definitely Like': 'badge-green',
   'Might Like': 'badge-blue',
   'Questionable': 'badge-yellow',
   'Will Not Like': 'badge-red',
 };
 
-export function RecommendationCard({ show, existingFeedback, onFeedback }: RecommendationCardProps) {
+export function RecommendationCard({ show, verified, existingFeedback, onFeedback }: RecommendationCardProps) {
   const [selectedStars, setSelectedStars] = useState(existingFeedback?.stars ?? 0);
+  const [hoverStar, setHoverStar] = useState(0);
   const [comment, setComment] = useState(existingFeedback?.feedback ?? '');
   const [showComment, setShowComment] = useState(false);
   const [submitted, setSubmitted] = useState(!!existingFeedback);
@@ -47,18 +50,39 @@ export function RecommendationCard({ show, existingFeedback, onFeedback }: Recom
       </div>
       <p class="card-explanation">{show.explanation}</p>
 
+      <div class="card-badges">
+        {show.genre && <span class="badge genre-badge">{show.genre}</span>}
+        {verified && <span class="badge badge-verified">Verified</span>}
+      </div>
+
+      {(show.pros?.length > 0 || show.cons?.length > 0) && (
+        <div class="pros-cons-table">
+          <div class="pros-col">
+            {show.pros?.map((p) => <span class="pro-item">+ {p}</span>)}
+          </div>
+          <div class="cons-col">
+            {show.cons?.map((c) => <span class="con-item">- {c}</span>)}
+          </div>
+        </div>
+      )}
+
       <div class="card-feedback">
         <div class="star-rating">
-          {[1, 2, 3].map((star) => (
+          {[1, 2, 3].map((star) => {
+            const effective = hoverStar || selectedStars;
+            return (
               <button
                 key={star}
-              class={`star-btn${selectedStars >= star ? ' star-active' : ''}`}
+                class={`star-btn${effective >= star ? ' star-active' : ''}`}
                 onClick={() => handleStarClick(star)}
+                onMouseEnter={() => setHoverStar(star)}
+                onMouseLeave={() => setHoverStar(0)}
                 aria-label={`Rate ${star} star${star > 1 ? 's' : ''}`}
               >
-              {selectedStars >= star ? '★' : '☆'}
+                {effective >= star ? '★' : '☆'}
               </button>
-          ))}
+            );
+          })}
           {submitted && <span class="feedback-saved">Saved</span>}
         </div>
 

packages/frontend/src/components/Sidebar.css

@@ -2,16 +2,20 @@
 
 .sidebar {
   width: var(--sidebar-width);
-  background: var(--bg-surface);
+  background:
-  border-right: 1px solid var(--border);
+    linear-gradient(180deg, rgba(255, 255, 255, 0.04), transparent 100%),
+    rgba(28, 25, 42, 0.82);
+  backdrop-filter: blur(20px) saturate(160%);
+  border-right: 1px solid rgba(255, 255, 255, 0.07);
   display: flex;
   flex-direction: column;
   overflow: hidden;
   flex-shrink: 0;
+  box-shadow: inset -1px 0 0 rgba(255, 255, 255, 0.02);
 }
 
 .sidebar-header {
-  padding: 20px 16px 12px;
+  padding: var(--sp-5) var(--sp-4) var(--sp-3);
   border-bottom: 1px solid var(--border);
   text-align: center;
 }
@@ -20,7 +24,7 @@
   font-size: 24px;
   font-weight: 700;
   color: var(--text);
-  letter-spacing: -0.3px;
+  letter-spacing: -0.03em;
   text-decoration: none;
   display: inline-block;
   transition: color 0.15s ease;
@@ -32,28 +36,31 @@
 }
 
 .btn-new {
-  margin: 12px;
+  margin: var(--sp-3);
   padding: 10px 14px;
   background: var(--accent);
-  color: #fff;
+  color: #1a141f;
   border: none;
   border-radius: var(--radius);
   font-size: 13px;
   font-weight: 600;
   cursor: pointer;
-  transition: background 0.15s;
+  transition: background 0.15s ease, transform 0.15s ease, box-shadow 0.15s ease;
   text-align: left;
+  box-shadow: 0 8px 24px rgba(232, 168, 124, 0.18);
 }
 
 .btn-new:hover {
   background: var(--accent-hover);
+  transform: translateY(-1px);
+  box-shadow: 0 12px 30px rgba(232, 168, 124, 0.22);
 }
 
 .sidebar-section-label {
   padding: 8px 16px 4px;
   font-size: 11px;
   font-weight: 600;
-  letter-spacing: 0.06em;
+  letter-spacing: 0.08em;
   text-transform: uppercase;
   color: var(--text-dim);
 }
@@ -76,21 +83,27 @@
   align-items: center;
   gap: 8px;
   padding: 8px 10px;
-  border-radius: var(--radius-sm);
+  border-radius: 10px;
   cursor: pointer;
   color: var(--text-muted);
   font-size: 13px;
-  transition: background 0.1s, color 0.1s;
+  border: 1px solid transparent;
+  transition: background 0.15s ease, color 0.15s ease, border-color 0.15s ease, transform 0.15s ease;
   user-select: none;
 }
 
 .sidebar-item:hover {
-  background: var(--bg-surface-2);
+  background: rgba(255, 255, 255, 0.05);
+  border-color: rgba(255, 255, 255, 0.06);
   color: var(--text);
+  transform: translateX(1px);
 }
 
 .sidebar-item.selected {
-  background: var(--accent-dim);
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.04), transparent 100%),
+    rgba(232, 168, 124, 0.12);
+  border-color: rgba(232, 168, 124, 0.22);
   color: var(--text);
 }
 
