diff --git a/CHANGELOG.md b/CHANGELOG.md
index e53a321..38c9629 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,9 @@
 # Changelog
 
+### `0.2.4`
+
+- Fix minor incompatibility with Redshift; contributed by [@steelcd](https://github.com/steelcd).
+
 ### `0.2.3`
 
 - Added Postgres support in integration tests + fixed bugs that prevented Postgres from working.
diff --git a/README.md b/README.md
index c4a0e06..d143ada 100644
--- a/README.md
+++ b/README.md
@@ -277,6 +277,7 @@ There are a few reasons why this method is discouraged over the `chol` method:
 - 🐌 It tends to be much slower in OLAP systems, and struggles to efficiently calculate large number of columns.
 - 📊 It does not calculate standard errors.
 - 😕 For ridge regression, coefficients are not accurate; they tend to be off by a magnitude of ~0.01%.
+- ⚠️ It does not work in all databases because it relies on `COVAR_POP`.
 
 So when should you use `fwl`? The main use case is in OLTP systems (e.g. Postgres) for unregularized coefficient estimation. Long story short, the `chol` method relies on subquery optimization to be more performant than `fwl`; however, OLTP systems do not benefit at all from subquery optimization. This means that `fwl` is slightly more performant in this context.
 
diff --git a/dbt_project.yml b/dbt_project.yml
index 59a675e..6eb1606 100644
--- a/dbt_project.yml
+++ b/dbt_project.yml
@@ -1,5 +1,5 @@
 name: "dbt_linreg"
-version: "0.2.3"
+version: "0.2.4"
 
 # 1.2 is required because of modules.itertools.
 require-dbt-version: [">=1.2.0", "<2.0.0"]