diff --git a/.env.example b/.env.example
new file mode 100644
index 0000000..ad38a4a
--- /dev/null
+++ b/.env.example
@@ -0,0 +1,13 @@
+# LLM (OpenAI-compatible gateway)
+LLM_BASE_URL=http://129.204.192.37:23000/v1
+LLM_API_KEY=sk-REDACTED
+LLM_MODEL=gpt-5.2
+
+# DB
+DATABASE_URL=sqlite:///./eventflow.sqlite3
+
+# Ingest tuning
+FETCH_UA=Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/123 Safari/537.36
+FETCH_RPS_PER_DOMAIN=0.2
+FETCH_TIMEOUT_SECS=15
+FETCH_MAX_RETRIES=3
diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..0b21aa8
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,7 @@
+.env
+*.sqlite
+*.sqlite3
+__pycache__/
+*.pyc
+.pytest_cache/
+*.log
diff --git a/backend/__init__.py b/backend/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/fastapi_app/__init__.py b/backend/fastapi_app/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/fastapi_app/kb/__init__.py b/backend/fastapi_app/kb/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/fastapi_app/kb/rules/__init__.py b/backend/fastapi_app/kb/rules/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/fastapi_app/kb/rules/rules.yml b/backend/fastapi_app/kb/rules/rules.yml
new file mode 100644
index 0000000..229c31d
--- /dev/null
+++ b/backend/fastapi_app/kb/rules/rules.yml
@@ -0,0 +1 @@
+# V1 rules placeholder
diff --git a/backend/fastapi_app/main.py b/backend/fastapi_app/main.py
new file mode 100644
index 0000000..6ea73ad
--- /dev/null
+++ b/backend/fastapi_app/main.py
@@ -0,0 +1,27 @@
+import threading
+
+from fastapi import FastAPI
+
+from .routers import analyze, ingest, query
+from .services.retry_worker import run_retry_loop
+
+app = FastAPI(title="eventflow-fastapi", version="0.1.0")
+
+app.include_router(ingest.router, prefix="/ingest", tags=["ingest"])
+app.include_router(analyze.router, prefix="/analyze", tags=["analyze"])
+app.include_router(query.router, prefix="/query", tags=["query"])
+
+
+_stop = threading.Event()
+
+
+@app.on_event("startup")
+def _startup() -> None:
+    # Background retry loop so transient gateway 5xx doesn't permanently stall parsing.
+    t = threading.Thread(target=run_retry_loop, args=(_stop,), daemon=True)
+    t.start()
+
+
+@app.on_event("shutdown")
+def _shutdown() -> None:
+    _stop.set()
diff --git a/backend/fastapi_app/routers/__init__.py b/backend/fastapi_app/routers/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/backend/fastapi_app/routers/analyze.py b/backend/fastapi_app/routers/analyze.py
new file mode 100644
index 0000000..cf590b8
--- /dev/null
+++ b/backend/fastapi_app/routers/analyze.py
@@ -0,0 +1,58 @@
+import json
+from fastapi import APIRouter, HTTPException
+
+from ..services.llm_extract import extract_event
+from ..services.store import ensure_schema, get_store, insert_event_result, insert_raw_item
+
+router = APIRouter()
+
+
+@router.post("/event")
+def analyze_event(payload: dict):
+    """Analyze one ad-hoc item and persist it.
+
+    This powers the UI while we build full ingesters.
+    """
+
+    title = (payload.get("title") or "").strip()
+    content = (payload.get("content") or "").strip()
+    if not title and not content:
+        raise HTTPException(status_code=400, detail="missing title/content")
+
+    st = get_store()
+    conn = st.connect()
+    ensure_schema(conn)
+
+    raw_item_id = insert_raw_item(
+        conn,
+        source=str(payload.get("source") or "manual"),
+        item_date=str(payload.get("date") or "") or "manual",
+        title=title[:500],
+        content=content[:20_000],
+        url=payload.get("url"),
+        published_at=payload.get("published_at"),
+        lang=payload.get("lang"),
+    )
+
+    res = extract_event(title=title, content=content, lang_hint=payload.get("lang"))
+
+    if res.get("ok") is True:
+        insert_event_result(
+            conn,
+            raw_item_id=raw_item_id,
+            model=str(res.get("model") or ""),
+            ok=True,
+            event_json=json.dumps(res.get("event"), ensure_ascii=True),
+            error=None,
+        )
+    else:
+        insert_event_result(
+            conn,
+            raw_item_id=raw_item_id,
+            model=str(res.get("model") or ""),
+            ok=False,
+            event_json=None,
+            error=str(res.get("error") or "unknown"),
+        )
+
+    return {"raw_item_id": raw_item_id, **res}
diff --git a/backend/fastapi_app/routers/ingest.py b/backend/fastapi_app/routers/ingest.py
new file mode 100644
index 0000000..3530dad
--- /dev/null
+++ b/backend/fastapi_app/routers/ingest.py
@@ -0,0 +1,214 @@
+import json
+from datetime import date
+
+from fastapi import APIRouter, HTTPException
+
+from ..services.llm_extract import extract_event
+from ..services.market_moves import fetch_moves_via_qfr
+from ..services.store import (
+    ensure_schema,
+    get_store,
+    insert_event_result,
+    insert_raw_item,
+)
+
+router = APIRouter()
+
+
+def _today() -> str:
+    return date.today().isoformat()
+
+
+@router.post("/rss")
+def ingest_rss(payload: dict):
+    """Ingest one or many RSS items.
