fix(tts): exclude donations and accounting entries

scripts/minimax_tts.py

@@ -93,7 +93,15 @@ def get_post_text(pid):
                    check=True, capture_output=True)
     subprocess.run(["docker", "cp", f"{CONTAINER}:/tmp/fea_es.json", "/tmp/fea_es.json"], check=True)
     d = json.load(open("/tmp/fea_es.json"))
-    raw = re.sub(r"(?i)</p>|<br\s*/?>|</h[1-6]>", "\n", d["content"])
+    # Hard gate: pages and operational/accounting entries are never TTS input.
+    # This also protects explicit --ids queues, which bypass author-backlog SQL.
+    raw_content = d.get("content", "")
+    blocked_markers = ("fea-don-wrap", "fea-ledger", "Haz tu donación")
+    if d.get("post_type") != "post":
+        raise ValueError(f"post #{pid} no es un artículo (post_type={d.get('post_type')!r}); TTS excluido")
+    if d.get("post_name") == "numeros" or any(marker in raw_content for marker in blocked_markers):
+        raise ValueError(f"post #{pid} es contenido de cuentas/donaciones; TTS excluido")
+    raw = re.sub(r"(?i)</p>|<br\s*/?>|</h[1-6]>", "\n", raw_content)
     raw = re.sub(r"<[^>]+>", "", raw)
     raw = re.sub(r"\[[^\]]+\]", "", raw)
     raw = html.unescape(raw)