diff --git a/doc/bucket.texi b/doc/bucket.texi
index 0b4a810..9a581bb 100644
--- a/doc/bucket.texi
+++ b/doc/bucket.texi
@@ -28,7 +28,89 @@
 @node Bucket Diff
 @section Bucket Diff
 @cindex Bucket diff
-@helpwanted
+Changes to the bucket are represented by an array with certain conventions:
+
+@enumerate
+@item A change to some index@tie{}@math{k} is represented by the same
+        index@tie{}@math{k} in the diff.
+@item A value of @code{undefined} indicates that the respective index
+        has not changed.
+      Holes in the array (indexes not assigned any value) are treated in
+        the same manner as @code{undefined}.
+@item A @code{null} in the last index of the vector marks a
+        truncation@tie{}point;
+          it is used to delete one or more indexes.
+      The vector will be truncated at that point.
+      Any preceding @code{null} values are treated as if they were
+        @code{undefined}.@footnote{
+          The reason for this seemingly redundant (and inconvenient) choice
+            is that JSON encodes @code{undefined} values as @code{null}.
+          Consequently, when serializing a diff, @code{undefined}s are lost.
+          To address this,
+            any @code{null} that is not in the tail position is treated as
+            @code{undefined}.
+          We cannot truncate at the first null,
+            because @samp{[null,null,null]} may actually represent
+            @samp{[undefined,undefined,null]}.}
+@end enumerate
+
+Diffs are only tracked at the vector (array of scalars) level@mdash{
+  }if there is a change to a nested structure assigned to an index,
+    that index of the outer vector will be tracked as modified.
+It will, however, recursively compare nested changes to determine if a
+  modification @emph{has taken place}.@footnote{
+    See @srcrefjs{bucket,StagingBucket} method @code{#_parseChanges}.}
+Examples appear in @ref{f:diff-ex}.
+
+@float Figure, f:diff-ex
+@multitable @columnfractions 0.30 0.30 0.40
+@headitem Original @tab Diff @tab Interpretation
+@item @samp{["foo", "bar"]}
+  @tab @samp{["baz", "quux"]}
+  @tab Index@tie{}0 changed to @samp{baz}.
+       Index@tie{}1 changed to @samp{quux}.
+
+@item @samp{["foo", "bar"]}
+  @tab @samp{[undefined, "quux"]}
+  @tab Index@tie{}0 did not change.
+       Index@tie{}1 changed to @samp{quux}.
+
+@item @samp{["foo", "bar"]}
+  @tab @samp{[, "quux"]}
+  @tab Index@tie{}0 did not change.
+       Index@tie{}1 changed to @samp{quux}.
+
+@item @samp{["foo", "bar"]}
+  @tab @samp{["baz", null]}
+  @tab Index@tie{}0 changed to @samp{baz}.
+       Index@tie{}1 was removed.
+
+@item @samp{["foo", "bar", "baz"]}
+  @tab @samp{[undefined, null]}
+  @tab Index@tie{}0 was not changed.
+       Index@tie{}1 was removed.
+       Index@tie{}2 was removed.
+
+@item @samp{["foo", "bar", "baz"]}
+  @tab @samp{[null, undefined, null]}
+  @tab Index@tie{}0 was not changed.
+       Index@tie{}1 was not changed.
+       Index@tie{}2 was removed.
+
+@item @samp{["foo", "bar", "baz"]}
+  @tab @samp{[null, null, null]}
+  @tab Index@tie{}0 was not changed.
+       Index@tie{}1 was not changed.
+       Index@tie{}2 was removed.
+@end multitable
+@caption{Bucket diff examples.}
+@end float
+
+Diffs are generated by @srcrefjs{bucket,StagingBucket}.
+@code{null} truncation is understood both by @code{StagingBucket}
+  and by @srcrefjs{bucket,QuoteDataBucket}.
+A diff is applied to the underlying bucket by invoking
+  @code{StagingBucket#commit}.
 
 
 @node Calculated Values