+
+    Expected payload:
+      {"items": [{"title":..., "url":..., "published_at":..., "summary":..., "lang":...}, ...]}
+    """
+    items = payload.get("items")
+    if not isinstance(items, list) or not items:
+        raise HTTPException(status_code=400, detail="payload.items must be a non-empty list")
+
+    st = get_store()
+    conn = st.connect()
+    ensure_schema(conn)
+
+    n = 0
+    for it in items:
+        if not isinstance(it, dict):
+            continue
+        insert_raw_item(
+            conn,
+            source="rss",
+            item_date=payload.get("date") or _today(),
+            title=(it.get("title") or "")[:500],
+            content=(it.get("summary") or it.get("content") or "")[:20_000],
+            url=it.get("url"),
+            published_at=it.get("published_at"),
+            lang=it.get("lang"),
+        )
+        n += 1
+
+    return {"ok": True, "inserted": n}
+
+
+@router.post("/macro")
+def ingest_macro(payload: dict):
+    items = payload.get("items")
+    if not isinstance(items, list) or not items:
+        raise HTTPException(status_code=400, detail="payload.items must be a non-empty list")
+
+    st = get_store()
+    conn = st.connect()
+    ensure_schema(conn)
+
+    n = 0
+    for it in items:
+        if not isinstance(it, dict):
+            continue
+        insert_raw_item(
+            conn,
+            source="macro",
+            item_date=payload.get("date") or _today(),
+            title=(it.get("title") or "")[:500],
+            content=(it.get("content") or "")[:20_000],
+            url=it.get("url"),
+            published_at=it.get("published_at"),
+            lang=it.get("lang"),
+        )
+        n += 1
+
+    return {"ok": True, "inserted": n}
+
+
+@router.post("/market_moves")
+def ingest_market_moves(payload: dict):
+    items = payload.get("items")
+    if not isinstance(items, list) or not items:
+        raise HTTPException(status_code=400, detail="payload.items must be a non-empty list")
+
+    st = get_store()
+    conn = st.connect()
+    ensure_schema(conn)
+
+    n = 0
+    for it in items:
+        if not isinstance(it, dict):
+            continue
+        insert_raw_item(
+            conn,
+            source="market_moves",
+            item_date=payload.get("date") or _today(),
+            title=(it.get("title") or "")[:500],
+            content=(it.get("content") or "")[:20_000],
+            url=it.get("url"),
+            published_at=it.get("published_at"),
+            lang=it.get("lang"),
+        )
+        n += 1
+
+    return {"ok": True, "inserted": n}
+
+
+@router.post("/market_moves/run")
+def run_market_moves(payload: dict | None = None):
+    """Generate daily market-move items from QFR raw data and parse them into events."""
+
+    payload = payload or {}
+    day = str(payload.get("date") or _today())
+    # QFR raw data uses trade_date like YYYYMMDD.
+    trade_date = day.replace("-", "")
+
+    st = get_store()
+    conn = st.connect()
+    ensure_schema(conn)
+
+    data = fetch_moves_via_qfr(trade_date=trade_date, symbols=payload.get("symbols"))
+    if not data.get("ok"):
+        raise HTTPException(status_code=500, detail=data)
+
+    inserted = 0
+    parsed_ok = 0
+    parsed_err = 0
+
+    for mv in data.get("moves", []):
+        sym = mv.get("symbol")
+        td = mv.get("trade_date")
+        ret_1d = mv.get("ret_1d")
+        vol_20d = mv.get("vol_20d")
+        z_1d = mv.get("z_1d")
+
+        title = f"Market move {sym} {td}: ret_1d={ret_1d:.4f} z_1d={z_1d:.2f}" if isinstance(ret_1d, (int, float)) and isinstance(z_1d, (int, float)) else f"Market move {sym} {td}"
+        content = (
+            f"symbol={sym}\n"
+            f"trade_date={td}\n"
+            f"prev_trade_date={mv.get('prev_trade_date')}\n"
+            f"close={mv.get('close')} prev_close={mv.get('prev_close')}\n"
+            f"ret_1d={ret_1d} vol_20d={vol_20d} z_1d={z_1d}\n"
+            "Interpretation task: explain the most likely macro/industry drivers for this move and which assets could be affected."
