From 2af654d3445f9dae89a9feba4d42dff139303a6f Mon Sep 17 00:00:00 2001
From: ChemaVX <jmivanez@gmail.com>
Date: Tue, 21 Jul 2026 07:15:45 +0000
Subject: [PATCH] vendor: sincroniza el motor SEO (stopwords ES + tokenizador
 sin acentos)

make sync-seo tras chemavx-seo-tools 31c9d3e. Solo afecta a topic_collision;
las RULES de check_post no usan el tokenizador. Los 9 tests de seo pasan.
---
 src/seo/.rules.sha256 |  2 +-
 src/seo/rules.py      | 39 ++++++++++++++++++++++++++++++++++++++-
 2 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/src/seo/.rules.sha256 b/src/seo/.rules.sha256
index 287383b..6e62471 100644
--- a/src/seo/.rules.sha256
+++ b/src/seo/.rules.sha256
@@ -1 +1 @@
-73872aaf4079e5a0b690fbbe7de7e7967edc2297c8bad79142308714452c60ef
+d373af317c287d1516d4487e1bcb874e98d0b76a402b9115c59280cef0050b97
diff --git a/src/seo/rules.py b/src/seo/rules.py
index 9e5e850..ad1a139 100644
--- a/src/seo/rules.py
+++ b/src/seo/rules.py
@@ -30,6 +30,7 @@ rule, write a function (post) -> list[Violation] and append it to RULES. The aud
 and the validator both just call check_post(); they never re-implement a check.
 """
 import re
+import unicodedata
 from collections import namedtuple
 
 SITE_HOST = "theexclusionzone.com"
@@ -218,6 +219,33 @@ TOPIC_STOPWORDS = {
     "unexplained", "encounter", "sighting", "sightings", "alien", "aliens",
     "phenomena", "aerial", "unidentified", "extraordinary", "americas",
     "american", "video", "footage",
+    # spanish glue (added 2026-07-21 with the ES site — zonadeexclusion.com).
+    # Without these, "que"/"los"/"del" counted as case identity: Kenneth Arnold
+    # 1947 "collided" with Roswell 1947 on nothing but «que» + the shared year.
+    # Cost: "los" no longer identifies Los Alamos on EN — "alamos" still does,
+    # and the EN corpus reports the same collisions before and after.
+    "los", "las", "una", "unos", "unas", "del", "por", "para", "con", "sin",
+    "sus", "que", "cual", "cuales", "quien", "quienes", "donde", "cuando",
+    "como", "pero", "porque", "aunque", "sobre", "entre", "hasta", "desde",
+    "tras", "ante", "bajo", "durante", "segun", "este", "esta", "esto",
+    "estos", "estas", "ese", "esa", "eso", "esos", "esas", "aquel", "aquella",
+    "otro", "otra", "otros", "otras", "todo", "toda", "todos", "todas",
+    "mismo", "misma", "cada", "algo", "alguien", "nada", "nadie", "mas", "muy",
+    "aun", "solo", "tambien", "siempre", "nunca", "jamas", "casi", "menos",
+    "fue", "fueron", "era", "eran", "ser", "son", "estan", "estaba",
+    "estaban", "haber", "habia", "han", "hay", "hizo", "hacer", "hace",
+    "tiene", "tienen", "tenia", "puede", "pueden", "podria", "sigue",
+    "siguen", "sabe", "dice", "dicen", "ano", "anos", "dia", "dias", "vez",
+    "veces", "despues", "antes", "hoy", "ahora",
+    # domain-generic ES — mirror of the English block above
+    "ovni", "ovnis", "fenomeno", "fenomenos", "caso", "casos", "incidente",
+    "incidentes", "misterio", "misterios", "expediente", "expedientes",
+    "archivo", "archivos", "documento", "documentos", "desclasificado",
+    "desclasificados", "desclasificacion", "gobierno", "militar", "militares",
+    "ejercito", "secreto", "secretos", "investigacion", "testigo", "testigos",
+    "avistamiento", "avistamientos", "encuentro", "encuentros",
+    "extraterrestre", "extraterrestres", "alienigena", "alienigenas",
+    "inexplicable", "inexplicables", "aereo", "aerea", "videos",
 }
 # 0.70 calibrated 2026-07-10: the "When Nuclear Weapons Go / Arsenal Went
 # Silent" near-twin pair scores 0.742 (char-level penalizes weapons/arsenal);
@@ -232,8 +260,17 @@ NEWS_YEAR_MIN = 2020
 _YEAR_RE = re.compile(r"\b(19|20)\d{2}\b")
 
 
+def _deaccent(text):
+    """Fold accents to ASCII. Required for Spanish: the [a-z0-9]+ tokenizer
+    SPLITS on any accented char, so "Pentágono" became {pent, gono} and
+    "Fenómenos" became {fen, menos} — 3-char garbage that no stopword list can
+    ever cover, and that never matched the (already accent-free) Ghost slug.
+    No-op on EN, whose only non-ASCII are dashes and curly apostrophes."""
+    return unicodedata.normalize("NFKD", text).encode("ascii", "ignore").decode()
+
+
 def _tokens(text):
-    return set(re.findall(r"[a-z0-9]+", _s(text).lower()))
+    return set(re.findall(r"[a-z0-9]+", _deaccent(_s(text)).lower()))
 
 
 def _case_years(title, slug):