Add background task persistence for debate & council operations

Decouple debate/council execution from SSE connection lifecycle so tasks
survive browser disconnects. Backend runs work as asyncio.Tasks with
progressive disk persistence; frontend can reconnect and recover state.

- New backend/app/services/tasks.py: task registry, broadcast pattern,
  disk persistence at milestones, stale task cleanup on startup
- New endpoints: POST start_debate/start_council, GET task stream/poll
- Frontend stores taskId on nodes, recovers running tasks on page load
- _applyPartialEvents rebuilds stage text + data from accumulated events

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

2 files changed, 658 insertions, 1 deletions

diff --git a/backend/app/main.py b/backend/app/main.py
index 89c5dd0..746b731 100644
--- a/backend/app/main.py
+++ b/backend/app/main.py
@@ -12,6 +12,10 @@ from app.schemas import NodeRunRequest, NodeRunResponse, MergeStrategy, Role, Me
 from app.services.llm import llm_streamer, generate_title, get_openai_client, get_anthropic_client, resolve_provider
 from app.services.council import council_event_stream
 from app.services.debate import debate_event_stream
+from app.services.tasks import (
+    register_task, get_task, get_task_status, run_task_in_background,
+    cleanup_stale_tasks, TaskStatus, _CHUNK_EVENT_TYPES, _task_registry,
+)
 from app.auth import auth_router, get_current_user, get_current_user_optional, init_db, User, get_db
 from app.auth.utils import get_password_hash
 from dotenv import load_dotenv
@@ -57,7 +61,12 @@ app.add_middleware(
 async def startup_event():
     """Initialize database and create default test user if not exists"""
     init_db()
-    
+
+    # Mark any in-progress tasks from prior run as interrupted
+    stale_count = cleanup_stale_tasks(DATA_ROOT)
+    if stale_count:
+        logger.info("Cleaned up %d stale background tasks", stale_count)
+
     # Create test user if not exists
     from app.auth.models import SessionLocal
     db = SessionLocal()
@@ -711,6 +720,350 @@ async def run_debate_stream(
     )
 
 
+# --------------- Background Task Helpers ---------------
+
+async def _prepare_council_args(request: CouncilRunRequest, resolved: User | None, username: str) -> dict:
+    """Extract shared council request processing into a reusable helper.
+    Returns kwargs suitable for council_event_stream()."""
+    raw_messages = []
+    for ctx in request.incoming_contexts:
+        raw_messages.extend(ctx.messages)
+    if request.merge_strategy == MergeStrategy.SMART:
+        final_messages = smart_merge_messages(raw_messages)
+    else:
+        final_messages = raw_messages
+    execution_context = Context(messages=final_messages)
+
+    images, non_image_file_ids = extract_image_attachments(username, request.attached_file_ids)
+    openrouter_key = get_user_api_key(resolved, "openrouter")
+
+    member_configs: list[LLMConfig] = []
+    attachments_per_model: list[list[dict] | None] = []
+    tools_per_model: list[list[dict] | None] = []
+    contexts_per_model: list[Context | None] = []
+
+    for member in request.council_models:
+        provider = resolve_provider(member.model_name)
+        provider_str = provider.value
+        api_key = get_user_api_key(resolved, provider_str)
+        config = LLMConfig(
+            provider=provider,
+            model_name=member.model_name,
+            temperature=member.temperature if member.temperature is not None else request.temperature,
+            system_prompt=request.system_prompt,
+            api_key=api_key,
+            reasoning_effort=member.reasoning_effort if member.reasoning_effort is not None else request.reasoning_effort,
+            enable_google_search=member.enable_google_search if member.enable_google_search is not None else request.enable_google_search,
+        )
+        member_configs.append(config)
+
+        tools: list[dict] = []
+        attachments: list[dict] = []
+        scoped_file_ids = resolve_scoped_file_ids(username, request.scopes, non_image_file_ids)
+
+        if provider == ModelProvider.OPENAI:
+            vs_ids, debug_refs, filters = await prepare_openai_vector_search(
+                user=username, attached_ids=non_image_file_ids,
+                scopes=request.scopes, llm_config=config,
+            )
+            if not vs_ids:
+                try:
+                    client = get_openai_client(config.api_key)
+                    vs_id = await ensure_user_vector_store(username, client)
+                    if vs_id:
+                        vs_ids = [vs_id]
+                except Exception:
+                    pass
+            if vs_ids:
+                tool_def = {"type": "file_search", "vector_store_ids": vs_ids}
+                if filters:
+                    tool_def["filters"] = filters
+                tools.append(tool_def)
+        elif provider == ModelProvider.GOOGLE:
+            attachments = await prepare_attachments(
+                user=username, target_provider=provider,
+                attached_ids=scoped_file_ids, llm_config=config,
+            )
+        elif provider == ModelProvider.CLAUDE:
+            attachments = await prepare_attachments(
+                user=username, target_provider=provider,
+                attached_ids=scoped_file_ids, llm_config=config,
+            )
+
+        attachments_per_model.append(attachments or None)
+        tools_per_model.append(tools or None)
+
+        if member.incoming_contexts:
+            raw = []
+            for ctx in member.incoming_contexts:
+                raw.extend(ctx.messages)
+            if request.merge_strategy == MergeStrategy.SMART:
+                merged = smart_merge_messages(raw)
+            else:
+                merged = raw
+            contexts_per_model.append(Context(messages=merged))
+        else:
+            contexts_per_model.append(None)
+
+    chairman = request.chairman_model
+    chairman_provider = resolve_provider(chairman.model_name)
+    chairman_api_key = get_user_api_key(resolved, chairman_provider.value)
+    chairman_config = LLMConfig(
+        provider=chairman_provider,
+        model_name=chairman.model_name,
+        temperature=chairman.temperature if chairman.temperature is not None else request.temperature,
+        system_prompt=request.system_prompt,
+        api_key=chairman_api_key,
+        reasoning_effort=chairman.reasoning_effort if chairman.reasoning_effort is not None else request.reasoning_effort,
+        enable_google_search=chairman.enable_google_search if chairman.enable_google_search is not None else request.enable_google_search,
+    )
+
+    return dict(
+        user_prompt=request.user_prompt,
+        context=execution_context,
+        member_configs=member_configs,
+        chairman_config=chairman_config,
+        attachments_per_model=attachments_per_model,
+        tools_per_model=tools_per_model,
+        openrouter_api_key=openrouter_key,
+        images=images,
+        contexts_per_model=contexts_per_model,
+    )
+
+
+async def _prepare_debate_args(request: DebateRunRequest, resolved: User | None, username: str) -> dict:
+    """Extract shared debate request processing into a reusable helper.
+    Returns kwargs suitable for debate_event_stream()."""
+    raw_messages = []
+    for ctx in request.incoming_contexts:
+        raw_messages.extend(ctx.messages)
+    if request.merge_strategy == MergeStrategy.SMART:
+        final_messages = smart_merge_messages(raw_messages)
+    else:
+        final_messages = raw_messages
+    execution_context = Context(messages=final_messages)
+
+    images, non_image_file_ids = extract_image_attachments(username, request.attached_file_ids)
+    openrouter_key = get_user_api_key(resolved, "openrouter")
+
+    member_configs: list[LLMConfig] = []
+    attachments_per_model: list[list[dict] | None] = []
+    tools_per_model: list[list[dict] | None] = []
+
+    for member in request.debate_models:
+        provider = resolve_provider(member.model_name)
+        provider_str = provider.value
+        api_key = get_user_api_key(resolved, provider_str)
+        config = LLMConfig(
+            provider=provider,
+            model_name=member.model_name,
+            temperature=member.temperature if member.temperature is not None else request.temperature,
+            system_prompt=request.system_prompt,
+            api_key=api_key,
+            reasoning_effort=member.reasoning_effort if member.reasoning_effort is not None else request.reasoning_effort,
+            enable_google_search=member.enable_google_search if member.enable_google_search is not None else request.enable_google_search,
+        )
+        member_configs.append(config)
+
+        tools: list[dict] = []
+        attachments: list[dict] = []
+        scoped_file_ids = resolve_scoped_file_ids(username, request.scopes, non_image_file_ids)
+
+        if provider == ModelProvider.OPENAI:
+            vs_ids, debug_refs, filters = await prepare_openai_vector_search(
+                user=username, attached_ids=non_image_file_ids,
+                scopes=request.scopes, llm_config=config,
+            )
+            if not vs_ids:
+                try:
+                    client = get_openai_client(config.api_key)
+                    vs_id = await ensure_user_vector_store(username, client)
+                    if vs_id:
+                        vs_ids = [vs_id]
+                except Exception:
+                    pass
+            if vs_ids:
+                tool_def = {"type": "file_search", "vector_store_ids": vs_ids}