+        )
+
+        raw_item_id = insert_raw_item(
+            conn,
+            source="market_moves",
+            item_date=day,
+            title=title[:500],
+            content=content[:20_000],
+            url=None,
+            published_at=None,
+            lang="en",
+        )
+        inserted += 1
+
+        try:
+            res = extract_event(title=title, content=content, lang_hint="en")
+        except Exception as e:
+            # Network/provider errors should not abort the whole batch.
+            insert_event_result(
+                conn,
+                raw_item_id=raw_item_id,
+                model="",
+                ok=False,
+                event_json=None,
+                error=f"llm_exception:{type(e).__name__}",
+            )
+            parsed_err += 1
+            continue
+
+        if res.get("ok") is True:
+            insert_event_result(
+                conn,
+                raw_item_id=raw_item_id,
+                model=str(res.get("model") or ""),
+                ok=True,
+                event_json=json.dumps(res.get("event"), ensure_ascii=True),
+                error=None,
+            )
+            parsed_ok += 1
+        else:
+            err = str(res.get("error") or "unknown")
+            if err == "llm_failed":
+                # Keep a short hint to debug gateway flakiness without dumping secrets.
+                exc = str(res.get("exc") or "")
+                if exc:
+                    err = f"{err}:{exc}"
+            insert_event_result(
+                conn,
+                raw_item_id=raw_item_id,
+                model=str(res.get("model") or ""),
+                ok=False,
+                event_json=None,
+                error=err,
+            )
+            parsed_err += 1
+
+    return {
+        "ok": True,
+        "date": day,
+        "inserted": inserted,
+        "parsed_ok": parsed_ok,
+        "parsed_err": parsed_err,
+        "errors": data.get("errors", []),
+        "symbols": data.get("symbols"),
+    }
diff --git a/backend/fastapi_app/routers/query.py b/backend/fastapi_app/routers/query.py
new file mode 100644
index 0000000..207e8b2
--- /dev/null
+++ b/backend/fastapi_app/routers/query.py
@@ -0,0 +1,215 @@
+from datetime import date
+
+from fastapi import APIRouter
+from fastapi.responses import HTMLResponse
+
+from ..services.store import counts, get_store, list_events, list_raw_items, sources_today
+
+router = APIRouter()
+
+
+@router.get("/health")
+def health():
+    return {"ok": True}
+
+
+@router.get("/status")
+def status(day: str | None = None):
+    st = get_store()
+    conn = st.connect()
+    item_date = day or date.today().isoformat()
+    return {
+        "date": item_date,
+        "counts": counts(conn),
+        "sources": sources_today(conn, item_date),
+    }
+
+
+@router.get("/raw_items")
+def raw_items(limit: int = 20):
+    st = get_store()
+    conn = st.connect()
+    return {"items": list_raw_items(conn, limit=limit)}
+
+
+@router.get("/events")
+def events(limit: int = 20):
+    st = get_store()
+    conn = st.connect()
+    return {"items": list_events(conn, limit=limit)}
+
+
+@router.get("/ui", response_class=HTMLResponse)
+def ui():
+    # Tiny no-build UI for early validation.
+    html = """<!doctype html>
+<html>
+<head>
+  <meta charset=\"utf-8\" />
+  <meta name=\"viewport\" content=\"width=device-width, initial-scale=1\" />
+  <title>EventFlow V1</title>
+  <style>
+    body { font-family: ui-sans-serif, system-ui, Arial; margin: 18px; }
+    h1 { margin: 0 0 8px 0; }
+    .row { display: flex; gap: 16px; flex-wrap: wrap; }
+    .card { border: 1px solid #ddd; border-radius: 8px; padding: 12px; min-width: 320px; }
+    table { border-collapse: collapse; width: 100%; }
+    th, td { border-bottom: 1px solid #eee; padding: 6px 8px; font-size: 13px; vertical-align: top; }
+    th { text-align: left; color: #444; }
+    code { background: #f6f6f6; padding: 2px 4px; border-radius: 4px; }
+    .muted { color: #666; }
+    .ok { color: #0a7; font-weight: 600; }
+    .bad { color: #b00; font-weight: 600; }
+    .mono { font-family: ui-monospace, SFMono-Regular, Menlo, monospace; }
+    button { padding: 6px 10px; }
+    input, textarea { width: 100%; box-sizing: border-box; }
+    textarea { height: 90px; }
+  </style>
+</head>
+<body>
+  <h1>EventFlow V1</h1>
+  <div class=\"muted\">Status page: sources, dates, counts, and parsed events.</div>
