diff --git a/tests/_data/snapshots/README.md b/tests/_data/snapshots/README.md
index f9743244..6ef8f6eb 100644
--- a/tests/_data/snapshots/README.md
+++ b/tests/_data/snapshots/README.md
@@ -1,6 +1,28 @@
-# TEST FIXTURES
+# Test fixtures
 
-## RE-CREATION
+## Model fixtures
+
+Serialization and deserialization tests share BOM factories from
+[`tests/_data/models.py`](../models.py). To cover a model change:
+
+1. Add a descriptively named factory that starts with `get_bom_` and returns a
+   `Bom`. The function is discovered automatically; its name also becomes the
+   snapshot filename prefix.
+2. Populate every field relevant to the change so the JSON and XML output
+   tests exercise the complete model. Name a fixture ending in `_invalid` only
+   when serialization is expected to fail.
+3. Register a factory in `all_get_bom_funct_with_incomplete_deps` when its
+   dependency graph is intentionally incomplete. Otherwise, include the
+   complete dependency graph in the returned BOM.
+4. Re-create the snapshots, review the changed files, then run the tests again
+   without snapshot re-creation enabled.
+
+The output tests serialize each discovered model for the applicable schema
+versions, validate the generated JSON and XML, and compare it with the stored
+snapshots. The deserialization tests also use these fixtures for round-trip
+coverage.
+
+## Re-creation
 
 Some assets here can be (re-)created automatically, by setting the env var `CDX_TEST_RECREATE_SNAPSHOTS=1`.  
 It might also help to set `PYTHONHASHSEED=0`!  
@@ -10,4 +32,6 @@ As a shortcut just run:
 CDX_TEST_RECREATE_SNAPSHOTS=1 poetry run tox -e py
 ```
 
-The files will be written as is, which might not be human-readable. feel free to reformat the files manually.
+Only commit snapshots that belong to the model change. Snapshot tests compare
+the serialized text with the stored fixture using exact string equality, so do
+not reformat snapshots manually.