@@ -121,4 +134,55 @@
   overflow: hidden;
   text-overflow: ellipsis;
   white-space: nowrap;
+  flex: 1;
+}
+
+.sidebar-type-badge {
+  flex-shrink: 0;
+  font-size: 10px;
+  font-weight: 600;
+  padding: 2px 5px;
+  border-radius: var(--radius-xs);
+  text-transform: uppercase;
+  letter-spacing: 0.04em;
+}
+
+.sidebar-type-tv_show {
+  background: rgba(99, 102, 241, 0.15);
+  color: #818cf8;
+}
+
+.sidebar-type-movie {
+  background: rgba(245, 158, 11, 0.15);
+  color: #fbbf24;
+}
+
+.sidebar-item-delete {
+  flex-shrink: 0;
+  background: none;
+  border: none;
+  color: var(--text-dim);
+  cursor: pointer;
+  font-size: 14px;
+  padding: 0 2px;
+  line-height: 1;
+  opacity: 0;
+  transition: opacity 0.15s, color 0.15s;
+}
+
+.sidebar-item:hover .sidebar-item-delete {
+  opacity: 1;
+}
+
+.sidebar-item-delete:hover {
+  color: var(--red);
+}
+
+@media (max-width: 720px) {
+  .sidebar {
+    width: 100%;
+    max-height: 42vh;
+    border-right: none;
+    border-bottom: 1px solid rgba(255, 255, 255, 0.07);
+  }
 }

packages/frontend/src/components/Sidebar.tsx

@@ -1,4 +1,5 @@
 import './Sidebar.css';
+import { route } from 'preact-router';
 import type { RecommendationSummary } from '../types/index.js';
 
 interface SidebarProps {
@@ -30,10 +31,27 @@ export function Sidebar({ list, selectedId, onSelect, onNewClick }: SidebarProps
   return (
     <aside class="sidebar">
       <div class="sidebar-header">
-        <a href="/" class="sidebar-title">Recommender</a>
+        <a
+          href="/"
+          class="sidebar-title"
+          onClick={(event) => {
+            event.preventDefault();
+            route('/');
+          }}
+        >
+          Recommender
+        </a>
       </div>
 
-      <button class="btn-new" onClick={onNewClick}>
+      <button
+        type="button"
+        class="btn-new"
+        onClick={(event) => {
+          event.preventDefault();
+          event.stopPropagation();
+          onNewClick();
+        }}
+      >
         + New Recommendation
       </button>
 
@@ -53,6 +71,9 @@ export function Sidebar({ list, selectedId, onSelect, onNewClick }: SidebarProps
               {statusIcon(item.status)}
             </span>
             <span class="sidebar-item-title">{item.title}</span>
+            <span class={`sidebar-type-badge sidebar-type-${item.media_type}`}>
+              {item.media_type === 'movie' ? 'Film' : 'TV'}
+            </span>
           </li>
         ))}
       </ul>

packages/frontend/src/hooks/useRecommendations.ts

@@ -1,13 +1,37 @@
 import { useState, useEffect, useCallback } from 'preact/hooks';
-import type { RecommendationSummary, FeedbackEntry } from '../types/index.js';
+import type { MediaType, RecommendationSummary, FeedbackEntry } from '../types/index.js';
 import {
   listRecommendations,
   createRecommendation,
+  createContinuousRecommendation,
   rerankRecommendation,
   submitFeedback,
   getFeedback,
+  deleteRecommendation,
 } from '../api/client.js';
 
+type GenerationMode = 'brainstorm' | 'continuous';
+
+interface CreateBody {
+  main_prompt: string;
+  liked_series: string;
+  disliked_series: string;
+  themes: string;
+  requirements?: string;
+  avoid?: string;
+  brainstorm_count?: number;
+  media_type: MediaType;
+  use_web_search?: boolean;
+  use_validator?: boolean;
+  hard_requirements?: boolean;
+  self_expansive?: boolean;
+  expansive_passes?: number;
+  expansive_mode?: 'soft' | 'extreme';
+  generation_mode?: GenerationMode;
+  total_count?: number;
+  validate_results?: boolean;
+}
+
 export function useRecommendations() {
   const [list, setList] = useState<RecommendationSummary[]>([]);
   const [selectedId, setSelectedId] = useState<string | null>(null);
@@ -29,14 +53,27 @@ export function useRecommendations() {
   }, [refreshList, refreshFeedback]);
 
   const createNew = useCallback(
-    async (body: {
+    async (body: CreateBody) => {
-      main_prompt: string;
+      let id = '';
-      liked_shows: string;
+
-      disliked_shows: string;
+      if (body.generation_mode === 'continuous') {
-      themes: string;
+        const result = await createContinuousRecommendation({
-      brainstorm_count?: number;
+          liked_series: body.liked_series,
-    }) => {
+          disliked_series: body.disliked_series,
-      const { id } = await createRecommendation(body);
+          themes: body.themes,
+          requirements: body.requirements ?? '',
+          avoid: body.avoid ?? '',
+          total_count: body.total_count ?? 30,
+          media_type: body.media_type,
+          use_web_search: body.use_web_search,
+          validate_results: body.validate_results,
+        });
+        id = result.id;
+      } else {
+        const result = await createRecommendation(body);
+        id = result.id;
+      }
+
       await refreshList();
       setSelectedId(id);
       return id;
@@ -47,7 +84,6 @@ export function useRecommendations() {
   const rerank = useCallback(
     async (id: string) => {
       await rerankRecommendation(id);
-      // Update local list to show pending status
       setList((prev) =>
         prev.map((r) => (r.id === id ? { ...r, status: 'pending' as const } : r)),
       );
@@ -56,7 +92,7 @@ export function useRecommendations() {
   );
 
   const handleSubmitFeedback = useCallback(
-    async (body: { tv_show_name: string; stars: number; feedback?: string }) => {
+    async (body: { item_name: string; stars: number; feedback?: string }) => {
       await submitFeedback(body);
       await refreshFeedback();
     },
@@ -70,6 +106,15 @@ export function useRecommendations() {
     [],
   );
 
