fix(crawler): FlareSolverr persistent sessions + DB data field + scraper schema

- flaresolverr.py: refactor stateless (create_session/destroy_session/get)
  L'ancienne approche warmup→cookies→requests était bloquée par Cloudflare
  (JA3 TLS fingerprinting). Toutes les requêtes passent maintenant via une
  session Chrome persistante FlareSolverr.

- ma_http.py: MASession utilise les sessions FS persistantes. Ajout de
  _extract_json (html.unescape + extraction <pre>) et _build_url.
  Gestion du refresh sur 403/429 avec recréation de session.

- db.py: ajout du champ data JSONB dans upsert_bands INSERT+ON CONFLICT.
  Avant ce fix, l'URL n'était jamais stockée → get_bands_to_enrich retournait
  toujours 0 résultats → enrichissement mort.

- scraper_band.py: schéma band_page aligné sur l'historique DB (lineup
  structuré avec URLs artistes, discography_url, lyrical_themes, location,
  info_raw). Ajout de "Themes" dans _pick pour couvrir les deux variantes
  de clé HTML (MA utilise parfois "Themes", parfois "Lyrical themes").

- 005_fix_checkpoints_and_data.sql: renomme last_additions_check →
  last_created_check (clé morte vs clé utilisée par le code). Backfill
  enriched=true + colonnes top-level (formed_year, themes, status,
  location_text) depuis band_page JSONB pour les 85k bands de l'ancien scraper.

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

apps/api/migrations/005_fix_checkpoints_and_data.sql

@@ -0,0 +1,47 @@
+-- 005: corriger les checkpoints + backfill enriched depuis les données de l'ancien scraper
+
+-- 1. Renommer la clé checkpoint morte 'last_additions_check' → 'last_created_check'
+--    (le code utilise last_{order_by}_check donc 'last_created_check', pas 'last_additions_check')
+INSERT INTO crawl_checkpoint (key, value)
+  VALUES ('last_created_check', NULL)
+  ON CONFLICT (key) DO NOTHING;
+
+DELETE FROM crawl_checkpoint WHERE key = 'last_additions_check';
+
+-- 2. Backfill depuis les données de l'ancien scraper :
+--    85 829 bands ont data->'band_page' rempli mais enriched=false et colonnes top-level vides.
+--    On remplit les colonnes à partir du JSONB existant.
+UPDATE bands SET
+  enriched   = true,
+  status     = COALESCE(
+                 NULLIF(trim(status), ''),
+                 data->'band_page'->>'status'
+               ),
+  genre      = COALESCE(
+                 NULLIF(trim(genre), ''),
+                 data->'band_page'->>'genre',
+                 data->'band_page'->'info_raw'->>'genre'
+               ),
+  themes     = COALESCE(
+                 NULLIF(trim(themes), ''),
+                 NULLIF(data->'band_page'->>'lyrical_themes', 'N/A'),
+                 NULLIF(data->'band_page'->'info_raw'->>'themes', 'N/A')
+               ),
+  formed_year = COALESCE(
+                 formed_year,
+                 CASE
+                   WHEN (data->'band_page'->>'formed_in') ~ '^\d{4}$'
+                     THEN (data->'band_page'->>'formed_in')::int
+                   WHEN (data->'band_page'->>'formed_in') IS NOT NULL
+                     THEN (regexp_match(data->'band_page'->>'formed_in', '\d{4}'))[1]::int
+                   WHEN (data->'band_page'->'info_raw'->>'formed in') IS NOT NULL
+                     THEN (regexp_match(data->'band_page'->'info_raw'->>'formed in', '\d{4}'))[1]::int
+                   ELSE NULL
+                 END
+               ),
+  location_text = COALESCE(
+                    NULLIF(trim(location_text), ''),
+                    data->'band_page'->>'location',
+                    data->'band_page'->'info_raw'->>'location'
+                  )
+WHERE data->'band_page' IS NOT NULL;

apps/crawler/src/db.py

@@ -1,6 +1,7 @@
 """
 Écriture directe en DB. Toutes les opérations passent par des upserts idempotents.
 """
+import json
 import logging
 from contextlib import contextmanager
 from datetime import datetime, timezone
@@ -62,13 +63,14 @@ def upsert_bands(bands: List[Dict[str, Any]]) -> Dict[str, int]:
             b.get("ma_modified_at"),
             ts,   # first_seen_at (ignoré si déjà set)
             ts,   # created_at (ignoré si déjà set)
+            psycopg2.extras.Json(b.get("data") or {}),
         ))
 
