diff --git a/backfill.py b/backfill.py
new file mode 100644
index 0000000..d5d4664
--- /dev/null
+++ b/backfill.py
@@ -0,0 +1,162 @@
+# backfill_from_ghost.py
+from __future__ import annotations
+import os, re, sys, html
+from typing import Dict, List, Optional
+import requests
+from bs4 import BeautifulSoup
+
+# Reuse your existing GhostAdmin client (same headers/base/proxy behavior)
+# Adjust the import path if your Ghost client lives elsewhere.
+from presquegratos import GhostAdmin
+
+from storage import Storage
+from keys import xgp_key, egs_key, psplus_key
+
+# ---------------- Ghost helpers (reusing your admin client) ----------------
+def ghost_list_posts(ghost: GhostAdmin, page: int = 1) -> Dict:
+    # Minimal params: avoid 'filter' and 'fields' to dodge 400 behind __bot proxy
+    url = ghost.base + "posts/"
+    params = {
+        "limit": "50",
+        "page": str(page),
+        "order": "published_at DESC",
+        "formats": "lexical,html",  # <-- IMPORTANT
+    }
+    r = requests.get(url, headers=ghost._headers(), params=params, timeout=30)
+    r.raise_for_status()
+    return r.json()
+
+def list_recap_posts(ghost: GhostAdmin, hard_limit: int = 2000) -> List[Dict]:
+    posts: List[Dict] = []
+    page = 1
+    while True:
+        data = ghost_list_posts(ghost, page=page)
+        batch = data.get("posts", [])
+        if not batch:
+            break
+        # client-side filter to be robust to proxy quirks
+        for p in batch:
+            title = (p.get("title") or "").strip()
+            if title.startswith("Récap hebdo"):
+                posts.append(p)
+        if len(batch) < 50 or len(posts) >= hard_limit:
+            break
+        page += 1
+    return posts
+
+# ---------------- Parsing helpers (unchanged) ----------------
+#MS_STORE_RE = re.compile(r"(?:microsoft|xbox)\.com/.*/store/.*/([0-9A-Z]{12,})", re.I)
+MS_STORE_RE = re.compile(r"(?:xbox|microsoft)\.com/.*/store/.*/([0-9A-Z]{12,16})", re.I)
+EPIC_RE     = re.compile(r"epicgames\.com/store/.*/p/([\w\-]+)", re.I)
+PSBLOG_RE   = re.compile(r"blog\.playstation\.com/.*", re.I)
+
+def clean_text(s: str) -> str:
+    return re.sub(r"\s+", " ", html.unescape(s or "")).strip()
+
+def extract_sections(soup: BeautifulSoup) -> Dict[str, BeautifulSoup]:
+    sections: Dict[str, BeautifulSoup] = {}
+    current = None
+    current_key = None
+    for node in soup.find_all(["h2","h3","h4","p","ul","ol","div","section"]):
+        if node.name in ("h2","h3","h4"):
+            title = clean_text(node.get_text())
+            key = None
+            tl = title.lower()
+            if "game pass" in tl:
+                key = "xgp"
+            elif "egs" in tl or "epic" in tl:
+                key = "egs"
+            elif "ps plus" in tl or "ps+" in tl:
+                key = "psplus"
+            if key:
+                current_key = key
+                current = sections[key] = soup.new_tag("div")
+                continue
+        if current_key and current is not None:
+            current.append(node)
+    return sections
+
+def parse_xgp(section: BeautifulSoup) -> List[Dict]:
+    items = []
+    for a in section.find_all("a", href=True):
+        href = a["href"]
+        m = MS_STORE_RE.search(href)
+        title = clean_text(a.get_text())
+        if m or title:
+            productId = m.group(1) if m else None
+            items.append({"title": title, "productId": productId})
+    uniq, seen = [], set()
+    for it in items:
+        k = xgp_key(it)
+        if k not in seen:
+            uniq.append(it); seen.add(k)
+    return uniq
+
+def parse_egs(section: BeautifulSoup) -> List[Dict]:
+    items = []
+    for a in section.find_all("a", href=True):
+        if not EPIC_RE.search(a["href"]):
+            continue
+        title = clean_text(a.get_text()) or clean_text(a.get("title"))
+        items.append({"title": title, "start": ""})
+    uniq, seen = [], set()
+    for it in items:
+        k = egs_key(it)
+        if k not in seen:
+            uniq.append(it); seen.add(k)
+    return uniq
+
+def parse_psplus(section: BeautifulSoup, post_title: str) -> Optional[Dict]:
+    a = section.find("a", href=PSBLOG_RE)
+    url = a["href"] if a else ""
+    m = re.search(r"(\d{2})-(\d{2})-(\d{4})", post_title)
+    iso = ""
+    if m:
+        d, mth, y = m.group(1), m.group(2), m.group(3)
+        iso = f"{y}-{mth}-{d}"
+    return {"url": url, "date": iso}
+
+# ---------------- Main backfill ----------------
+def backfill():
+    # Use the same env your main script uses; GhostAdmin will read them internally or
+    # you can pass them explicitly if your class expects (base_url, admin_key).