+                if filters:
+                    tool_def["filters"] = filters
+                tools.append(tool_def)
+        elif provider == ModelProvider.GOOGLE:
+            attachments = await prepare_attachments(
+                user=username, target_provider=provider,
+                attached_ids=scoped_file_ids, llm_config=config,
+            )
+        elif provider == ModelProvider.CLAUDE:
+            attachments = await prepare_attachments(
+                user=username, target_provider=provider,
+                attached_ids=scoped_file_ids, llm_config=config,
+            )
+
+        attachments_per_model.append(attachments or None)
+        tools_per_model.append(tools or None)
+
+    judge_config = None
+    if request.judge_mode == DebateJudgeMode.EXTERNAL_JUDGE and request.judge_model:
+        judge = request.judge_model
+        judge_provider = resolve_provider(judge.model_name)
+        judge_api_key = get_user_api_key(resolved, judge_provider.value)
+        judge_config = LLMConfig(
+            provider=judge_provider,
+            model_name=judge.model_name,
+            temperature=judge.temperature if judge.temperature is not None else request.temperature,
+            system_prompt=request.system_prompt,
+            api_key=judge_api_key,
+            reasoning_effort=judge.reasoning_effort if judge.reasoning_effort is not None else request.reasoning_effort,
+            enable_google_search=judge.enable_google_search if judge.enable_google_search is not None else request.enable_google_search,
+        )
+
+    return dict(
+        user_prompt=request.user_prompt,
+        context=execution_context,
+        member_configs=member_configs,
+        judge_config=judge_config,
+        judge_mode=request.judge_mode,
+        debate_format=request.debate_format,
+        max_rounds=request.max_rounds,
+        custom_format_prompt=request.custom_format_prompt,
+        attachments_per_model=attachments_per_model,
+        tools_per_model=tools_per_model,
+        openrouter_api_key=openrouter_key,
+        images=images,
+    )
+
+
+# --------------- Background Task Endpoints ---------------
+
+@app.post("/api/task/start_council")
+async def start_council_task(
+    request: CouncilRunRequest,
+    user: str = DEFAULT_USER,
+    current_user: User | None = Depends(get_current_user_optional),
+):
+    """Start council as a background task. Returns {task_id}."""
+    resolved = resolve_user(current_user, user)
+    username = resolved.username if resolved else DEFAULT_USER
+
+    # Cancel existing task on same node if still running
+    for existing in list(_task_registry.values()):
+        if existing.node_id == request.node_id and existing.user == username and existing.status == TaskStatus.RUNNING:
+            if existing.asyncio_task and not existing.asyncio_task.done():
+                existing.asyncio_task.cancel()
+                logger.info("Cancelled previous task %s for node %s", existing.task_id, request.node_id)
+
+    kwargs = await _prepare_council_args(request, resolved, username)
+    generator = council_event_stream(**kwargs)
+
+    task_id = str(uuid4())
+    info = register_task(task_id, username, request.node_id, "council")
+    info.asyncio_task = asyncio.create_task(run_task_in_background(info, generator))
+    return {"task_id": task_id}
+
+
+@app.post("/api/task/start_debate")
+async def start_debate_task(
+    request: DebateRunRequest,
+    user: str = DEFAULT_USER,
+    current_user: User | None = Depends(get_current_user_optional),
+):
+    """Start debate as a background task. Returns {task_id}."""
+    resolved = resolve_user(current_user, user)
+    username = resolved.username if resolved else DEFAULT_USER
+
+    # Cancel existing task on same node if still running
+    for existing in list(_task_registry.values()):
+        if existing.node_id == request.node_id and existing.user == username and existing.status == TaskStatus.RUNNING:
+            if existing.asyncio_task and not existing.asyncio_task.done():
+                existing.asyncio_task.cancel()
+                logger.info("Cancelled previous task %s for node %s", existing.task_id, request.node_id)
+
+    kwargs = await _prepare_debate_args(request, resolved, username)
+    generator = debate_event_stream(**kwargs)
+
+    task_id = str(uuid4())
+    info = register_task(task_id, username, request.node_id, "debate")
+    info.asyncio_task = asyncio.create_task(run_task_in_background(info, generator))
+    return {"task_id": task_id}
+
+
+@app.get("/api/task/{task_id}/stream")
+async def stream_task_events(
+    task_id: str,
+    from_event: int = 0,
+    user: str = DEFAULT_USER,
+    current_user: User | None = Depends(get_current_user_optional),
+):
+    """SSE stream: replay missed events then stream live."""