+  <div style=\"height:12px\"></div>
+
+  <div class=\"row\">
+    <div class=\"card\" style=\"flex: 1\">
+      <div style=\"display:flex; justify-content:space-between; align-items:center; gap:12px;\">
+        <div>
+          <div><b>Today</b>: <span id=\"today\"></span></div>
+          <div class=\"muted\">Auto-refresh every 5s</div>
+        </div>
+        <button onclick=\"refreshAll()\">Refresh</button>
+      </div>
+      <div style=\"height:10px\"></div>
+      <div id=\"counts\" class=\"mono\"></div>
+      <div style=\"height:10px\"></div>
+      <div><b>Sources (today)</b></div>
+      <table id=\"sources\"></table>
+    </div>
+
+    <div class=\"card\" style=\"flex: 1\">
+      <div><b>Quick Analyze (manual)</b></div>
+      <div class=\"muted\">This will save a raw item + parsed event into SQLite.</div>
+      <div style=\"height:8px\"></div>
+      <label>Title</label>
+      <input id=\"m_title\" placeholder=\"e.g. Fed signals higher-for-longer rates\" />
+      <div style=\"height:6px\"></div>
+      <label>Content</label>
+      <textarea id=\"m_content\" placeholder=\"Paste a paragraph...\"></textarea>
+      <div style=\"height:6px\"></div>
+      <button onclick=\"runAnalyze()\">Analyze</button>
+      <div id=\"m_out\" class=\"mono\" style=\"white-space:pre-wrap; margin-top:10px;\"></div>
+    </div>
+  </div>
+
+  <div style=\"height:16px\"></div>
+
+  <div class=\"row\">
+    <div class=\"card\" style=\"flex: 1\">
+      <div><b>Latest Raw Items</b> <span class=\"muted\">(limit 20)</span></div>
+      <table id=\"raw\"></table>
+    </div>
+    <div class=\"card\" style=\"flex: 1\">
+      <div><b>Latest Parsed Events</b> <span class=\"muted\">(limit 20)</span></div>
+      <table id=\"events\"></table>
+    </div>
+  </div>
+
+<script>
+function esc(s) {
+  return (s ?? '').toString().replaceAll('&', '&amp;').replaceAll('<', '&lt;').replaceAll('>', '&gt;');
+}
+
+async function jget(url) {
+  const r = await fetch(url);
+  if (!r.ok) throw new Error('HTTP ' + r.status);
+  return await r.json();
+}
+
+function setTable(el, headers, rows) {
+  let h = '<tr>' + headers.map(x => `<th>${esc(x)}</th>`).join('') + '</tr>';
+  let b = rows.map(r => '<tr>' + r.map(x => `<td>${x}</td>`).join('') + '</tr>').join('');
+  el.innerHTML = h + b;
+}
+
+async function refreshStatus() {
+  const st = await jget('/query/status');
+  document.getElementById('today').textContent = st.date;
+  const c = st.counts;
+  document.getElementById('counts').textContent = `raw_items=${c.raw_items} events=${c.events} ok=${c.events_ok} err=${c.events_err}`;
+
+  const srows = (st.sources || []).map(x => [esc(x.source), esc(x.count)]);
+  setTable(document.getElementById('sources'), ['source', 'count'], srows);
+}
+
+async function refreshRaw() {
+  const data = await jget('/query/raw_items?limit=20');
+  const rows = (data.items || []).map(it => {
+    const u = it.url ? `<a href="${esc(it.url)}" target="_blank">link</a>` : '';
+    return [
+      `<span class="mono">${esc(it.id)}</span>`,
+      esc(it.source),
+      esc(it.item_date),
+      esc(it.lang || ''),
+      esc((it.title || '').slice(0, 120)),
+      u,
+    ];
+  });
+  setTable(document.getElementById('raw'), ['id', 'source', 'date', 'lang', 'title', 'url'], rows);
+}
+
+async function refreshEvents() {
+  const data = await jget('/query/events?limit=20');
+  const rows = (data.items || []).map(it => {
+    const cls = it.ok ? 'ok' : 'bad';
+    let ev = '';
+    try {
+      if (it.event_json) {
+        const obj = JSON.parse(it.event_json);
+        ev = `${esc(obj.event_type)} dir=${esc(obj.impact_direction)} conf=${esc(obj.confidence)}<br/><span class="muted">${esc((obj.summary_en || obj.summary_zh || '').slice(0, 140))}</span>`;
+      }
+    } catch (e) {
+      ev = '<span class="bad">bad json</span>';
+    }
+    return [
+      `<span class="mono">${esc(it.id)}</span>`,
+      esc(it.source),
+      `<span class="${cls}">${it.ok ? 'ok' : 'err'}</span>`,
+      esc((it.title || '').slice(0, 90)),
+      ev,
+    ];
+  });
+  setTable(document.getElementById('events'), ['id', 'source', 'ok', 'raw_title', 'event'], rows);
+}
+
+async function runAnalyze() {
+  const title = document.getElementById('m_title').value;
+  const content = document.getElementById('m_content').value;
+  const out = document.getElementById('m_out');