     sql = """
         INSERT INTO bands
           (ma_id, name, country, location_text, status, genre, formed_year,
            themes, enriched, crawled_at, crawled_hash, ma_created_at, ma_modified_at,
-           first_seen_at, created_at)
+           first_seen_at, created_at, data)
         VALUES %s
         ON CONFLICT (ma_id) DO UPDATE SET
           name          = COALESCE(EXCLUDED.name, bands.name),
@@ -82,7 +84,8 @@ def upsert_bands(bands: List[Dict[str, Any]]) -> Dict[str, int]:
           crawled_at    = COALESCE(EXCLUDED.crawled_at, bands.crawled_at),
           crawled_hash  = COALESCE(EXCLUDED.crawled_hash, bands.crawled_hash),
           ma_created_at = COALESCE(EXCLUDED.ma_created_at, bands.ma_created_at),
-          ma_modified_at= COALESCE(EXCLUDED.ma_modified_at, bands.ma_modified_at)
+          ma_modified_at= COALESCE(EXCLUDED.ma_modified_at, bands.ma_modified_at),
+          data          = bands.data || EXCLUDED.data
         RETURNING (xmax = 0) AS was_inserted
     """
 

apps/crawler/src/flaresolverr.py

@@ -1,16 +1,17 @@
 """
-Client FlareSolverr pour bypasser Cloudflare.
+Client FlareSolverr — interface stateless.
 
 Workflow :
   1. fs = FlareSolverr(url)
-  2. fs.warmup(ma_url)         → solve le challenge CF, stocke les cookies
-  3. session = fs.build_session() → requests.Session avec les cookies CF
-  4. Toutes les requêtes suivantes passent par session (requests standard, rapide)
-  5. Si 403 : appeler fs.warmup() à nouveau pour rafraîchir les cookies
+  2. session_id = fs.create_session()
+  3. fs.get(warmup_url, session_id)      → résout le challenge CF
+  4. status, body = fs.get(ajax_url, session_id) → réutilise le même Chrome
+  5. fs.destroy_session(session_id)      → libère la ressource Chrome
 """
 import logging
+from typing import Optional, Tuple
+
 import requests
-from typing import Dict, List, Optional
 
 log = logging.getLogger(__name__)
 
@@ -23,42 +24,6 @@ class FlareSolverr:
     def __init__(self, base_url: str, timeout_ms: int = 90_000):
         self.base_url = base_url.rstrip("/")
         self.timeout_ms = timeout_ms
-        self._cookies: List[Dict] = []
-        self._user_agent: Optional[str] = None
-
-    # ------------------------------------------------------------------
-    # Public
-    # ------------------------------------------------------------------
-
-    def warmup(self, url: str):
-        """Charge une page MA pour résoudre le challenge CF et stocker les cookies."""
-        log.info(f"[fs] warmup on {url}")
-        resp = self._request("request.get", url)
-        solution = resp.get("solution", {})
-        self._cookies = solution.get("cookies", [])
-        self._user_agent = solution.get("userAgent")
-        log.info(f"[fs] warmup ok, got {len(self._cookies)} cookies, ua={self._user_agent[:40] if self._user_agent else None}")
-
-    def build_session(self) -> requests.Session:
-        """Retourne un requests.Session avec les cookies et user-agent CF."""
-        if not self._cookies:
-            raise FlareSolverrError("No cookies — call warmup() first")
-
-        s = requests.Session()
-        s.headers.update({
-            "User-Agent": self._user_agent or "Mozilla/5.0",
-            "Accept-Language": "en-US,en;q=0.9",
-            "Accept": "text/html,application/xhtml+xml,application/xml;q=0.9,*/*;q=0.8",
-            "Referer": "https://www.metal-archives.com/",
-        })
-        for c in self._cookies:
-            s.cookies.set(c["name"], c["value"], domain=c.get("domain", ""), path=c.get("path", "/"))
-        return s
-
-    def get_raw(self, url: str) -> str:
-        """Fait une requête via FlareSolverr et retourne le HTML rendu."""
-        resp = self._request("request.get", url)
-        return resp.get("solution", {}).get("response", "")
 
     def healthy(self) -> bool:
         try:
@@ -67,19 +32,47 @@ class FlareSolverr:
         except Exception:
             return False
 