+  const updateTitle = useCallback((id: string, title: string) => {
+    setList((prev) => prev.map((r) => (r.id === id ? { ...r, title } : r)));
+  }, []);
+
+  const deleteRec = useCallback(async (id: string) => {
+    await deleteRecommendation(id);
+    setList((prev) => prev.filter((r) => r.id !== id));
+  }, []);
+
   return {
     list,
     selectedId,
@@ -79,6 +124,8 @@ export function useRecommendations() {
     rerank,
     submitFeedback: handleSubmitFeedback,
     updateStatus,
+    updateTitle,
     refreshList,
+    deleteRec,
   };
 }

packages/frontend/src/hooks/useSSE.ts

@@ -4,10 +4,13 @@ import type { SSEEvent } from '../types/index.js';
 export function useSSE(
   url: string | null,
   onEvent: (event: SSEEvent) => void,
+  onClose?: () => void,
 ): { close: () => void } {
   const esRef = useRef<EventSource | null>(null);
   const onEventRef = useRef(onEvent);
   onEventRef.current = onEvent;
+  const onCloseRef = useRef(onClose);
+  onCloseRef.current = onClose;
 
   useEffect(() => {
     if (!url) return;
@@ -32,6 +35,7 @@ export function useSSE(
     es.onerror = () => {
       es.close();
       esRef.current = null;
+      onCloseRef.current?.();
     };
 
     return () => {

packages/frontend/src/index.css

@@ -7,29 +7,47 @@
 }
 
 :root {
-  --bg: #0f1117;
+  --bg: #13111a;
-  --bg-surface: #1a1d27;
+  --bg-surface: rgba(28, 25, 42, 0.72);
-  --bg-surface-2: #222535;
+  --bg-surface-2: rgba(36, 33, 56, 0.65);
-  --bg-surface-3: #2a2e42;
+  --bg-surface-3: rgba(48, 45, 72, 0.6);
-  --border: #2e3248;
+  --border: rgba(245, 240, 232, 0.08);
-  --text: #e2e4ed;
+  --border-strong: rgba(245, 240, 232, 0.16);
-  --text-muted: #8a8fa8;
+  --text: #f5f0e8;
-  --text-dim: #5a5f77;
+  --text-muted: #a09890;
-  --accent: #6366f1;
+  --text-dim: #686070;
-  --accent-hover: #4f52d9;
+  --accent: #e8a87c;
-  --accent-dim: rgba(99, 102, 241, 0.15);
+  --accent-hover: #d4956a;
+  --accent-dim: rgba(232, 168, 124, 0.12);
   --green: #22c55e;
   --blue: #3b82f6;
   --yellow: #eab308;
   --red: #ef4444;
-  --radius: 8px;
+  --magenta: #d946ef;
-  --radius-sm: 4px;
+  --radius-xs: 4px;
+  --radius: 12px;
+  --radius-md: 16px;
+  --radius-lg: 20px;
+  --radius-xl: 28px;
+  --radius-pill: 999px;
+  --sp-1: 4px;
+  --sp-2: 8px;
+  --sp-3: 12px;
+  --sp-4: 16px;
+  --sp-5: 20px;
+  --sp-6: 24px;
+  --sp-8: 32px;
+  --sp-10: 40px;
+  --sp-12: 48px;
   --sidebar-width: 260px;
-  font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
+  font-family: 'Sora', -apple-system, BlinkMacSystemFont, 'Segoe UI', Roboto, sans-serif;
   font-size: 14px;
   color: var(--text);
   background: var(--bg);
   -webkit-font-smoothing: antialiased;
+  -moz-osx-font-smoothing: grayscale;
+  font-weight: 400;
+  letter-spacing: -0.01em;
 }
 
 html,
@@ -40,6 +58,43 @@ body,
   overflow: hidden;
 }
 
+body {
+  position: relative;
+  color: var(--text);
+  background:
+    radial-gradient(ellipse 80% 60% at 15% 10%, rgba(232, 168, 124, 0.07), transparent 55%),
+    radial-gradient(ellipse 70% 50% at 85% 90%, rgba(245, 240, 232, 0.05), transparent 50%),
+    var(--bg);
+}
+
+body::before {
+  content: '';
+  position: fixed;
+  inset: 0;
+  z-index: -1;
+  pointer-events: none;
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.02), transparent 40%),
+    radial-gradient(circle at top left, rgba(255, 214, 170, 0.05), transparent 28%);
+}
+
+a {
+  color: inherit;
+}
+
+button,
+input,
+textarea {
+  font: inherit;
+}
+
+button:focus-visible,
+input:focus-visible,
+textarea:focus-visible {
+  outline: none;
+  box-shadow: 0 0 0 3px rgba(232, 168, 124, 0.12);
+}
+
 /* ── Layout ────────────────────────────────────────────── */
 
 .layout {
@@ -51,7 +106,7 @@ body,
 .main-content {
   flex: 1;
   overflow-y: auto;
-  padding: 32px;
+  padding: var(--sp-8);
 }
 
 /* ── Shared content areas ───────────────────────────────── */
@@ -62,7 +117,7 @@ body,
   align-items: center;
   justify-content: center;
   height: 100%;
-  gap: 12px;
+  gap: var(--sp-3);
   color: var(--text-muted);
   text-align: center;
 }
@@ -73,26 +128,56 @@ body,
   color: var(--text);
 }
 
+.surface-glass {
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.04), transparent 100%),
+    var(--bg-surface);
+  backdrop-filter: blur(18px) saturate(160%);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-lg);
+  box-shadow:
+    0 8px 40px rgba(0, 0, 0, 0.28),
+    inset 0 1px 0 rgba(255, 255, 255, 0.04);
+}
+
+.label-eyebrow {
+  font-size: 11px;
+  font-weight: 600;
+  letter-spacing: 0.08em;
+  text-transform: uppercase;
+  color: var(--text-dim);
+}
+
 /* ── Buttons ────────────────────────────────────────────── */
 
 .btn-primary {
   padding: 10px 20px;
   background: var(--accent);
-  color: #fff;
+  color: #1a141f;
   border: none;
   border-radius: var(--radius);
   font-size: 14px;
   font-weight: 600;
   cursor: pointer;
-  transition: background 0.15s;
+  transition:
+    background 0.15s ease,
+    transform 0.15s ease,
+    box-shadow 0.15s ease;
 }
 
 .btn-primary:hover:not(:disabled) {
   background: var(--accent-hover);
+  transform: translateY(-1px);
+  box-shadow: 0 6px 20px rgba(232, 168, 124, 0.28);
+}
+
+.btn-primary:active:not(:disabled) {
+  transform: translateY(0);
+  box-shadow: 0 2px 8px rgba(232, 168, 124, 0.15);
 }
 
 .btn-primary:disabled {
-  opacity: 0.5;
+  opacity: 0.45;
   cursor: not-allowed;
 }
 