+  out.textContent = 'running...';
+  const r = await fetch('/analyze/event', {
+    method: 'POST',
+    headers: {'Content-Type': 'application/json'},
+    body: JSON.stringify({source: 'manual', date: new Date().toISOString().slice(0,10), title, content, lang: ''}),
+  });
+  const j = await r.json();
+  out.textContent = JSON.stringify(j, null, 2);
+  await refreshAll();
+}
+
+async function refreshAll() {
+  try {
+    await refreshStatus();
+    await refreshRaw();
+    await refreshEvents();
+  } catch (e) {
+    console.error(e);
+  }
+}
+
+refreshAll();
+setInterval(refreshAll, 5000);
+</script>
+</body>
+</html>"""
+    return HTMLResponse(content=html)
diff --git a/backend/fastapi_app/services/__init__.py b/backend/fastapi_app/services/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/eventflow/backend/fastapi_app/services/llm_extract.py b/backend/fastapi_app/services/llm_extract.py
similarity index 100%
rename from eventflow/backend/fastapi_app/services/llm_extract.py
rename to backend/fastapi_app/services/llm_extract.py
diff --git a/backend/fastapi_app/services/market_moves.py b/backend/fastapi_app/services/market_moves.py
new file mode 100644
index 0000000..b55651d
--- /dev/null
+++ b/backend/fastapi_app/services/market_moves.py
@@ -0,0 +1,68 @@
+import json
+import os
+import subprocess
+from typing import Any
+
+
+WATCHLIST_DEFAULT = [
+    # A-share ETF proxies for indices (QFR raw data uses ETF parquets)
+    # HS300
+    "510300.SH",
+    # ZZ500
+    "510500.SH",
+    # ChiNext
+    "159915.SZ",
+    # SSE50 proxy (may not exist in rawdir unless downloaded)
+    "510050.SH",
+
+    # Futures placeholders (not in QFR rawdir by default; will show as errors until sourced)
+    "AU.SHF",
+    "CU.SHF",
+    "M.DCE",
+    "TA.CZCE",
+    "SC.INE",
+]
+
+
+def fetch_moves_via_qfr(*, trade_date: str | None = None, symbols: list[str] | None = None) -> dict[str, Any]:
+    """Fetch day-level moves by shelling out to the existing qfr conda env.
+
+    Reason: eventflow env is kept minimal; qfr env already has pandas/pyarrow.
+    """
+
+    sym_list = symbols or WATCHLIST_DEFAULT
+    rawdir = os.environ.get("QFR_RAWDIR", "/home/openclaw/projects/quant-factor-research/data/raw")
+
+    env = os.environ.copy()
+    env["QFR_RAWDIR"] = rawdir
+    env["QFR_SYMBOLS"] = ",".join(sym_list)
+    if trade_date:
+        env["QFR_TRADE_DATE"] = trade_date
+
+    conda = os.environ.get("CONDA_BIN", "/home/openclaw/miniconda3/bin/conda")
+    script = os.path.join(os.path.dirname(__file__), "market_moves_qfr.py")
+
+    cmd = [conda, "run", "-n", "qfr", "python", script]
+    proc = subprocess.run(cmd, env=env, stdout=subprocess.PIPE, stderr=subprocess.PIPE, text=True)
+
+    if proc.returncode != 0:
+        return {
+            "ok": False,
+            "error": "qfr_subprocess_failed",
+            "returncode": proc.returncode,
+            "stderr": proc.stderr[-4000:],
+        }
+
+    try:
+        data = json.loads(proc.stdout)
+    except json.JSONDecodeError:
+        return {
+            "ok": False,
+            "error": "invalid_json_from_qfr",
+            "stdout": proc.stdout[-2000:],
+            "stderr": proc.stderr[-2000:],
+        }
+
+    data["ok"] = True
+    data["symbols"] = sym_list
+    return data
diff --git a/backend/fastapi_app/services/market_moves_qfr.py b/backend/fastapi_app/services/market_moves_qfr.py
new file mode 100644
index 0000000..1891fe1
--- /dev/null
+++ b/backend/fastapi_app/services/market_moves_qfr.py
@@ -0,0 +1,108 @@
+import json
+import os
+from dataclasses import asdict, dataclass
+from typing import Any
+
+import pandas as pd
+
+
+@dataclass
+class Move:
+    symbol: str
+    trade_date: str
+    prev_trade_date: str
+    close: float
+    prev_close: float
+    ret_1d: float
+    vol_20d: float
+    z_1d: float
+
+
+def _read_one(rawdir: str, symbol: str) -> pd.DataFrame:
+    # QFR stores parquet as e.g. 510300SH.parquet / 159915SZ.parquet
+    fn = symbol.replace(".", "") + ".parquet"
+    p = os.path.join(rawdir, fn)
+    df = pd.read_parquet(p)
+    # Standardize
+    if "trade_date" not in df.columns or "close" not in df.columns:
+        raise RuntimeError(f"unexpected parquet schema for {symbol}: {df.columns.tolist()}")
+    df = df.sort_values("trade_date").reset_index(drop=True)
+    return df
+
+
+def _calc_move(df: pd.DataFrame, symbol: str, trade_date: str | None) -> Move | None:
