diff --git a/docs/tutorials/tutorial2.md b/docs/tutorials/tutorial2.md
index af565da60..c2dd4fb88 100644
--- a/docs/tutorials/tutorial2.md
+++ b/docs/tutorials/tutorial2.md
@@ -246,6 +246,18 @@ runBeamSqliteDebug putStrLn conn $ runInsert $
   insertValues [ james, betty, sam ]
 ```
 
+!!! tip "Tip"
+    The `let` bindings for `james`, `betty`, and `sam` above are reused at two
+    different types: as concrete rows (`UserT Identity`) inside
+    `insertValues`, and as expression-level values (here as `pk james`, of
+    type `PrimaryKey UserT (QExpr Sqlite s)`) inside `insertExpressions`.
+    For the same binding to be usable at both types, GHC must infer a
+    polymorphic type for it. Inside GHCi this happens by default, but in a
+    compiled module you may need to enable `NoMonomorphismRestriction` (or
+    give each binding an explicit polymorphic signature). Otherwise the
+    monomorphism restriction will pin `james` to whichever type it is first
+    used at, and the second use will fail to typecheck.
+
 Now that we have some `User` objects, we can create associated addresses. Notice
 that above, we used `insertValues` to insert concrete `User` rows. This worked
 because we could determine every field of `User` before insertion. `Address`es
diff --git a/docs/tutorials/tutorial3.md b/docs/tutorials/tutorial3.md
index aa1acb95c..64a1cff4f 100644
--- a/docs/tutorials/tutorial3.md
+++ b/docs/tutorials/tutorial3.md
@@ -483,7 +483,25 @@ Some RDBMSs, like Postgres, given such a query will be unable to utilize availab
 to perform join operations - this translates to *extremely* poor perfomance for even moderately
 sized data.
 
+!!! warning "Nullable columns and `maybe_` on a left-joined table"
+    There is also a *correctness* gotcha here whenever the left-joined table
+    contains a schema-level nullable column. `maybe_` (and `isJust_`) on a
+    nullable table is implemented as "every column of the row is non-`NULL`",
+    so a real, matched row whose nullable column happens to be `NULL` is
+    treated as `Nothing`. In our schema `OrderT` has the nullable
+    `_orderShippingInfo` column, so for any of James's unshipped orders
+    `maybe_ (val_ False) ... order` evaluates to `False`, and the chained
+    `lineItem` and `product` left joins return no rows for him. The query
+    above will compute `Just 0` for James instead of `Just 27000`.
+
+    The next variant uses `leftJoin_'` and avoids `maybe_` over a nullable
+    table entirely, so it is both faster *and* correct in this case. As a
+    rule of thumb, when a left-joined table contains nullable columns,
+    prefer `leftJoin_'` with `==?.` (and other `SqlBool` operators) over
+    `leftJoin_` with `maybe_`.
+
 Luckily, Beam also provides an alternate way to phrase things that directly maps to SQL semantics
+(and avoids the gotcha above)
 
 !beam-query
 ```haskell
@@ -589,7 +607,7 @@ shippingInformationByUser <-
     do user <- all_ (shoppingCartDb ^. shoppingCartUsers)
 
        (userEmail, unshippedCount) <-
-         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 countAll_)) $
+         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 (count_ (_orderId order)))) $
          do user  <- all_ (shoppingCartDb ^. shoppingCartUsers)
             order <- leftJoin_ (all_ (shoppingCartDb ^. shoppingCartOrders))
                                (\order -> _orderForUser order `references_` user &&. isNothing_ (_orderShippingInfo order))
@@ -598,7 +616,7 @@ shippingInformationByUser <-
        guard_ (userEmail `references_` user)
 
        (userEmail, shippedCount) <-
-         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 countAll_)) $
+         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 (count_ (_orderId order)))) $
          do user  <- all_ (shoppingCartDb ^. shoppingCartUsers)
             order <- leftJoin_ (all_ (shoppingCartDb ^. shoppingCartOrders))
                                (\order -> _orderForUser order `references_` user &&. isJust_ (_orderShippingInfo order))
@@ -629,7 +647,7 @@ shippingInformationByUser <-
 
        (userEmail, unshippedCount) <-
          subselect_ $
-         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 countAll_)) $
+         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 (count_ (_orderId order)))) $
          do user  <- all_ (shoppingCartDb ^. shoppingCartUsers)
             order <- leftJoin_ (all_ (shoppingCartDb ^. shoppingCartOrders))
                                (\order -> _orderForUser order `references_` user &&. isNothing_ (_orderShippingInfo order))
@@ -639,7 +657,7 @@ shippingInformationByUser <-
 
        (userEmail, shippedCount) <-
          subselect_ $
-         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 countAll_)) $
+         aggregate_ (\(userEmail, order) -> (group_ userEmail, as_ @Int32 (count_ (_orderId order)))) $
          do user  <- all_ (shoppingCartDb ^. shoppingCartUsers)
             order <- leftJoin_ (all_ (shoppingCartDb ^. shoppingCartOrders))
                                (\order -> _orderForUser order `references_` user &&. isJust_ (_orderShippingInfo order))