diff --git a/seo_exceptions.py b/seo_exceptions.py
index d18b1dd..2ff468d 100644
--- a/seo_exceptions.py
+++ b/seo_exceptions.py
@@ -7,7 +7,13 @@ moves a matching violation into a separate "Accepted exceptions (won't fix —
 documented)" section so the headline "issues to fix" count reflects only
 genuinely-actionable items. Each waiver carries its REASON (no silent skips).
 
-Structure:   EXCEPTIONS[rule_family][slug] = "documented reason"
+Structure:   EXCEPTIONS[site][rule_family][slug] = "documented reason"
+
+El ámbito por SITIO ("en"/"es") es deliberado: los waivers se indexaban solo por
+slug, así que un slug repetido entre blogs habría aplicado en silencio la
+excepción del sitio equivocado — y ese post habría dejado de auditarse sin que
+nadie se enterase. Hoy no hay colisiones (los slugs ES van en español), pero
+nada lo impedía. Con el ámbito, los dos sitios son independientes de verdad.
 
 Scope:
   - A waiver applies ONLY to the named rule family for the named slug. Any OTHER
@@ -23,6 +29,7 @@ held to the full standard.
 """
 
 EXCEPTIONS = {
+  "en": {
     # internal_links < 2 — these posts are intentionally left as-is because there
     # is no honest 2nd internal-link target in their prose; forcing one would hurt
     # quality. (Decided 2026-06-23.)
@@ -83,17 +90,38 @@ EXCEPTIONS = {
             "(Malmstrom, Levelland, Manises) appear nowhere in the text. "
             "(Reviewed 2026-07-18.)",
     },
+  },
+
+  # El blog ES se auditó por primera vez el 2026-07-20 (al crear ghst-es) y
+  # arrastra backlog: 28 posts sin alt-text, 27 con enlazado flojo. Es deuda
+  # acumulada por no haberse revisado nunca, NO regresión — se irá waiveando o
+  # arreglando post a post, con la misma vara de medir que el EN.
+  "es": {},
 }
 
 
-def accepted_reason(rule, slug):
-    """Return the documented waiver reason if (rule, slug) is an accepted
-    exception, else None.
+def accepted_reason(rule, slug, site="en"):
+    """Devuelve la razón documentada si (rule, slug) está waiveado EN ESE SITIO.
 
-    Matches a rule-family key to a Violation whose `rule` is the key itself or
-    "<key>.<detail>" (e.g. family "internal_links" matches "internal_links.too_few").
+    Empareja la familia de reglas con una Violation cuyo `rule` sea la clave
+    misma o "<clave>.<detalle>" (la familia "internal_links" cubre
+    "internal_links.too_few").
+
+    `site` por defecto "en" para no romper a seo_audit.py, que es EN-only.
+    Quien audite el ES DEBE pasar site="es" explícitamente.
     """
-    for family, slugs in EXCEPTIONS.items():
+    for family, slugs in EXCEPTIONS.get(site, {}).items():
         if (rule == family or rule.startswith(family + ".")) and slug in slugs:
             return slugs[slug]
     return None
+
+
+def collisions():
+    """Slugs waiveados en más de un sitio. Debe estar SIEMPRE vacío: si algo
+    aparece aquí, hay una excepción ambigua que revisar a mano."""
+    from collections import Counter
+    c = Counter()
+    for fams in EXCEPTIONS.values():
+        for slugs in fams.values():
+            c.update(slugs.keys())
+    return sorted(s for s, n in c.items() if n > 1)
diff --git a/seo_finish.py b/seo_finish.py
index a274570..3d0236c 100755
--- a/seo_finish.py
+++ b/seo_finish.py
@@ -89,10 +89,10 @@ def fetch(slug, cli):
     sys.exit(f"ERROR: no encuentro el post '{slug}' en {cli}")
 
 
-def violations_of(post):
+def violations_of(post, site):
     out = []
     for v in R.check_post(post):
-        if v.severity == R.INFO or X.accepted_reason(v.rule, post.get("slug")):
+        if v.severity == R.INFO or X.accepted_reason(v.rule, post.get("slug"), site):
             continue
         out.append(v)
     return out
@@ -142,7 +142,7 @@ def cmd_prep(a):
         if v:
             print(f"       {v[:150]}")
 
-    viol = violations_of(p)
+    viol = violations_of(p, a.site)
     print(f"\n─── violaciones vigentes: {len(viol)} ───")
     for v in viol:
         print(f"  [{v.severity}] {v.rule}: {v.message}")
@@ -199,7 +199,7 @@ def cmd_apply(a):
     print("✓ aplicado")
 
     after = fetch(a.slug, cfg["cli"])
-    viol = violations_of(after)
+    viol = violations_of(after, a.site)
     if viol:
         print(f"\n⚠ el post SIGUE con {len(viol)} violación(es):")
         for v in viol:
diff --git a/seo_watch.py b/seo_watch.py
index a78e100..a09d1e8 100644
--- a/seo_watch.py
+++ b/seo_watch.py
@@ -216,14 +216,14 @@ def check_sitemap():
     return len(urls), malos
 
 
-def check_rules(posts):
+def check_rules(posts, site):
     """Violaciones accionables (excluye waivers e INFO), agrupadas por regla."""
     out = {}
     for p in posts:
         if p.get("status") != "published":
             continue
         for v in R.check_post(p):
-            if v.severity == R.INFO or X.accepted_reason(v.rule, p["slug"]):
+            if v.severity == R.INFO or X.accepted_reason(v.rule, p["slug"], site):
                 continue
             out.setdefault(v.rule, []).append(p["slug"])
     return out
@@ -287,7 +287,7 @@ def run_site(site):
     roto, nocanon, prematuro = check_links(posts)
     envejecido = check_link_aging(posts)
     n_sm, sm_malos = check_sitemap()
-    reglas = check_rules(posts)
+    reglas = check_rules(posts, site)
 
     body = build_report(roto, nocanon, prematuro, envejecido, n_sm, sm_malos, reglas)
     pub = sum(1 for p in posts if p.get("status") == "published")