+    if df.empty:
+        return None
+    # pick last available <= trade_date if given
+    if trade_date:
+        df2 = df[df["trade_date"] <= trade_date]
+        if df2.empty:
+            return None
+        df = df2
+
+    if len(df) < 2:
+        return None
+
+    # compute returns
+    close = df["close"].astype(float)
+    ret = close.pct_change()
+
+    i = len(df) - 1
+    prev_i = i - 1
+
+    td = str(df.iloc[i]["trade_date"])
+    ptd = str(df.iloc[prev_i]["trade_date"])
+
+    close_i = float(close.iloc[i])
+    prev_close_i = float(close.iloc[prev_i])
+    ret_1d = float(ret.iloc[i])
+
+    # vol over last 20 returns (excluding NaN)
+    vol_20 = float(ret.iloc[max(0, i - 20 + 1) : i + 1].std(skipna=True))
+    if not (vol_20 > 0):
+        vol_20 = 0.0
+    z = float(ret_1d / vol_20) if vol_20 > 0 else 0.0
+
+    return Move(
+        symbol=symbol,
+        trade_date=td,
+        prev_trade_date=ptd,
+        close=close_i,
+        prev_close=prev_close_i,
+        ret_1d=ret_1d,
+        vol_20d=vol_20,
+        z_1d=z,
+    )
+
+
+def main() -> None:
+    rawdir = os.environ.get("QFR_RAWDIR") or "/home/openclaw/projects/quant-factor-research/data/raw"
+    symbols = (os.environ.get("QFR_SYMBOLS") or "").strip()
+    trade_date = (os.environ.get("QFR_TRADE_DATE") or "").strip() or None
+
+    if not symbols:
+        raise SystemExit("QFR_SYMBOLS is required")
+
+    out: dict[str, Any] = {
+        "rawdir": rawdir,
+        "trade_date": trade_date,
+        "moves": [],
+        "errors": [],
+    }
+
+    for sym in [s.strip() for s in symbols.split(",") if s.strip()]:
+        try:
+            df = _read_one(rawdir, sym)
+            mv = _calc_move(df, sym, trade_date)
+            if mv is None:
+                out["errors"].append({"symbol": sym, "error": "no_data"})
+                continue
+            out["moves"].append(asdict(mv))
+        except Exception as e:
+            out["errors"].append({"symbol": sym, "error": str(e)})
+
+    print(json.dumps(out, ensure_ascii=True))
+
+
+if __name__ == "__main__":
+    main()
diff --git a/backend/fastapi_app/services/retry_worker.py b/backend/fastapi_app/services/retry_worker.py
new file mode 100644
index 0000000..fdad3a3
--- /dev/null
+++ b/backend/fastapi_app/services/retry_worker.py
@@ -0,0 +1,99 @@
+import json
+import os
+import random
+import threading
+import time
+
+from .llm_extract import extract_event
+from .store import ensure_schema, get_store, insert_event_result
+
+
+def _pick_pending(conn, *, batch: int, max_attempts: int, min_age_s: int, retry_after_s: int):
+    now = int(time.time())
+    # Select items that are not successfully parsed yet.
+    # We rely on raw_items.status/parse_attempts which are updated by insert_event_result.
+    rows = conn.execute(
+        """
+        SELECT id, title, content, lang, status, parse_attempts, last_parse_at
+        FROM raw_items
+        WHERE status != 'parsed_ok'
+          AND COALESCE(parse_attempts, 0) < ?
+          AND (? - created_at) >= ?
+          AND (last_parse_at IS NULL OR (? - last_parse_at) >= ?)
+        ORDER BY id ASC
+        LIMIT ?
+        """,
+        (max_attempts, now, min_age_s, now, retry_after_s, batch),
+    ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def run_retry_loop(stop: threading.Event) -> None:
+    if os.environ.get("PARSE_RETRY_ENABLED", "1") not in ("1", "true", "TRUE", "yes", "YES"):
+        return
+
+    interval_s = int(os.environ.get("PARSE_RETRY_INTERVAL_S", "120"))
+    batch = int(os.environ.get("PARSE_RETRY_BATCH", "5"))
+    max_attempts = int(os.environ.get("PARSE_RETRY_MAX_ATTEMPTS", "6"))
+    min_age_s = int(os.environ.get("PARSE_RETRY_MIN_AGE_S", "2"))
+    retry_after_s = int(os.environ.get("PARSE_RETRY_AFTER_S", "300"))
+    sleep_between_s = float(os.environ.get("PARSE_RETRY_SLEEP_BETWEEN_S", "0.5"))
+
+    st = get_store()
+
+    while not stop.is_set():
+        try:
+            conn = st.connect()
+            ensure_schema(conn)
+
+            items = _pick_pending(
+                conn,
+                batch=batch,
+                max_attempts=max_attempts,
+                min_age_s=min_age_s,
+                retry_after_s=retry_after_s,
+            )
+
+            for it in items:
+                if stop.is_set():
+                    break
+
+                # Jitter to reduce gateway burstiness.