+    resolved = resolve_user(current_user, user)
+    username = resolved.username if resolved else DEFAULT_USER
+
+    info = get_task(task_id)
+    if not info:
+        raise HTTPException(status_code=404, detail="Task not found")
+    if info.user != username:
+        raise HTTPException(status_code=403, detail="Not your task")
+
+    async def event_generator():
+        cursor = from_event
+        # 1. Replay accumulated events from cursor
+        while cursor < len(info.events):
+            evt = info.events[cursor]
+            yield f"data: {json.dumps(evt)}\n\n"
+            cursor += 1
+
+        # 2. If already done, send terminal and return
+        if info.status in (TaskStatus.COMPLETED, TaskStatus.FAILED, TaskStatus.INTERRUPTED):
+            yield f"data: {json.dumps({'type': 'task_status', 'data': {'status': info.status.value}})}\n\n"
+            return
+
+        # 3. Stream live events
+        while True:
+            # Wait for new events with timeout (keepalive)
+            try:
+                await asyncio.wait_for(
+                    asyncio.shield(info.new_event_signal.wait()),
+                    timeout=15.0,
+                )
+            except asyncio.TimeoutError:
+                # Send keepalive comment
+                yield ": keepalive\n\n"
+                # Check if task finished while waiting
+                if info.status in (TaskStatus.COMPLETED, TaskStatus.FAILED, TaskStatus.INTERRUPTED):
+                    break
+                continue
+
+            # Yield any new events since cursor
+            while cursor < len(info.events):
+                evt = info.events[cursor]
+                yield f"data: {json.dumps(evt)}\n\n"
+                cursor += 1
+
+            if info.status in (TaskStatus.COMPLETED, TaskStatus.FAILED, TaskStatus.INTERRUPTED):
+                break
+
+        yield f"data: {json.dumps({'type': 'task_status', 'data': {'status': info.status.value}})}\n\n"
+
+    return StreamingResponse(event_generator(), media_type="text/event-stream")
+
+
+@app.get("/api/task/{task_id}")
+async def poll_task(
+    task_id: str,
+    user: str = DEFAULT_USER,
+    current_user: User | None = Depends(get_current_user_optional),
+):
+    """Poll: return status + accumulated results."""
+    resolved = resolve_user(current_user, user)
+    username = resolved.username if resolved else DEFAULT_USER
+
+    status_data = get_task_status(username, task_id)
+    if not status_data:
+        raise HTTPException(status_code=404, detail="Task not found")
+    if status_data.get("user") != username:
+        raise HTTPException(status_code=403, detail="Not your task")
+    return status_data
+
+
 class TitleRequest(BaseModel):
     user_prompt: str
     response: str
diff --git a/backend/app/services/tasks.py b/backend/app/services/tasks.py
new file mode 100644
index 0000000..003c02d
--- /dev/null
+++ b/backend/app/services/tasks.py
@@ -0,0 +1,304 @@
+"""
+Background task management for debate & council operations.
+
+Decouples execution from SSE delivery so tasks survive browser disconnects.
+Results are persisted progressively to disk.
+"""
+
+import asyncio
+import json
+import logging
+import os
+import time
+from dataclasses import dataclass, field
+from enum import Enum
+from typing import Any, AsyncGenerator, Dict, List, Optional
+
+logger = logging.getLogger("contextflow.tasks")
+
+DATA_ROOT = os.path.abspath(os.getenv("DATA_ROOT", os.path.join(os.getcwd(), "data")))
+
+
+class TaskStatus(str, Enum):
+    PENDING = "pending"
+    RUNNING = "running"
+    COMPLETED = "completed"
+    FAILED = "failed"
+    INTERRUPTED = "interrupted"
+
+
+# Chunk event types that are NOT persisted to disk (redundant with *_complete events)
+_CHUNK_EVENT_TYPES = {"final_chunk", "stage3_chunk"}
+
+
+@dataclass
+class TaskInfo:
+    task_id: str
+    user: str
+    node_id: str
+    task_type: str  # "debate" | "council"
+    status: TaskStatus
+    created_at: float
+    updated_at: float
+    events: List[Dict] = field(default_factory=list)
+    result: Optional[Dict] = None
+    error: Optional[str] = None
+    # In-memory only (not serialized):
+    new_event_signal: asyncio.Event = field(default_factory=asyncio.Event, repr=False)
+    asyncio_task: Optional[asyncio.Task] = field(default=None, repr=False)
+
+
+# --------------- In-memory registry ---------------
+_task_registry: Dict[str, TaskInfo] = {}
+
+
+def register_task(task_id: str, user: str, node_id: str, task_type: str) -> TaskInfo:
+    """Create and register a new task in memory + persist initial state to disk."""