@@ -103,16 +188,154 @@ body,
 
 .btn-secondary {
   padding: 10px 20px;
-  background: var(--bg-surface-2);
+  background: rgba(255, 255, 255, 0.05);
   color: var(--text-muted);
   border: 1px solid var(--border);
   border-radius: var(--radius);
   font-size: 14px;
   font-weight: 500;
   cursor: pointer;
-  transition: background 0.15s;
+  transition:
+    background 0.15s ease,
+    border-color 0.15s ease,
+    color 0.15s ease;
 }
 
 .btn-secondary:hover:not(:disabled) {
-  background: var(--bg-surface-3);
+  background: rgba(255, 255, 255, 0.09);
+  border-color: var(--border-strong);
+  color: var(--text);
+}
+
+.btn-ghost {
+  padding: 8px 16px;
+  background: transparent;
+  color: var(--text-dim);
+  border: 1px solid rgba(255, 255, 255, 0.06);
+  border-radius: var(--radius);
+  font-size: 13px;
+  font-weight: 400;
+  cursor: pointer;
+  transition: color 0.15s ease, border-color 0.15s ease;
+}
+
+.btn-ghost:hover:not(:disabled) {
+  color: var(--text-muted);
+  border-color: rgba(255, 255, 255, 0.12);
+}
+
+.btn-icon {
+  display: inline-flex;
+  align-items: center;
+  justify-content: center;
+  width: 36px;
+  height: 36px;
+  background: rgba(255, 255, 255, 0.04);
+  border: 1px solid rgba(255, 255, 255, 0.08);
+  border-radius: 10px;
+  color: var(--text-muted);
+  cursor: pointer;
+  font-size: 16px;
+  transition:
+    background 0.15s ease,
+    border-color 0.15s ease,
+    color 0.15s ease,
+    transform 0.15s ease;
+}
+
+.btn-icon:hover:not(:disabled) {
+  background: rgba(255, 255, 255, 0.08);
+  border-color: rgba(255, 255, 255, 0.14);
+  color: var(--text);
+  transform: translateY(-1px);
+}
+
+.btn-danger {
+  padding: 8px 16px;
+  background: transparent;
+  color: #f87171;
+  border: 1px solid rgba(239, 68, 68, 0.3);
+  border-radius: var(--radius);
+  font-size: 13px;
+  font-weight: 600;
+  cursor: pointer;
+  transition: background 0.15s ease, border-color 0.15s ease;
+}
+
+.btn-danger:hover:not(:disabled) {
+  background: rgba(239, 68, 68, 0.1);
+  border-color: rgba(239, 68, 68, 0.5);
+}
+
+input[type='text'],
+input[type='search'],
+textarea {
+  background: rgba(18, 16, 26, 0.85);
+  border: 1px solid rgba(255, 255, 255, 0.1);
+  border-radius: var(--radius);
+  color: var(--text);
+  padding: 10px 14px;
+  font-size: 14px;
+  transition: border-color 0.15s ease, box-shadow 0.15s ease, background 0.15s ease;
+  outline: none;
+}
+
+input[type='text']:focus,
+input[type='search']:focus,
+textarea:focus {
+  border-color: rgba(232, 168, 124, 0.5);
+  box-shadow: 0 0 0 3px rgba(232, 168, 124, 0.12);
+}
+
+input::placeholder,
+textarea::placeholder {
+  color: var(--text-dim);
+}
+
+input[type='range'] {
+  -webkit-appearance: none;
+  appearance: none;
+  height: 4px;
+  background: rgba(255, 255, 255, 0.1);
+  border-radius: 100px;
+  outline: none;
+}
+
+input[type='range']::-webkit-slider-thumb {
+  -webkit-appearance: none;
+  width: 16px;
+  height: 16px;
+  background: var(--accent);
+  border-radius: 50%;
+  cursor: pointer;
+  box-shadow: 0 0 0 3px rgba(232, 168, 124, 0.2);
+  transition: box-shadow 0.15s ease;
+}
+
+input[type='range']::-webkit-slider-thumb:hover {
+  box-shadow: 0 0 0 5px rgba(232, 168, 124, 0.3), 0 0 12px rgba(232, 168, 124, 0.3);
+}
+
+@media (max-width: 860px) {
+  .main-content {
+    padding: var(--sp-6);
+  }
+}
+
+@media (max-width: 720px) {
+  html,
+  body,
+  #app {
+    overflow: auto;
+  }
+
+  .layout {
+    flex-direction: column;
+    height: auto;
+    min-height: 100vh;
+  }
+
+  .main-content {
+    padding: var(--sp-5);
+  }
 }

packages/frontend/src/pages/Home.css

@@ -5,22 +5,30 @@
   height: 100vh;
   position: relative;
   overflow: hidden;
+  isolation: isolate;
 }
 
 .landing-bg {
   position: absolute;
-  inset: -6px;
+  inset: -10px;
-  /* overflow slightly to hide blur edges */
   background: url('/wallpaper.png') center / cover no-repeat;
-  filter: blur(3px) brightness(0.35);
+  filter: blur(5px) brightness(0.28) saturate(0.85);
   z-index: 0;
 }
 
+.landing-layout::after {
+  content: '';
+  position: absolute;
+  inset: 0;
+  background:
+    linear-gradient(90deg, rgba(10, 8, 15, 0.34) 0%, rgba(10, 8, 15, 0.08) 32%, rgba(10, 8, 15, 0.28) 100%);
+  z-index: 0;
+  pointer-events: none;
+}
+
 .landing-layout .sidebar {
   position: relative;
   z-index: 1;
-  background: rgba(26, 29, 39, 0.85);
-  backdrop-filter: blur(8px);
 }
 
 .landing-main {
@@ -33,41 +41,67 @@
   text-align: center;
   position: relative;
   z-index: 1;
-  padding: 40px;
+  padding: var(--sp-10);
 }
 
 .landing-title {
-  font-size: 5rem;
+  font-size: clamp(3.8rem, 9vw, 5.6rem);
-  font-weight: 800;
+  font-weight: 300;
-  color: #fff;
+  color: var(--text);
-  letter-spacing: -2px;
+  letter-spacing: -0.03em;
   line-height: 1;
+  text-shadow: 0 18px 42px rgba(0, 0, 0, 0.28);
 }
 
 .landing-tagline {
-  font-size: 1.4rem;
+  max-width: 620px;
-  color: rgba(255, 255, 255, 0.6);
+  font-size: clamp(1rem, 2vw, 1.3rem);
+  color: rgba(245, 240, 232, 0.7);
   font-weight: 400;
+  line-height: 1.7;
 }
 