+                time.sleep(sleep_between_s + random.random() * 0.4)
+
+                title = (it.get("title") or "")[:500]
+                content = (it.get("content") or "")[:20000]
+                lang_hint = it.get("lang")
+
+                res = extract_event(title=title, content=content, lang_hint=lang_hint)
+
+                if res.get("ok") is True:
+                    insert_event_result(
+                        conn,
+                        raw_item_id=int(it["id"]),
+                        model=str(res.get("model") or ""),
+                        ok=True,
+                        event_json=json.dumps(res.get("event"), ensure_ascii=True),
+                        error=None,
+                    )
+                else:
+                    # Preserve hint for debugging (no secrets).
+                    err = str(res.get("error") or "unknown")
+                    if err == "llm_failed":
+                        exc = str(res.get("exc") or "")
+                        if exc:
+                            err = f"{err}:{exc}"
+                    insert_event_result(
+                        conn,
+                        raw_item_id=int(it["id"]),
+                        model=str(res.get("model") or ""),
+                        ok=False,
+                        event_json=None,
+                        error=err,
+                    )
+
+        except Exception:
+            # Keep the loop alive; details are in DB (last_error) or uvicorn logs.
+            pass
+
+        stop.wait(interval_s)
diff --git a/backend/fastapi_app/services/store.py b/backend/fastapi_app/services/store.py
new file mode 100644
index 0000000..e4d4342
--- /dev/null
+++ b/backend/fastapi_app/services/store.py
@@ -0,0 +1,205 @@
+import os
+import sqlite3
+import time
+from dataclasses import dataclass
+from typing import Any
+
+
+@dataclass(frozen=True)
+class Store:
+    db_path: str
+
+    def connect(self) -> sqlite3.Connection:
+        # check_same_thread=False allows use across simple dev reload threads.
+        conn = sqlite3.connect(self.db_path, check_same_thread=False)
+        conn.row_factory = sqlite3.Row
+        return conn
+
+
+def _sqlite_path_from_database_url(url: str) -> str:
+    # V1 supports only sqlite. Examples:
+    # - sqlite:///./eventflow.sqlite3
+    # - sqlite:////abs/path/eventflow.sqlite3
+    if not url.startswith("sqlite:///"):
+        raise ValueError("Only sqlite DATABASE_URL is supported in V1")
+    return url[len("sqlite:///") :]
+
+
+def get_store() -> Store:
+    url = os.environ.get("DATABASE_URL", "sqlite:///./eventflow.sqlite3")
+    path = _sqlite_path_from_database_url(url)
+    return Store(db_path=path)
+
+
+def _ensure_column(conn: sqlite3.Connection, *, table: str, col: str, ddl: str) -> None:
+    cols = [r[1] for r in conn.execute(f"PRAGMA table_info({table})").fetchall()]
+    if col not in cols:
+        conn.execute(f"ALTER TABLE {table} ADD COLUMN {ddl}")
+
+
+def ensure_schema(conn: sqlite3.Connection) -> None:
+    conn.executescript(
+        """
+        PRAGMA journal_mode=WAL;
+
+        CREATE TABLE IF NOT EXISTS raw_items (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            source TEXT NOT NULL,
+            item_date TEXT NOT NULL,
+            published_at TEXT,
+            url TEXT,
+            title TEXT,
+            content TEXT,
+            lang TEXT,
+            status TEXT NOT NULL DEFAULT 'new',
+            created_at INTEGER NOT NULL
+        );
+
+        CREATE INDEX IF NOT EXISTS idx_raw_items_source_date ON raw_items(source, item_date);
+        CREATE INDEX IF NOT EXISTS idx_raw_items_created_at ON raw_items(created_at);
+
+        CREATE TABLE IF NOT EXISTS events (
+            id INTEGER PRIMARY KEY AUTOINCREMENT,
+            raw_item_id INTEGER NOT NULL,
+            model TEXT,
+            ok INTEGER NOT NULL,
+            event_json TEXT,
+            error TEXT,
+            created_at INTEGER NOT NULL,
+            FOREIGN KEY(raw_item_id) REFERENCES raw_items(id)
+        );
+
+        CREATE INDEX IF NOT EXISTS idx_events_raw_item_id ON events(raw_item_id);
+        CREATE INDEX IF NOT EXISTS idx_events_created_at ON events(created_at);
+        """
+    )
+
+    # Light migrations for V1: keep raw_items parse state for retry.
