diff --git a/codex/training/magic-alch-handoff.yaml b/codex/training/magic-alch-handoff.yaml
index 92157d2..c3da286 100644
--- a/codex/training/magic-alch-handoff.yaml
+++ b/codex/training/magic-alch-handoff.yaml
@@ -582,7 +582,11 @@ run_control:
   #   are independent by construction, and restocking did NOT shift the break.
   # Step 3 — break #2 at cast 1,100 (exactly 550 after #1 — monotonic,
   #   unaffected by the 9-cast restock cadence); break #3 at 1,650; restocks
-  #   continue at 1,107, 1,116, ... (every 9). The SII window ends at Magic 55:
+  #   continue at 1,107, 1,116, ... (every 9). On the alch tiers (LA and the
+  #   post-55 HLA tier) the restock cadence is the 901-cast batch: restock #1
+  #   at cast 901, restock #2 at cast 1,802 — monotonic, independent pins just
+  #   like the breaks (restocking never resets or shifts the break counter).
+  #   The SII window ends at Magic 55:
   #   with the default target 55 the session completes at the LA/SII -> HLA
   #   handoff; with a target above 55 the HLA tier (a8, 65 xp/cast, 901-cast
   #   batch model) executes until the target.