 .btn-gradient {
-  padding: 16px 42px;
+  padding: 14px 30px;
-  font-size: 1.20rem;
+  font-size: 1rem;
-  font-weight: 700;
+  font-weight: 600;
-  color: rgba(255, 255, 255, 0.9);
+  color: #1a141f;
-  border: 1px solid rgba(255, 255, 255, 0.1);
+  border: 1px solid rgba(255, 244, 232, 0.14);
-  border-radius: 12px;
+  border-radius: var(--radius);
   cursor: pointer;
-  background: linear-gradient(135deg, rgba(113, 88, 226, 0.8), rgba(61, 59, 243, 0.8));
+  background:
-  backdrop-filter: blur(6px);
+    linear-gradient(180deg, rgba(255, 255, 255, 0.12), transparent 100%),
-  box-shadow: 0 4px 16px rgba(0, 0, 0, 0.2), inset 0 1px 1px rgba(255, 255, 255, 0.1);
+    linear-gradient(135deg, rgba(232, 168, 124, 0.96), rgba(214, 149, 106, 0.94));
-  transition: all 0.6s cubic-bezier(0.25, 0.8, 0.25, 1);
+  backdrop-filter: blur(10px);
-  letter-spacing: 0.02em;
+  box-shadow:
+    0 10px 30px rgba(232, 168, 124, 0.22),
+    inset 0 1px 0 rgba(255, 255, 255, 0.22);
+  transition:
+    transform 0.2s ease,
+    box-shadow 0.2s ease,
+    background 0.2s ease;
+  letter-spacing: -0.01em;
 }
 
 .btn-gradient:hover {
   transform: translateY(-2px);
-  color: #fff;
+  box-shadow:
-  background: linear-gradient(135deg, rgba(113, 88, 226, 0.95), rgba(41, 121, 255, 0.95));
+    0 16px 38px rgba(232, 168, 124, 0.28),
-  box-shadow: 0 8px 24px rgba(113, 88, 226, 0.3), inset 0 1px 2px rgba(255, 255, 255, 0.2);
+    inset 0 1px 0 rgba(255, 255, 255, 0.24);
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.14), transparent 100%),
+    linear-gradient(135deg, rgba(232, 168, 124, 1), rgba(214, 149, 106, 0.98));
+}
+
+@media (max-width: 720px) {
+  .landing-layout {
+    flex-direction: column;
+    min-height: 100vh;
+    height: auto;
+  }
+
+  .landing-main {
+    min-height: 55vh;
+    padding: var(--sp-8) var(--sp-5);
+  }
 }

packages/frontend/src/pages/Home.tsx

@@ -13,13 +13,27 @@ export function Home() {
 
   const handleCreateNew = async (body: {
     main_prompt: string;
-    liked_shows: string;
+    liked_series: string;
-    disliked_shows: string;
+    disliked_series: string;
     themes: string;
+    requirements?: string;
+    avoid?: string;
     brainstorm_count?: number;
+    media_type: import('../types/index.js').MediaType;
+    use_web_search?: boolean;
+    use_validator?: boolean;
+    hard_requirements?: boolean;
+    self_expansive?: boolean;
+    expansive_passes?: number;
+    expansive_mode?: 'soft' | 'extreme';
+    generation_mode?: 'brainstorm' | 'continuous';
+    total_count?: number;
+    validate_results?: boolean;
   }) => {
     const id = await createNew(body);
+    if (id) {
       route(`/recom/${id}`);
+    }
   };
 
   return (
@@ -33,7 +47,7 @@ export function Home() {
       />
       <main class="landing-main">
         <h1 class="landing-title">Recommender</h1>
-        <p class="landing-tagline">Discover your next favorite show, powered by AI.</p>
+        <p class="landing-tagline">Discover your next favorite show or movie, powered by AI.</p>
         <button class="btn-gradient" onClick={() => setShowModal(true)}>
           Get Started →
         </button>

packages/frontend/src/pages/Recom.css

@@ -16,6 +16,131 @@
   margin-bottom: 24px;
 }
 
+/* ── Request Info Panel ─────────────────────────────────── */
+
+.rec-info-panel {
+  background:
+    linear-gradient(180deg, rgba(255, 255, 255, 0.04), transparent 100%),
+    var(--bg-surface);
+  backdrop-filter: blur(18px) saturate(160%);
+  border: 1px solid var(--border);
+  border-radius: var(--radius-lg);
+  margin-bottom: var(--sp-6);
+  overflow: hidden;
+  box-shadow:
+    0 12px 34px rgba(0, 0, 0, 0.2),
+    inset 0 1px 0 rgba(255, 255, 255, 0.04);
+}
+
+.rec-info-toggle {
+  width: 100%;
+  display: flex;
+  align-items: center;
+  justify-content: space-between;
+  padding: 14px 16px;
+  background: none;
+  border: none;
+  cursor: pointer;
+  text-align: left;
+  gap: 8px;
+}
+
+.rec-info-toggle:hover {
+  background: rgba(255, 255, 255, 0.04);
+}
+
+.rec-info-title {
+  font-size: 15px;
+  font-weight: 700;
+  color: var(--text);
+}
+
+.rec-info-chevron {
+  font-size: 10px;
+  color: var(--text-dim);
+  flex-shrink: 0;
+}
+
+.rec-info-body {
+  padding: 0 16px 14px;
+  display: flex;
+  flex-direction: column;
+  gap: 10px;
+  border-top: 1px solid var(--border);
+  padding-top: 14px;
+}
+
+.rec-info-row {
+  display: flex;
+  gap: 12px;
+  align-items: baseline;
+}
+
+.rec-info-label {
+  font-size: 11px;
+  font-weight: 600;
+  text-transform: uppercase;
+  letter-spacing: 0.08em;
+  color: var(--text-dim);
+  flex-shrink: 0;
+  width: 72px;
+}
+
+.rec-info-value {
+  font-size: 14px;
+  color: var(--text-muted);
+  line-height: 1.6;
+}
+
+.rec-info-badge {
+  font-size: 11px;
+  font-weight: 600;
+  padding: 2px 8px;
+  border-radius: 100px;
+  background: rgba(232, 168, 124, 0.12);
+  color: var(--accent);
+}
+
+.rec-info-delete-row {
+  padding-top: 4px;
+  align-items: center;
+}
+
+/* Shared sizing for the two action buttons in the info panel */
+.btn-danger,
+.rec-info-delete-row .btn-rerun {
+  padding: 5px 14px;
+  font-size: 12px;
+  font-weight: 600;
+  border-radius: var(--radius);
+  border: 1px solid;
+  background: none;
+  cursor: pointer;
+  transition: background 0.15s, border-color 0.15s;
+}
+
+.btn-danger {
+  border-color: rgba(239, 68, 68, 0.4);
+  color: #f87171;
+}
+
+.btn-danger:hover {
+  background: rgba(239, 68, 68, 0.12);
+  border-color: #f87171;
+}
+
+/* Re-run button — same shape as Delete, neutral tone */
+.rec-info-delete-row .btn-rerun {
+  border-color: var(--border);
+  color: var(--text-muted);
+}
+
+.rec-info-delete-row .btn-rerun:hover {
+  background: rgba(255, 255, 255, 0.08);
+  border-color: var(--border-strong);
+  color: var(--text);
+}
+
 .error-state {
   color: var(--text-muted);
 }
@@ -28,3 +153,23 @@
 .error-state p {
   margin-bottom: 16px;
 }
+
+@media (max-width: 720px) {
+  .content-area {
+    max-width: 100%;
+  }
+
+  .rec-info-row {
+    flex-direction: column;
+    gap: 4px;
+    align-items: flex-start;
+  }
+
+  .rec-info-label {
+    width: auto;
+  }
+
+  .rec-info-delete-row {
+    gap: 10px;
+  }
+}

