diff --git a/docs/REPLICATION.md b/docs/REPLICATION.md
new file mode 100644
index 0000000..8458c3a
--- /dev/null
+++ b/docs/REPLICATION.md
@@ -0,0 +1,18 @@
+The key thing to getting replication right is having the ability to negotiate
+consistency. Distributed systems are, as has been noted, hard to make, and even
+harder to make correct. For a website, this is generally handled by having a
+backing database that has replication functionality, and then having multiple
+flat copies of a site accessing it; normally, with a well-defined schema,
+there's not a problem if updating the site takes a minute longer on a few nodes,
+because the backing data structure (the DB) can handle (or throw away) the odd
+requests.
+
+If you're rolling your own data structure, though, it's more complicated. In
+this case you're probably best served by not doing replication, but sharding.
+It's far easier to spread work out on some known identifiers than working out
+how to synchronize various tables, but you don't get the performance.
+
+If you can manage it, a forum is actually a great place to use a hybridized
+system. Live data can be sharded, and old data can be replicated - since it
+doesn't change much, it doesn't matter if it lives in several places.
+