diff --git a/evalharness/cli.py b/evalharness/cli.py
index d1f9f61..000ede1 100644
--- a/evalharness/cli.py
+++ b/evalharness/cli.py
@@ -546,11 +546,17 @@ def _compose_judge_spec(args):
     url = getattr(args, 'judge_api_url', '') or ''
     provider = getattr(args, 'judge_provider', 'openai-chat') or 'openai-chat'
     internal = {'openai-chat': 'openai', 'openai-pool': 'openai-pool'}.get(provider, provider)
-    if judge and not url and '/' not in judge:
-        raise SystemExit(
-            f'--judge-model {judge!r} needs --judge-api-url (or pass a full '
-            f'spec like openai/http://host:8000/v1?{judge})')
-    if url and judge and '/' not in judge:
+    if judge and not url:
+        if '/' not in judge and '?' not in judge:
+            raise SystemExit(
+                f'--judge-model {judge!r} needs --judge-api-url (or pass a '
+                f'full spec like openai/http://host:8000/v1?model)')
+        return judge   # full legacy spec, used as-is
+    if url and judge:
+        # url+model ALWAYS combines (same as the main model flags): a bare
+        # model name may itself contain slashes (/data/hf_models/...), which
+        # the old 'no slash' heuristic misread as a full spec and passed
+        # through un-prefixed
         judge = f'{internal}/{url.rstrip("/")}?{judge}'
     return judge or None
 
@@ -644,11 +650,17 @@ def _compose_judge_spec(args):
     url = getattr(args, 'judge_api_url', '') or ''
     provider = getattr(args, 'judge_provider', 'openai-chat') or 'openai-chat'
     internal = {'openai-chat': 'openai', 'openai-pool': 'openai-pool'}.get(provider, provider)
-    if judge and not url and '/' not in judge:
-        raise SystemExit(
-            f'--judge-model {judge!r} needs --judge-api-url (or pass a full '
-            f'spec like openai/http://host:8000/v1?{judge})')
-    if url and judge and '/' not in judge:
+    if judge and not url:
+        if '/' not in judge and '?' not in judge:
+            raise SystemExit(
+                f'--judge-model {judge!r} needs --judge-api-url (or pass a '
+                f'full spec like openai/http://host:8000/v1?model)')
+        return judge   # full legacy spec, used as-is
+    if url and judge:
+        # url+model ALWAYS combines (same as the main model flags): a bare
+        # model name may itself contain slashes (/data/hf_models/...), which
+        # the old 'no slash' heuristic misread as a full spec and passed
+        # through un-prefixed
         judge = f'{internal}/{url.rstrip("/")}?{judge}'
     return judge or None