+    _ensure_column(conn, table="raw_items", col="parse_attempts", ddl="parse_attempts INTEGER NOT NULL DEFAULT 0")
+    _ensure_column(conn, table="raw_items", col="last_parse_at", ddl="last_parse_at INTEGER")
+    _ensure_column(conn, table="raw_items", col="last_error", ddl="last_error TEXT")
+
+    conn.commit()
+
+
+def insert_raw_item(
+    conn: sqlite3.Connection,
+    *,
+    source: str,
+    item_date: str,
+    title: str | None,
+    content: str | None,
+    url: str | None = None,
+    published_at: str | None = None,
+    lang: str | None = None,
+) -> int:
+    now = int(time.time())
+    cur = conn.execute(
+        """
+        INSERT INTO raw_items (source, item_date, published_at, url, title, content, lang, status, created_at)
+        VALUES (?, ?, ?, ?, ?, ?, ?, 'new', ?)
+        """,
+        (source, item_date, published_at, url, title, content, lang, now),
+    )
+    conn.commit()
+    return int(cur.lastrowid)
+
+
+def insert_event_result(
+    conn: sqlite3.Connection,
+    *,
+    raw_item_id: int,
+    model: str,
+    ok: bool,
+    event_json: str | None,
+    error: str | None,
+) -> int:
+    now = int(time.time())
+    cur = conn.execute(
+        """
+        INSERT INTO events (raw_item_id, model, ok, event_json, error, created_at)
+        VALUES (?, ?, ?, ?, ?, ?)
+        """,
+        (raw_item_id, model, 1 if ok else 0, event_json, error, now),
+    )
+
+    # Update raw item parse state (best-effort).
+    status = "parsed_ok" if ok else "parsed_err"
+    conn.execute(
+        """
+        UPDATE raw_items
+        SET status = ?,
+            parse_attempts = COALESCE(parse_attempts, 0) + 1,
+            last_parse_at = ?,
+            last_error = ?
+        WHERE id = ?
+        """,
+        (status, now, error, raw_item_id),
+    )
+
+    conn.commit()
+    return int(cur.lastrowid)
+
+
+def counts(conn: sqlite3.Connection) -> dict[str, Any]:
+    ensure_schema(conn)
+    row_raw = conn.execute("SELECT COUNT(1) AS n FROM raw_items").fetchone()
+    row_ev = conn.execute("SELECT COUNT(1) AS n FROM events").fetchone()
+    row_ok = conn.execute("SELECT COUNT(1) AS n FROM events WHERE ok=1").fetchone()
+    row_bad = conn.execute("SELECT COUNT(1) AS n FROM events WHERE ok=0").fetchone()
+    return {
+        "raw_items": int(row_raw["n"]),
+        "events": int(row_ev["n"]),
+        "events_ok": int(row_ok["n"]),
+        "events_err": int(row_bad["n"]),
+    }
+
+
+def sources_today(conn: sqlite3.Connection, item_date: str) -> list[dict[str, Any]]:
+    ensure_schema(conn)
+    rows = conn.execute(
+        """
+        SELECT source, COUNT(1) AS n
+        FROM raw_items
+        WHERE item_date = ?
+        GROUP BY source
+        ORDER BY source
+        """,
+        (item_date,),
+    ).fetchall()
+    return [{"source": r["source"], "count": int(r["n"])} for r in rows]
+
+
+def list_raw_items(conn: sqlite3.Connection, limit: int = 20) -> list[dict[str, Any]]:
+    ensure_schema(conn)
+    rows = conn.execute(
+        """
+        SELECT id, source, item_date, published_at, url, title, lang, status, created_at
+        FROM raw_items
+        ORDER BY id DESC
+        LIMIT ?
+        """,
+        (limit,),
+    ).fetchall()
+    return [dict(r) for r in rows]
+
+
+def list_events(conn: sqlite3.Connection, limit: int = 20) -> list[dict[str, Any]]:
+    ensure_schema(conn)
+    rows = conn.execute(
+        """
+        SELECT e.id, e.raw_item_id, e.model, e.ok, e.event_json, e.error, e.created_at,
+               r.source, r.item_date, r.title
+        FROM events e
+        JOIN raw_items r ON r.id = e.raw_item_id
+        ORDER BY e.id DESC
+        LIMIT ?
+        """,
+        (limit,),
+    ).fetchall()
+    out: list[dict[str, Any]] = []
+    for r in rows:
+        d = dict(r)
+        # keep event_json as string; UI can parse if needed
+        out.append(d)
+    return out
diff --git a/eventflow.sqlite3-shm b/eventflow.sqlite3-shm
new file mode 100644
index 0000000..2eb1696
Binary files /dev/null and b/eventflow.sqlite3-shm differ
diff --git a/eventflow.sqlite3-wal b/eventflow.sqlite3-wal
new file mode 100644
index 0000000..abbd90a
Binary files /dev/null and b/eventflow.sqlite3-wal differ
diff --git a/requirements.txt b/requirements.txt
new file mode 100644
index 0000000..d80b5b9
--- /dev/null
+++ b/requirements.txt
@@ -0,0 +1,8 @@
+django>=5.0,<6.0
+fastapi>=0.110
+uvicorn[standard]>=0.27
+requests>=2.31
+feedparser>=6.0
+python-dotenv>=1.0
+openai>=1.40
+pydantic>=2.6