+    ghost = GhostAdmin(
+        admin_url=os.environ.get("GHOST_ADMIN_URL", "").rstrip("/") + "/",
+        admin_key=os.environ.get("GHOST_ADMIN_KEY", "")
+    )
+    store = Storage()
+
+    posts = list_recap_posts(ghost)
+    print(f"Found {len(posts)} recap posts.")
+
+    total_xgp = total_egs = total_ps = 0
+
+    dedup = []
+
+    for p in posts:
+        
+        pid = p["id"]
+        title = p.get("title") or ""
+        html_body = p.get("html") or ""
+
+        soup = BeautifulSoup(html_body, "html.parser")
+        sections = extract_sections(soup)
+
+        for it in parse_xgp(sections.get("xgp", BeautifulSoup("", "html.parser"))):
+            key = xgp_key(it)
+            if not key in dedup: 
+                store.remember("xgp", key, pid); total_xgp += 1
+                dedup.append(key)
+
+        # for it in parse_egs(sections.get("egs", BeautifulSoup("", "html.parser"))):
+        #     store.remember("egs", egs_key(it), pid); total_egs += 1
+        # if "psplus" in sections:
+        #     item = parse_psplus(sections["psplus"], title)
+        #     store.remember("psplus", psplus_key(item), pid); total_ps += 1
+
+        print(f"Backfilled from: {title}")
+
+    print(f"Done. Inserted ~ XGP:{total_xgp} | EGS:{total_egs} | PS+:{total_ps}")
+
+if __name__ == "__main__":
+    backfill()
diff --git a/keys.py b/keys.py
new file mode 100644
index 0000000..973ea51
--- /dev/null
+++ b/keys.py
@@ -0,0 +1,20 @@
+# keys.py (or inline in your main)
+def xgp_key(item) -> str:
+    # Prefer stable Microsoft Store productId if present; fallback to normalized title.