-    # ------------------------------------------------------------------
-    # Internal
-    # ------------------------------------------------------------------
+    def create_session(self) -> str:
+        """Ouvre une session Chrome persistante dans FlareSolverr, retourne le session_id."""
+        data = self._call({"cmd": "sessions.create"})
+        session_id = data["session"]
+        log.info(f"[fs] created session {session_id}")
+        return session_id
 
-    def _request(self, cmd: str, url: str) -> Dict:
-        payload = {"cmd": cmd, "url": url, "maxTimeout": self.timeout_ms}
+    def destroy_session(self, session_id: str):
+        """Libère la session Chrome."""
         try:
-            r = requests.post(f"{self.base_url}/v1", json=payload, timeout=self.timeout_ms / 1000 + 10)
+            self._call({"cmd": "sessions.destroy", "session": session_id})
+            log.info(f"[fs] destroyed session {session_id}")
+        except Exception as e:
+            log.warning(f"[fs] destroy session {session_id} failed: {e}")
+
+    def get(self, url: str, session_id: Optional[str] = None) -> Tuple[int, str]:
+        """
+        GET via FlareSolverr. Retourne (http_status_code, response_body).
+        Si session_id fourni, réutilise le Chrome existant (pas de re-warmup).
+        """
+        payload: dict = {"cmd": "request.get", "url": url, "maxTimeout": self.timeout_ms}
+        if session_id:
+            payload["session"] = session_id
+        data = self._call(payload)
+        sol = data.get("solution", {})
+        status = sol.get("status", 0)
+        body = sol.get("response", "")
+        log.debug(f"[fs] GET {url} → {status}")
+        return status, body
+
+    def _call(self, payload: dict) -> dict:
+        try:
+            r = requests.post(
+                f"{self.base_url}/v1",
+                json=payload,
+                timeout=self.timeout_ms / 1000 + 10,
+            )
             r.raise_for_status()
             data = r.json()
         except Exception as e:
             raise FlareSolverrError(f"FlareSolverr request failed: {e}") from e
-
         if data.get("status") != "ok":
             raise FlareSolverrError(f"FlareSolverr error: {data.get('message', data)}")
         return data

apps/crawler/src/ma_http.py

@@ -1,79 +1,88 @@
 """
-Session HTTP vers Metal Archives.
+Session HTTP vers Metal Archives via FlareSolverr persistent sessions.
 
-Combine FlareSolverr (pour les cookies CF) et requests (pour les appels AJAX rapides).
-Si un appel retourne 403/429, on rafraîchit automatiquement les cookies.
+Cloudflare bloque les requêtes `requests` directes (fingerprint TLS/JA3).
+Toutes les requêtes passent par FlareSolverr qui utilise un Chrome headless réel.
+Une session Chrome est créée une fois et réutilisée pour toute la durée du crawl
+(refresh automatique toutes les 4h ou sur 403/429).
 """
+import json
 import logging
+import re
 import time
+from html import unescape as html_unescape
 from typing import Any, Dict, Optional
+from urllib.parse import urlencode
 
-import requests
-
-from .config import MA_BASE, FS_TIMEOUT_MS, AJAX_PAGE_SIZE, LIST_MIN_DELAY, LIST_MAX_DELAY
+from .config import MA_BASE, AJAX_PAGE_SIZE, LIST_MIN_DELAY, LIST_MAX_DELAY
 from .flaresolverr import FlareSolverr, FlareSolverrError
 from .polite import sleep_range
 
 log = logging.getLogger(__name__)
 
 _WARMUP_URL = f"{MA_BASE}/lists/FR"
+_SESSION_TTL = 3600 * 4  # refresh session toutes les 4h
 
 
 class MASession:
     def __init__(self, fs: FlareSolverr):
         self.fs = fs
-        self._session: Optional[requests.Session] = None
-        self._cookie_age = 0.0
+        self._session_id: Optional[str] = None
+        self._session_age = 0.0
 
     # ------------------------------------------------------------------
     # Session management
     # ------------------------------------------------------------------
 
     def ensure_session(self, force_refresh: bool = False):
-        age = time.time() - self._cookie_age
-        if self._session is None or force_refresh or age > 3600 * 4:  # refresh every 4h
-            log.info("[ma_http] acquiring Cloudflare cookies via FlareSolverr...")
-            self.fs.warmup(_WARMUP_URL)
-            self._session = self.fs.build_session()
-            self._cookie_age = time.time()
-            log.info("[ma_http] session ready")
+        age = time.time() - self._session_age
+        if self._session_id is None or force_refresh or age > _SESSION_TTL:
+            if self._session_id:
+                self.fs.destroy_session(self._session_id)
+                self._session_id = None
+            log.info("[ma_http] creating FlareSolverr persistent session...")
+            self._session_id = self.fs.create_session()
+            log.info(f"[ma_http] warming up CF cookies on {_WARMUP_URL}")
+            status, _ = self.fs.get(_WARMUP_URL, self._session_id)
+            log.info(f"[ma_http] warmup done (status={status}), session ready")
+            self._session_age = time.time()
 
