diff --git a/src/lib_crawler/layer_1.ml b/src/lib_crawler/layer_1.ml
index 524c59462d8441010a494c1c4ae545de19e7f53a..518f2a1dd736867f39504fcd0767de483168d829 100644
--- a/src/lib_crawler/layer_1.ml
+++ b/src/lib_crawler/layer_1.ml
@@ -215,7 +215,10 @@ let get_predecessor =
       (chain : Tezos_shell_services.Chain_services.chain)
       ancestor ->
     let open Lwt_result_syntax in
+    (* Don't read more than the hard limit in one RPC call. *)
     let max_read = min max_read hard_max_read in
+    (* But read at least two because the RPC also returns the head. *)
+    let max_read = max max_read 2 in
     match HM.find_opt cache ancestor with
     | Some pred -> return_some pred
     | None -> (
@@ -294,9 +297,10 @@ let get_tezos_reorg_for_new_head l1_state
     if old_head_level = new_head_level then
       return (old_head, new_head, Reorg.no_reorg)
     else if old_head_level < new_head_level then
+      let max_read = distance + 1 (* reading includes the head *) in
       let+ new_head, new_chain =
         nth_predecessor
-          ~get_predecessor:(get_predecessor ~max_read:distance l1_state)
+          ~get_predecessor:(get_predecessor ~max_read l1_state)
           distance
           new_head
       in
@@ -321,9 +325,10 @@ let get_tezos_reorg_for_new_head l1_state ?get_old_predecessor old_head new_head
       if new_head_level < l then return Reorg.no_reorg
       else
         let distance = Int32.sub new_head_level l |> Int32.to_int in
+        let max_read = distance + 1 (* reading includes the head *) in
         let* _block_at_l, new_chain =
           nth_predecessor
-            ~get_predecessor:(get_predecessor ~max_read:distance l1_state)
+            ~get_predecessor:(get_predecessor ~max_read l1_state)
             distance
             new_head
         in
diff --git a/src/lib_crawler/layer_1.mli b/src/lib_crawler/layer_1.mli
index f5573d58d6e3990b12ae9d2a970f8bd72a0b4723..794c566b596720301065330a1b8bd73907204046 100644
--- a/src/lib_crawler/layer_1.mli
+++ b/src/lib_crawler/layer_1.mli
@@ -67,8 +67,8 @@ val wait_first : t -> (Block_hash.t * Block_header.t) Lwt.t
 
 (** [get_predecessor_opt ?max_read state head] returns the predecessor of block
     [head], when [head] is not the genesis block. [max_read] (by default [8]) is
-    used to cache a number of predecessors to avoid some RPC calls when one need
-    to access multiple predecessors. *)
+    used to cache a number of predecessors (including [head]) to avoid some RPC
+    calls when one need to access multiple predecessors. *)
 val get_predecessor_opt :
   ?max_read:int ->
   t ->