diff --git a/run.py b/run.py
index 819f440..abf9ac0 100644
--- a/run.py
+++ b/run.py
@@ -382,9 +382,20 @@ def step_build_all(args, examples_cache: dict, freq_cache: dict, conjugations: d
 
 
 def step_conjugations(args):
-    """Step 6 — extract conjugations (returns data; building handled by step_build_all)."""
+    """Step 6 — extract conjugations (returns data; building handled by step_build_all).
+
+    --skip-conjugations skips re-extraction from pealim.com but still loads
+    from cache so conj deck variants are built correctly.
+    """
+    conj_cache = DATA_DIR / "conjugations.json"
+
     if args.skip_conjugations:
-        logger.info("[6] Skipping conjugations (--skip-conjugations)")
+        if conj_cache.exists():
+            logger.info("[6] --skip-conjugations: loading from cache …")
+            with open(conj_cache) as f:
+                import json as _json
+                return _json.load(f)
+        logger.info("[6] --skip-conjugations: no cache found, skipping conj decks")
         return None
 
     verbs_file = Path(__file__).parent / "verbs_input.txt"
@@ -392,9 +403,7 @@ def step_conjugations(args):
         logger.info("[6] verbs_input.txt not found — skipping conjugation deck")
         return None
 
-    # Use cached conjugations.json if available (skip re-extraction)
-    conj_cache = DATA_DIR / "conjugations.json"
-    if conj_cache.exists() and not getattr(args, 'refresh_conjugations', False):
+    if conj_cache.exists():
         logger.info("[6] Using cached conjugations.json …")
         with open(conj_cache) as f:
             import json as _json