+    now = time.time()
+    info = TaskInfo(
+        task_id=task_id,
+        user=user,
+        node_id=node_id,
+        task_type=task_type,
+        status=TaskStatus.PENDING,
+        created_at=now,
+        updated_at=now,
+    )
+    _task_registry[task_id] = info
+    _persist_task(info)
+    return info
+
+
+def get_task(task_id: str) -> Optional[TaskInfo]:
+    """Get task from in-memory registry."""
+    return _task_registry.get(task_id)
+
+
+def get_task_status(user: str, task_id: str) -> Optional[Dict]:
+    """
+    Get task status dict. In-memory first, fallback to disk.
+    If disk says "running" but not in memory → report "interrupted".
+    """
+    info = _task_registry.get(task_id)
+    if info:
+        return _task_to_dict(info)
+
+    # Fallback: read from disk
+    disk_data = _load_task_from_disk(user, task_id)
+    if not disk_data:
+        return None
+
+    # If disk says running but it's not in memory, backend must have restarted
+    if disk_data.get("status") == TaskStatus.RUNNING:
+        disk_data["status"] = TaskStatus.INTERRUPTED
+    return disk_data
+
+
+async def run_task_in_background(info: TaskInfo, event_generator: AsyncGenerator[str, None]) -> None:
+    """
+    Consume the async generator, appending events to info.events.
+    Signals new_event_signal for SSE consumers.
+    Persists at milestones, skips chunk events from persistence.
+    """
+    info.status = TaskStatus.RUNNING
+    info.updated_at = time.time()
+    _persist_task(info)
+
+    try:
+        async for raw_sse in event_generator:
+            # raw_sse is a string like 'data: {"type": "...", ...}\n\n'
+            evt = _parse_sse_event(raw_sse)
+            if evt is None:
+                continue
+
+            evt_type = evt.get("type", "")
+
+            # Always store in memory (even chunks, for live streaming)
+            info.events.append(evt)
+            info.updated_at = time.time()
+
+            # Signal waiting SSE consumers
+            info.new_event_signal.set()
+            info.new_event_signal.clear()
+
+            # Persist at milestones (skip chunk events)
+            if evt_type not in _CHUNK_EVENT_TYPES:
+                _persist_task(info)
+
+            # Detect completion / error
+            if evt_type in ("complete", "debate_complete"):
+                info.result = _extract_result(info.task_type, info.events)
+                info.status = TaskStatus.COMPLETED
+                info.updated_at = time.time()
+                _persist_task(info)
+                return
+            if evt_type == "error":
+                info.error = evt.get("data", {}).get("message", "Unknown error") if isinstance(evt.get("data"), dict) else str(evt.get("data", "Unknown error"))
+                info.status = TaskStatus.FAILED
+                info.updated_at = time.time()
+                _persist_task(info)
+                return
+
+        # Generator exhausted without explicit complete/error
+        if info.status == TaskStatus.RUNNING:
+            info.result = _extract_result(info.task_type, info.events)
+            info.status = TaskStatus.COMPLETED
+            info.updated_at = time.time()
+            _persist_task(info)
+
+    except asyncio.CancelledError:
+        info.status = TaskStatus.INTERRUPTED
+        info.updated_at = time.time()
+        _persist_task(info)
+        raise
+    except Exception as e:
+        logger.exception("Background task %s failed: %s", info.task_id, e)
+        info.error = str(e)
+        info.status = TaskStatus.FAILED
+        info.updated_at = time.time()
+        _persist_task(info)
+
+
+def cleanup_stale_tasks(data_root: Optional[str] = None) -> int:
+    """
+    On startup, scan task files on disk and mark any "running" as "interrupted".
+    Returns count of tasks marked interrupted.
