diff --git a/doc/adr/0022-preferred-identifier-algorithm.md b/doc/adr/0022-preferred-identifier-algorithm.md
new file mode 100644
index 0000000..e2573a3
--- /dev/null
+++ b/doc/adr/0022-preferred-identifier-algorithm.md
@@ -0,0 +1,39 @@
+# 23. Preferred Identifier Algorithm
+
+Date: 2024-11-28
+
+## Status
+
+Draft
+
+## Context
+
+When moving to a model where multiple documents have multiple identifiers, we need to understand how we choose which one to display for human identification purposes.
+
+## Decision
+
+Scores for individual components are multiplied together; IDs scoring 0 are never routinely considered preferred. (Scores for a component are 1 unless otherwise specified)
+
+#### Deprecation
+
+Identifier is deprecated: 0
+Document is deprecated: 0
+
+#### Schema
+
+Default: 0.5?
+NCN: 1
+BAILII: 0.8
+FCL SQID: 0.01
+
+#### Recentness
+
+Take the most recent (which measure of recentness? on publication? identifier?) if all other choices are even.
+
+### Other options
+
+We probably want to be able to limit to "only public documents" (for the PUI), "only human friendly identifiers", which can also add a 0 modifier to an identifier/document pair.
+
+## Consequences
+
+We have a modular system for determining the best identifier/document to return in multiple scenarios.