packages/frontend/src/pages/Recom.tsx

@@ -2,46 +2,92 @@ import { useState, useCallback, useEffect } from 'preact/hooks';
 import './Recom.css';
 import { route } from 'preact-router';
 import { PipelineProgress } from '../components/PipelineProgress.js';
+import type { StageGroup } from '../components/PipelineProgress.js';
 import { RecommendationCard } from '../components/RecommendationCard.js';
 import { Sidebar } from '../components/Sidebar.js';
 import { NewRecommendationModal } from '../components/NewRecommendationModal.js';
 import { useRecommendationsContext } from '../context/RecommendationsContext.js';
 import { useSSE } from '../hooks/useSSE.js';
 import { getRecommendation } from '../api/client.js';
-import type { Recommendation, SSEEvent, StageMap, PipelineStage } from '../types/index.js';
+import type { Recommendation, SSEEvent, StageMap, StageStatus } from '../types/index.js';
 
 interface RecomProps {
   id: string;
   path?: string;
 }
 
-const DEFAULT_STAGES: StageMap = {
+function buildDefaultStages(rec: Recommendation | null): StageMap {
+  const map: StageMap = {
     interpreter: 'pending',
     retrieval: 'pending',
+    validator: 'pending',
     ranking: 'pending',
     curator: 'pending',
-};
+  };
+  if (rec?.self_expansive && rec.expansive_passes > 0) {
+    for (let i = 0; i < rec.expansive_passes; i++) {
+      const p = i + 2;
+      map[`pass${p}:retrieval`] = 'pending';
+      if (rec.use_validator) map[`pass${p}:validator`] = 'pending';
+      map[`pass${p}:ranking`] = 'pending';
+      map[`pass${p}:curator`] = 'pending';
+    }
+  }
+  return map;
+}
 
-const STAGE_ORDER: (keyof StageMap)[] = ['interpreter', 'retrieval', 'ranking', 'curator'];
+function buildStageGroups(rec: Recommendation | null): StageGroup[] {
+  const baseStages = [
+    { key: 'interpreter', label: 'Interpreting Preferences' },
+    { key: 'retrieval', label: 'Generating Candidates' },
+    { key: 'validator', label: 'Validating Candidates' },
+    { key: 'ranking', label: 'Ranking Candidates' },
+    { key: 'curator', label: 'Curating Results' },
+  ];
+  const groups: StageGroup[] = [{ label: '', stages: baseStages }];
+  if (rec?.self_expansive && rec.expansive_passes > 0) {
+    for (let i = 0; i < rec.expansive_passes; i++) {
+      const p = i + 2;
+      groups.push({
+        label: `Pass ${p}`,
+        stages: [
+          { key: `pass${p}:retrieval`, label: 'Generating Candidates' },
+          ...(rec.use_validator ? [{ key: `pass${p}:validator`, label: 'Validating Candidates' }] : []),
+          { key: `pass${p}:ranking`, label: 'Ranking Candidates' },
+          { key: `pass${p}:curator`, label: 'Curating Results' },
+        ],
+      });
+    }
+  }
+  return groups;
+}
 
 export function Recom({ id }: RecomProps) {
-  const { list, feedback, submitFeedback, rerank, updateStatus, refreshList, createNew } = useRecommendationsContext();
+  const { list, feedback, submitFeedback, rerank, updateStatus, updateTitle, refreshList, createNew, deleteRec } = useRecommendationsContext();
 
   const [rec, setRec] = useState<Recommendation | null>(null);
-  const [stages, setStages] = useState<StageMap>(DEFAULT_STAGES);
+  const [stages, setStages] = useState<StageMap>(buildDefaultStages(null));
-  const [sseUrl, setSseUrl] = useState<string | null>(null);
+  // sseKey drives the SSE connection. null = inactive; a number = active.
+  // Using a timestamp nonce ensures the URL is always unique on (re)connect,
+  // so useSSE's useEffect always re-runs even if the base path hasn't changed.
+  const [sseKey, setSseKey] = useState<number | null>(null);
   const [showModal, setShowModal] = useState(false);
+  const [infoExpanded, setInfoExpanded] = useState(true);
+
+  // Derive the actual URL from the key; query param is ignored by the server
+  // but makes the string unique so React/Preact state always treats it as changed.
+  const sseUrl = sseKey !== null ? `/api/recommendations/${id}/stream?_k=${sseKey}` : null;
 