+    """
+    root = data_root or DATA_ROOT
+    count = 0
+    if not os.path.exists(root):
+        return count
+
+    for user_dir in os.listdir(root):
+        tasks_dir = os.path.join(root, user_dir, "tasks")
+        if not os.path.isdir(tasks_dir):
+            continue
+        for fname in os.listdir(tasks_dir):
+            if not fname.endswith(".json"):
+                continue
+            fpath = os.path.join(tasks_dir, fname)
+            try:
+                with open(fpath, "r", encoding="utf-8") as f:
+                    data = json.load(f)
+                if data.get("status") == TaskStatus.RUNNING:
+                    data["status"] = TaskStatus.INTERRUPTED
+                    data["updated_at"] = time.time()
+                    with open(fpath, "w", encoding="utf-8") as f:
+                        json.dump(data, f, ensure_ascii=False)
+                    count += 1
+                    logger.info("Marked stale task as interrupted: %s", fname)
+            except Exception as e:
+                logger.warning("Failed to process stale task file %s: %s", fpath, e)
+
+    return count
+
+
+# --------------- Persistence helpers ---------------
+
+def _task_dir(user: str) -> str:
+    d = os.path.join(DATA_ROOT, user, "tasks")
+    os.makedirs(d, exist_ok=True)
+    return d
+
+
+def _task_file_path(user: str, task_id: str) -> str:
+    return os.path.join(_task_dir(user), f"{task_id}.json")
+
+
+def _persist_task(info: TaskInfo) -> None:
+    """Write task state to disk. Excludes chunk events for smaller files."""
+    try:
+        data = _task_to_dict(info, exclude_chunks=True)
+        path = _task_file_path(info.user, info.task_id)
+        with open(path, "w", encoding="utf-8") as f:
+            json.dump(data, f, ensure_ascii=False)
+    except Exception as e:
+        logger.warning("Failed to persist task %s: %s", info.task_id, e)
+
+
+def _load_task_from_disk(user: str, task_id: str) -> Optional[Dict]:
+    path = _task_file_path(user, task_id)
+    if not os.path.exists(path):
+        return None
+    try:
+        with open(path, "r", encoding="utf-8") as f:
+            return json.load(f)
+    except Exception as e:
+        logger.warning("Failed to load task %s from disk: %s", task_id, e)
+        return None
+
+
+def _task_to_dict(info: TaskInfo, exclude_chunks: bool = False) -> Dict:
+    events = info.events
+    if exclude_chunks:
+        events = [e for e in events if e.get("type") not in _CHUNK_EVENT_TYPES]
+    return {
+        "task_id": info.task_id,
+        "user": info.user,
+        "node_id": info.node_id,
+        "task_type": info.task_type,
+        "status": info.status.value if isinstance(info.status, TaskStatus) else info.status,
+        "created_at": info.created_at,
+        "updated_at": info.updated_at,
+        "events": events,
+        "result": info.result,
+        "error": info.error,
+    }
+
+
+def _parse_sse_event(raw: str) -> Optional[Dict]:
+    """Parse a raw SSE string like 'data: {...}\\n\\n' into a dict."""
+    raw = raw.strip()
+    if not raw:
+        return None
+    for line in raw.split("\n"):
+        line = line.strip()
+        if line.startswith("data: "):
+            try:
+                return json.loads(line[6:])
+            except json.JSONDecodeError:
+                return None
+    return None
+
+
+def _extract_result(task_type: str, events: List[Dict]) -> Dict:
+    """Reconstruct final result data from accumulated events."""
+    if task_type == "council":
+        stage1 = None
+        stage2 = None
+        stage3 = None
+        for evt in events:
+            t = evt.get("type", "")
+            if t == "stage1_complete":
+                stage1 = evt.get("data")
+            elif t == "stage1_model_complete":
+                if stage1 is None:
+                    stage1 = []
+                stage1.append(evt.get("data"))
+            elif t == "stage2_complete":
+                stage2 = evt.get("data")
+            elif t == "stage3_complete":
+                stage3 = evt.get("data")
+        return {"councilData": {"stage1": stage1, "stage2": stage2, "stage3": stage3}}
+
+    elif task_type == "debate":
+        rounds: List[Dict] = []
+        final_verdict = None
+        for evt in events:
+            t = evt.get("type", "")
+            if t == "round_complete":
+                rounds.append(evt.get("data", {}))
+            elif t == "final_complete":
+                final_verdict = evt.get("data")
+            elif t == "judge_decision":
+                # Attach to last round
+                if rounds:
+                    rounds[-1]["judgeDecision"] = evt.get("data")
+            elif t == "model_eliminated":
+                if rounds:
+                    if "eliminated" not in rounds[-1]:
+                        rounds[-1]["eliminated"] = []
+                    rounds[-1]["eliminated"].append(evt.get("data"))
+        return {"debateRounds": rounds, "finalVerdict": final_verdict}
+
+    return {}