     def get_json(self, url: str, params: Dict = None, retries: int = 2) -> Any:
         """GET JSON depuis un endpoint AJAX MA."""
         self.ensure_session()
+        full_url = _build_url(url, params)
         for attempt in range(retries + 1):
-            try:
-                r = self._session.get(url, params=params, timeout=30)
-                if r.status_code in (403, 429, 503) and attempt < retries:
-                    log.warning(f"[ma_http] {r.status_code} on {url}, refreshing cookies...")
+            status, body = self.fs.get(full_url, self._session_id)
+            if status in (403, 429, 503) and attempt < retries:
+                log.warning(f"[ma_http] {status} on {url}, refreshing session...")
                 self.ensure_session(force_refresh=True)
                 sleep_range(LIST_MIN_DELAY * 2, LIST_MAX_DELAY * 2)
                 continue
-                r.raise_for_status()
-                return r.json()
-            except requests.exceptions.JSONDecodeError:
+            if status != 200:
+                raise RuntimeError(f"HTTP {status} on {full_url}")
+            try:
+                return _extract_json(body)
+            except (json.JSONDecodeError, ValueError):
                 log.warning(f"[ma_http] non-JSON response from {url}")
                 raise
         raise RuntimeError(f"Failed after {retries} retries: {url}")
 
     def get_html(self, url: str, retries: int = 2) -> str:
-        """GET HTML brut (pages band, etc.)."""
+        """GET HTML rendu (pages band, etc.)."""
         self.ensure_session()
         for attempt in range(retries + 1):
-            try:
-                r = self._session.get(url, timeout=30)
-                if r.status_code in (403, 429, 503) and attempt < retries:
-                    log.warning(f"[ma_http] {r.status_code} on {url}, refreshing...")
+            status, body = self.fs.get(url, self._session_id)
+            if status in (403, 429, 503) and attempt < retries:
+                log.warning(f"[ma_http] {status} on {url}, refreshing session...")
                 self.ensure_session(force_refresh=True)
                 sleep_range(LIST_MIN_DELAY * 2, LIST_MAX_DELAY * 2)
                 continue
-                r.raise_for_status()
-                return r.text
-            except Exception:
+            if status != 200:
                 if attempt < retries:
                     sleep_range(3, 6)
                     continue
-                raise
+                raise RuntimeError(f"HTTP {status} on {url}")
+            return body
         raise RuntimeError(f"Failed after {retries} retries: {url}")
 
     # ------------------------------------------------------------------
@@ -154,7 +163,6 @@ class MASession:
                 parsed = _parse_archive_row(row, order_by)
                 if not parsed:
                     continue
-                # Si on a déjà vu jusqu'à cette date, on s'arrête
                 if since_date and parsed.get("date_str") and parsed["date_str"] <= since_date:
                     stop = True
                     break
@@ -172,13 +180,29 @@ class MASession:
 
 
 # ------------------------------------------------------------------
-# Row parsers
+# Helpers internes
 # ------------------------------------------------------------------
 
-import re
-from typing import Optional as Opt
+def _build_url(base: str, params: Dict = None) -> str:
+    if not params:
+        return base
+    return f"{base}?{urlencode(params)}"
 
 
+def _extract_json(body: str) -> Any:
+    """
+    FlareSolverr wraps les réponses JSON dans <pre> avec HTML-encoding des <>.
+    On extrait le contenu, on unescape les entités HTML, puis on parse le JSON.
+    """
+    m = re.search(r"<pre[^>]*>([\s\S]*?)</pre>", body)
+    text = html_unescape(m.group(1)) if m else body
+    return json.loads(text)
+
+
+# ------------------------------------------------------------------
+# Row parsers
+# ------------------------------------------------------------------
+
 def _extract_link(html_str: str):
     """Extrait href et texte d'un fragment HTML '<a href="...">text</a>'."""
     m = re.search(r'href=["\']([^"\']+)["\']', html_str or "")
@@ -187,7 +211,7 @@ def _extract_link(html_str: str):
     return href, text
 