   useEffect(() => {
     setRec(null);
-    setStages(DEFAULT_STAGES);
+    setStages(buildDefaultStages(null));
-    setSseUrl(null);
+    setSseKey(null);
     getRecommendation(id)
       .then((data) => {
         setRec(data);
         if (data.status === 'running' || data.status === 'pending') {
-          setStages(DEFAULT_STAGES);
+          setStages(buildDefaultStages(data));
-          setSseUrl(`/api/recommendations/${id}/stream`);
+          setSseKey(Date.now());
         }
       })
       .catch(() => route('/'));
@@ -50,56 +96,117 @@ export function Recom({ id }: RecomProps) {
   const handleSSEEvent = useCallback(
     (event: SSEEvent) => {
       if (event.stage !== 'complete') {
-        const stageKey = event.stage as keyof StageMap;
+        const stageKey = event.stage as string;
-        if (STAGE_ORDER.includes(stageKey)) {
+        setStages((prev) => {
-          setStages((prev) => ({
+          if (!(stageKey in prev)) return prev;
-            ...prev,
+          const eventData = event.data as { skipped?: boolean } | undefined;
-            [stageKey]: event.status === 'start' ? 'running' : event.status === 'done' ? 'done' : 'error',
+          let newStatus: StageStatus;
-          }));
+          if (event.status === 'start') {
+            newStatus = 'running';
+          } else if (event.status === 'done') {
+            newStatus = eventData?.skipped ? 'skipped' : 'done';
+          } else {
+            newStatus = 'error';
           }
+          return { ...prev, [stageKey]: newStatus };
+        });
       }
 
       if (event.stage === 'complete' && event.status === 'done') {
-        setSseUrl(null);
+        const incoming = event.data as { title?: string } | undefined;
+        if (incoming?.title) {
+          updateTitle(id, incoming.title);
+          setRec((prev) => (prev ? { ...prev, title: incoming.title! } : prev));
+        }
+        setSseKey(null);
         updateStatus(id, 'done');
-        void getRecommendation(id).then(setRec);
+        void getRecommendation(id)
+          .then(setRec)
+          .catch(() => {
+            // Fetch failed after completion — poll once more via handleSSEClose logic
+            updateStatus(id, 'done');
+          });
         void refreshList();
       }
 
       if (event.status === 'error') {
-        setSseUrl(null);
+        setSseKey(null);
         updateStatus(id, 'error');
-        const stageKey = event.stage as PipelineStage;
+        setRec((prev) => (prev ? { ...prev, status: 'error' as const } : null));
+        const stageKey = event.stage as string;
         if (stageKey !== 'complete') {
-          setStages((prev) => ({ ...prev, [stageKey as keyof StageMap]: 'error' }));
+          setStages((prev) => (stageKey in prev ? { ...prev, [stageKey]: 'error' } : prev));
         }
       }
     },
-    [id, updateStatus, refreshList],
+    [id, updateStatus, updateTitle, refreshList],
   );
 
-  useSSE(sseUrl, handleSSEEvent);
+  const handleSSEClose = useCallback(() => {
+    void getRecommendation(id).then((data) => {
+      if (data.status === 'done') {
+        setSseKey(null);
+        setRec(data);
+        const allDone = Object.fromEntries(
+          Object.keys(buildDefaultStages(data)).map((k) => [k, 'done' as StageStatus])
+        );
+        setStages(allDone);
+        updateStatus(id, 'done');
+        void refreshList();
+      } else if (data.status === 'error') {
+        setSseKey(null);
+        setRec(data);
+        updateStatus(id, 'error');
+      } else {
+        // Pipeline still running — reconnect with a new unique key so that
+        // useSSE's useEffect always re-fires even if the base URL is the same.
+        setSseKey(Date.now());
+      }
+    });
+  }, [id, updateStatus, refreshList]);
+
+  useSSE(sseUrl, handleSSEEvent, handleSSEClose);
+
+  const handleRetry = async () => {
+    await rerank(id);
+    setStages(buildDefaultStages(rec));
+    setSseKey(Date.now());
+    setRec((prev) => (prev ? { ...prev, status: 'pending' as const } : null));
+  };
 
   const handleRerank = async () => {
     await rerank(id);
-    setStages(DEFAULT_STAGES);
+    setStages(buildDefaultStages(rec));
-    setSseUrl(`/api/recommendations/${id}/stream`);
+    setSseKey(Date.now());
-    setRec((prev) => (prev ? { ...prev, status: 'pending' } : null));
+    setRec((prev) => (prev ? { ...prev, status: 'pending' as const } : null));
   };
 
   const handleCreateNew = async (body: {
     main_prompt: string;
-    liked_shows: string;
+    liked_series: string;
-    disliked_shows: string;
+    disliked_series: string;
     themes: string;
     brainstorm_count?: number;
+    media_type: import('../types/index.js').MediaType;
+    use_web_search?: boolean;
+    use_validator?: boolean;
+    hard_requirements?: boolean;
+    self_expansive?: boolean;
+    expansive_passes?: number;
+    expansive_mode?: 'soft' | 'extreme';
+    generation_mode?: 'brainstorm' | 'continuous';
+    total_count?: number;
+    validate_results?: boolean;
   }) => {
     const newId = await createNew(body);
+    if (newId) {
       route(`/recom/${newId}`);
+    }
   };
 
-  const isRunning = rec?.status === 'running' || rec?.status === 'pending' || !!sseUrl;
+  const isRunning = rec?.status === 'running' || rec?.status === 'pending' || sseKey !== null;
-  const feedbackMap = new Map(feedback.map((f) => [f.tv_show_name, f]));
+  const feedbackMap = new Map(feedback.map((f) => [f.item_name, f]));
+  const stageGroups = buildStageGroups(rec);
 
   return (
     <div class="layout">
@@ -112,23 +219,107 @@ export function Recom({ id }: RecomProps) {
 