+    pid = (item.get("productId") or "").strip()
+    if pid:
+        return f"item:xgp:{pid}"
+    title = (item.get("title") or "").strip().lower()
+    return f"item:xgp:title:{title}"
+
+def egs_key(item) -> str:
+    # Use title + start window (your fetcher usually knows the free-week start)
+    title = (item.get("title") or "").strip()
+    start = (item.get("start") or "").strip()  # ISO or YYYY-MM-DD
+    return f"item:egs:{title}|{start}"
+
+def psplus_key(item) -> str:
+    # Use official PS Blog URL + the published month (or your computed date)
+    url = (item.get("url") or "").strip()
+    date = (item.get("date") or "").strip()
+    return f"item:psplus:{url}|{date}"
diff --git a/presquegratos.py b/presquegratos.py
index 7d4f3f1..b7f12bb 100644
--- a/presquegratos.py
+++ b/presquegratos.py
@@ -21,10 +21,13 @@ from bs4 import BeautifulSoup
 from logging.handlers import RotatingFileHandler
 from datetime import datetime, timedelta, timezone
 import zoneinfo
+from storage import Storage
+from keys import xgp_key
 
 LOG = logging.getLogger("bot_weekly")
 LOG_PATTERN = logging.Formatter("%(asctime)s:%(levelname)s: [%(filename)s] %(message)s")
 
+
 def setuplogger():
     stream_handler = logging.StreamHandler()
     stream_handler.setFormatter(LOG_PATTERN)
@@ -505,6 +508,7 @@ def build_html(egs: List[Dict[str, Any]],
         parts.append("</div>")
 
     # --- Xbox Game Pass
+    
     parts.append("<h3>🟩 Xbox Game Pass — Récemment ajoutés</h3>")
     if not xgp:
         parts.append("<p>Pas d'entrées détectées.</p>")
@@ -568,6 +572,23 @@ async def run_weekly():
         LOG.warning("XGP fetch failed: %s", e)
 
     # Build HTML
+    store = Storage()
+
+    def keep_new(items, key_fn):
+        fresh = []
+        for it in items:
+            k = key_fn(it)
+            
+            if not store.seen(k):
+
+                it["_dedup_key"] = k
+                fresh.append(it)
+        return fresh
+
+    # Fetch your three sources as you already do:
+    
+    xgp_items  = keep_new(xgp_items, xgp_key)
+
     html_body, feature = build_html(egs_items, psplus_data, xgp_items)
     # Title (FR)
     start = (datetime.now(TZ) - timedelta(days=6)).strftime("%d/%m/%Y")
diff --git a/storage.py b/storage.py
new file mode 100644
index 0000000..9a7a540
--- /dev/null
+++ b/storage.py
@@ -0,0 +1,55 @@
+# storage.py
+from __future__ import annotations
+import sqlite3, pathlib, datetime as dt
+from typing import Optional, Iterable, Tuple
+import os
+DB_PATH = "/data/published.db"  # bind-mount ./data:/data in docker
+
+_SCHEMA = """
+PRAGMA journal_mode = WAL;
+CREATE TABLE IF NOT EXISTS published_items(
+  platform TEXT NOT NULL,            -- e.g. xgp | egs | psplus
+  key      TEXT PRIMARY KEY,         -- your dedupe key (see below)
+  first_seen_utc TEXT NOT NULL,      -- ISO-8601
+  last_post_id   TEXT                -- Ghost post id that recorded it
+);
+CREATE INDEX IF NOT EXISTS idx_platform ON published_items(platform);
+"""
+
+class Storage:
+    def __init__(self, db_path: str = DB_PATH):
+
+        
+        if not os.path.isfile(db_path):
+            db_path = os.environ.get("DB_FILE_FALLBACK", r"f:\workspace\Substack_JV\data\published.db")
+
+        print(db_path)
+        pathlib.Path(db_path).parent.mkdir(parents=True, exist_ok=True)
+        self.conn = sqlite3.connect(db_path)
+        self.conn.execute("PRAGMA foreign_keys = ON;")
+        for stmt in filter(None, _SCHEMA.split(";")):
+            if stmt.strip():
+                self.conn.execute(stmt)
+
+    def seen(self, key: str) -> bool:
+        cur = self.conn.execute("SELECT 1 FROM published_items WHERE key=?", (key,))
+        return cur.fetchone() is not None
+
+    def remember(self, platform: str, key: str, post_id: Optional[str]):
+        self.conn.execute(
+            "INSERT OR IGNORE INTO published_items(platform,key,first_seen_utc,last_post_id) VALUES(?,?,?,?)",
+            (platform, key, dt.datetime.utcnow().isoformat(), post_id),
+        )
+        if post_id:
+            self.conn.execute("UPDATE published_items SET last_post_id=? WHERE key=?", (post_id, key))
+        self.conn.commit()
+
+    def bulk_remember(self, platform: str, pairs: Iterable[Tuple[str, Optional[str]]]):
+        rows = [(platform, k, dt.datetime.utcnow().isoformat(), pid) for (k, pid) in pairs]
+        self.conn.executemany(
+            "INSERT OR IGNORE INTO published_items(platform,key,first_seen_utc,last_post_id) VALUES(?,?,?,?)",
+            rows
+        )
+        self.conn.commit()
+
+