 
-def _extract_ma_id(url: str) -> Opt[int]:
+def _extract_ma_id(url: str) -> Optional[int]:
     m = re.search(r"/bands/[^/]+/(\d+)", url or "")
     return int(m.group(1)) if m else None
 
@@ -196,10 +220,8 @@ def _clean(s) -> str:
     return re.sub(r"<[^>]+>", "", str(s or "")).strip()
 
 
-def _parse_list_row(row, country_code: str) -> Opt[Dict]:
-    """
-    Ligne du listing pays : [name_html, genre, location, status]
-    """
+def _parse_list_row(row, country_code: str) -> Optional[Dict]:
+    """Ligne du listing pays : [name_html, genre, location, status]"""
     if len(row) < 3:
         return None
     href, name = _extract_link(row[0])
@@ -217,10 +239,8 @@ def _parse_list_row(row, country_code: str) -> Opt[Dict]:
     }
 
 
-def _parse_archive_row(row, order_by: str) -> Opt[Dict]:
-    """
-    Ligne des archives : [name_html, country_html, genre, date_str]
-    """
+def _parse_archive_row(row, order_by: str) -> Optional[Dict]:
+    """Ligne des archives : [name_html, country_html, genre, date_str]"""
     if len(row) < 4:
         return None
     href, name = _extract_link(row[0])
@@ -228,8 +248,7 @@ def _parse_archive_row(row, order_by: str) -> Opt[Dict]:
     if not ma_id or not name:
         return None
 
-    country_href, country_code = _extract_link(row[1])
-    # country_href est genre /lists/FR → extraire le code
+    country_href, _ = _extract_link(row[1])
     m = re.search(r"/lists/([A-Z]{1,3})", country_href or "")
     cc = m.group(1) if m else _clean(row[1])
 
@@ -240,5 +259,5 @@ def _parse_archive_row(row, order_by: str) -> Opt[Dict]:
         "country": cc,
         "genre": _clean(row[2]),
         "date_str": _clean(row[3]),
-        "date_type": order_by,  # 'created' ou 'modified'
+        "date_type": order_by,
     }

apps/crawler/src/scraper_band.py

@@ -1,5 +1,8 @@
 """
 Parser d'une page individuelle de band Metal Archives.
+
+Produit un dict compatible avec upsert_band_enriched() ET cohérent avec
+le schéma band_page historique stocké en DB (lineup structuré, discography_url, etc.)
 """
 import hashlib
 import re
@@ -7,85 +10,96 @@ from typing import Any, Dict, List, Optional
 
 from bs4 import BeautifulSoup
 
+# Section header → clé lineup
+_LINEUP_SECTIONS = {
+    "current": "current",
+    "past":    "past",
+    "live":    "live",
+    "session": "session",
+}
+
 
 def parse_band_page(html: str) -> Dict[str, Any]:
     soup = BeautifulSoup(html, "lxml")
     out: Dict[str, Any] = {}
 
-    title = soup.find("title")
-    out["title"] = _text(title)
-
-    h1 = (
-        soup.find("h1", class_=re.compile(r"band_name", re.I))
-        or soup.find("h1")
-    )
+    h1 = soup.find("h1", class_=re.compile(r"band_name", re.I)) or soup.find("h1")
     out["name"] = _text(h1)
 
-    # Bloc #band_stats : dt/dd pairs
-    stats = soup.find("div", id="band_stats") or soup.find("div", id="band_info") or soup
+    # --- band_stats : dt/dd pairs ---
+    stats_div = soup.find("div", id="band_stats") or soup.find("div", id="band_info") or soup
     kv: Dict[str, str] = {}
-    for dl in stats.find_all("dl"):
+    for dl in stats_div.find_all("dl"):
         for dt in dl.find_all("dt"):
             dd = dt.find_next_sibling("dd")
             k = _text(dt).rstrip(":")
-            v = _text(dd)
+            v = _text(dd) if dd else None
             if k and v:
                 kv[k] = v
+    out["info_raw"] = {k.lower(): v for k, v in kv.items()}
 