       <main class="main-content">
         <div class="recom-content">
+          {rec && (
+            <div class="content-area">
+              <div class="rec-info-panel">
+                <button class="rec-info-toggle" onClick={() => setInfoExpanded((e) => !e)}>
+                  <span class="rec-info-title">{rec.title}</span>
+                  <span class="rec-info-chevron">{infoExpanded ? '▲' : '▼'}</span>
+                </button>
+                {infoExpanded && (
+                  <div class="rec-info-body">
+                    {rec.main_prompt && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Description</span>
+                        <span class="rec-info-value">{rec.main_prompt}</span>
+                      </div>
+                    )}
+                    {rec.liked_series && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Liked</span>
+                        <span class="rec-info-value">{rec.liked_series}</span>
+                      </div>
+                    )}
+                    {rec.disliked_series && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Disliked</span>
+                        <span class="rec-info-value">{rec.disliked_series}</span>
+                      </div>
+                    )}
+                    {rec.themes && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Themes</span>
+                        <span class="rec-info-value">{rec.themes}</span>
+                      </div>
+                    )}
+                    <div class="rec-info-row">
+                      <span class="rec-info-label">Media</span>
+                      <span class="rec-info-value">{rec.media_type === 'tv_show' ? 'TV series' : 'Movies'}</span>
+                    </div>
+                    {rec.use_web_search && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Web Search</span>
+                        <span class="rec-info-badge">enabled</span>
+                      </div>
+                    )}
+                    {rec.use_validator && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Validator</span>
+                        <span class="rec-info-badge">enabled</span>
+                      </div>
+                    )}
+                    {rec.hard_requirements && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Hard Req.</span>
+                        <span class="rec-info-badge">enabled</span>
+                      </div>
+                    )}
+                    {rec.self_expansive && (
+                      <div class="rec-info-row">
+                        <span class="rec-info-label">Self Expansive</span>
+                        <span class="rec-info-badge">{rec.expansive_passes} pass{rec.expansive_passes !== 1 ? 'es' : ''} · {rec.expansive_mode}</span>
+                      </div>
+                    )}
+                    <div class="rec-info-row rec-info-delete-row">
+                      {!isRunning && (
+                        <button class="btn-rerun btn-rerank" onClick={handleRetry}>
+                          Re-run Pipeline
+                        </button>
+                      )}
+                      <button
+                        class="btn-danger"
+                        onClick={() => {
+                          if (confirm('Delete this recommendation?')) {
+                            void deleteRec(id).then(() => route('/'));
+                          }
+                        }}
+                      >
+                        Delete
+                      </button>
+                    </div>
+                  </div>
+                )}
+              </div>
+            </div>
+          )}
+
           {isRunning && (
             <div class="content-area">
-              <PipelineProgress stages={stages} />
+              <PipelineProgress stageGroups={stageGroups} stages={stages} onRetry={handleRetry} />
             </div>
           )}
 
           {!isRunning && rec?.status === 'done' && rec.recommendations && (
             <div class="content-area">
-              <h2 class="rec-title">{rec.title}</h2>
               <div class="cards-grid">
                 {rec.recommendations.map((show) => (
                   <RecommendationCard
                     key={show.title}
                     show={show}
+                    verified={rec.use_validator}
                     existingFeedback={feedbackMap.get(show.title)}
                     onFeedback={async (name, stars, comment) => {
-                      await submitFeedback({ tv_show_name: name, stars, feedback: comment });
+                      await submitFeedback({ item_name: name, stars, feedback: comment });
                     }}
                   />
                 ))}
@@ -149,9 +340,9 @@ export function Recom({ id }: RecomProps) {
           {!isRunning && rec?.status === 'error' && (
             <div class="content-area error-state">
               <h2>Something went wrong</h2>
-              <p>The pipeline encountered an error. You can try again by clicking Re-rank.</p>
+              <p>The pipeline encountered an error. You can try again by clicking the button below.</p>
-              <button class="btn-primary" onClick={handleRerank}>
+              <button class="btn-primary" onClick={handleRetry}>
-                Try Again
+                Re-run Pipeline
               </button>
             </div>
           )}

packages/frontend/src/types/index.ts

@@ -1,9 +1,14 @@
-export type CuratorCategory = 'Definitely Like' | 'Might Like' | 'Questionable' | 'Will Not Like';
+export type MediaType = 'tv_show' | 'movie';
+
+export type CuratorCategory = 'Full Match' | 'Definitely Like' | 'Might Like' | 'Questionable' | 'Will Not Like';
 
 export interface CuratorOutput {
   title: string;
   explanation: string;
   category: CuratorCategory;
+  genre: string;
+  pros: string[];
+  cons: string[];
 }
 
 export type RecommendationStatus = 'pending' | 'running' | 'done' | 'error';
@@ -12,9 +17,16 @@ export interface Recommendation {
   id: string;
   title: string;
   main_prompt: string;
-  liked_shows: string;
+  liked_series: string;
-  disliked_shows: string;
+  disliked_series: string;
   themes: string;
+  media_type: MediaType;
+  use_web_search: boolean;
+  use_validator: boolean;
+  hard_requirements: boolean;
+  self_expansive: boolean;
+  expansive_passes: number;
+  expansive_mode: 'soft' | 'extreme';
   recommendations: CuratorOutput[] | null;
   status: RecommendationStatus;
   created_at: string;
@@ -24,18 +36,30 @@ export interface RecommendationSummary {
   id: string;
   title: string;
   status: RecommendationStatus;
+  media_type: MediaType;
   created_at: string;
 }
 
 export interface FeedbackEntry {
   id: string;
-  tv_show_name: string;
+  item_name: string;
   stars: number;
   feedback: string;
   created_at: string;
 }
 
-export type PipelineStage = 'interpreter' | 'retrieval' | 'ranking' | 'curator' | 'complete';
+export type PipelineStage =
+  | 'interpreter'
+  | 'retrieval'
+  | 'validator'
+  | 'ranking'
+  | 'curator'
+  | 'complete'
+  | `pass${number}:retrieval`
+  | `pass${number}:validator`
+  | `pass${number}:ranking`
+  | `pass${number}:curator`;
+
 export type SSEStatus = 'start' | 'done' | 'error';
 
 export interface SSEEvent {
@@ -44,6 +68,6 @@ export interface SSEEvent {
   data?: unknown;
 }
 
-export type StageStatus = 'pending' | 'running' | 'done' | 'error';
+export type StageStatus = 'pending' | 'running' | 'done' | 'error' | 'skipped';
 
-export type StageMap = Record<Exclude<PipelineStage, 'complete'>, StageStatus>;
+export type StageMap = Record<string, StageStatus>;

renovate.json

@@ -0,0 +1,3 @@
+{
+  "$schema": "https://docs.renovatebot.com/renovate-schema.json"
+}