-    out["info"] = kv
     out["status"]       = _pick(kv, "Status")
-    out["formed_in"]    = _pick(kv, "Formed in", "Formed")
     out["genre"]        = _pick(kv, "Genre")
-    out["themes"]       = _pick(kv, "Lyrical themes", "Lyrical Themes")
-    out["label"]        = _pick(kv, "Current label", "Label")
+    out["formed_in"]    = _pick(kv, "Formed in", "Formed")
     out["years_active"] = _pick(kv, "Years active")
+    out["location"]     = _pick(kv, "Location")
+    # MA uses either "Lyrical themes", "Lyrical Themes", or simply "Themes"
+    out["lyrical_themes"] = _pick(kv, "Lyrical themes", "Lyrical Themes", "Themes")
+    # Alias for upsert_band_enriched which reads data.get("themes")
+    out["themes"]       = out["lyrical_themes"]
+    out["label"]        = _pick(kv, "Current label", "Last label", "Label")
 
-    # Dates "Added on" / "Last modified" (souvent dans #auditTrail ou en bas de page)
-    audit = soup.find(id="auditTrail") or soup.find("div", class_=re.compile(r"audit", re.I))
-    if audit:
-        trail_text = _text(audit)
-        m_added = re.search(r"Added on:\s*([^\n,]+)", trail_text, re.I)
-        m_modified = re.search(r"Last modified on:\s*([^\n,]+)", trail_text, re.I)
-        out["ma_created_at"] = m_added.group(1).strip() if m_added else None
-        out["ma_modified_at"] = m_modified.group(1).strip() if m_modified else None
-
-    # Membres
-    members: List[Dict] = []
-    for table in soup.find_all("table"):
-        cls = " ".join(table.get("class") or [])
-        if "lineupTable" not in cls:
-            continue
-        section_el = table.find_previous(["h2", "h3"])
-        section = _text(section_el) if section_el else None
-        headers = [_text(th) for th in table.find_all("th")]
+    # --- Lineup (grouped by section, with artist URLs) ---
+    lineup: Dict[str, List[Dict]] = {"current": [], "past": [], "live": [], "session": []}
+    for table in soup.find_all("table", class_="lineupTable"):
+        section_key = _lineup_section(table)
+        bucket = lineup.setdefault(section_key, [])
         for tr in table.find_all("tr")[1:]:
             tds = tr.find_all(["td", "th"])
             if not tds:
                 continue
-            cells = [_text(td) for td in tds]
-            row = dict(zip(headers, cells)) if headers and len(headers) == len(cells) else {"cells": cells}
-            if section:
-                row["section"] = section
-            members.append(row)
-    if members:
-        out["members"] = members
+            a = tds[0].find("a", href=re.compile(r"/artists/"))
+            if not a:
+                continue
+            role = _text(tds[1]) if len(tds) > 1 else None
+            bucket.append({
+                "name": _text(a),
+                "url":  a.get("href", ""),
+                "role": role,
+            })
+    out["lineup"] = lineup
 
-    # Liens externes
+    # --- Discography URL ---
+    disc = soup.find("a", href=re.compile(r"/band/discography/"))
+    out["discography_url"] = disc["href"] if disc else None
+
+    # --- Liens externes ---
     links_div = soup.find("div", id="band_links")
-    if links_div:
     out["links"] = [
         {"text": _text(a), "url": a["href"]}
         for a in links_div.find_all("a", href=True)
-        ]
+    ] if links_div else []
+
+    # --- Dates audit trail ---
+    audit = soup.find(id="auditTrail") or soup.find("div", class_=re.compile(r"audit", re.I))
+    if audit:
+        trail = _text(audit)
+        m_added    = re.search(r"Added on:\s*(\S+\s+\S+)", trail, re.I)
+        m_modified = re.search(r"Last modified on:\s*(\S+\s+\S+)", trail, re.I)
+        out["ma_created_at"]  = m_added.group(1).strip()    if m_added    else None
+        out["ma_modified_at"] = m_modified.group(1).strip() if m_modified else None
 
     return out
 
 
 def page_hash(html: str) -> str:
-    """MD5 du HTML pour détecter les changements."""
     return hashlib.md5(html.encode()).hexdigest()
 
 
+# ------------------------------------------------------------------
+# Helpers
+# ------------------------------------------------------------------
+
 def _text(el) -> str:
     return (el.get_text(" ", strip=True) if el else "").strip()
 
@@ -95,3 +109,15 @@ def _pick(kv: Dict, *keys: str) -> Optional[str]:
         if k in kv:
             return kv[k]
     return None
+
+
+def _lineup_section(table) -> str:
+    """Déduit la section lineup depuis le h2/h3 qui précède la table."""
+    el = table.find_previous(["h2", "h3"])
+    if not el:
+        return "current"
+    text = _text(el).lower()
+    for kw, key in _LINEUP_SECTIONS.items():
+        if kw in text:
+            return key
+    return "current"