diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
new file mode 100644
index 00000000..bd79b4ba
--- /dev/null
+++ b/CONTRIBUTING.md
@@ -0,0 +1,96 @@
+# Contributing to htmlparser
+
+## Adding new elements
+
+When adding new elements to the parser, you must regenerate the element name hash tables in `src/nu/validator/htmlparser/impl/ElementName.java`.
+
+### Step 1: Add the new element constant
+
+Add a new `static final ElementName` constant for your element, following the existing pattern:
+
+```java
+public static final ElementName MYNEWELEMENT = new ElementName(
+    "mynewelement", "mynewelement",
+    // CPPONLY: NS_NewHTMLElement,
+    // CPPONLY: NS_NewSVGUnknownElement,
+    TreeBuilder.OTHER);
+```
+
+The flags (like `TreeBuilder.OTHER`, `SPECIAL`, `SCOPING`, etc.) depend on how the element should be handled by the tree builder.
+
+### Step 2: Uncomment the code generation sections
+
+Uncomment three sections in `ElementName.java`:
+
+1. **The imports** near the top (~lines 26-39):
+   - `java.io.*`
+   - `java.util.*`
+   - `java.util.regex.*`
+
+2. **`implements Comparable<ElementName>`** on the class declaration (~line 49)
+
+3. **The code generation block** marked with:
+   `"START CODE ONLY USED FOR GENERATING CODE uncomment and run to regenerate"`
+   That includes the `main()` method and helper functions (~lines 272-659)
+
+### Step 3: Add case to treeBuilderGroupToName() if needed
+
+If your element uses a new `TreeBuilder` group constant, add a case for it in the `treeBuilderGroupToName()` method within the code generation block.
+
+### Step 4: Compile and run
+
+Compile the project:
+
+```bash
+mvn compile
+```
+
+Run the `ElementName` class with paths to the Gecko tag-list files:
+
+```bash
+java -cp target/classes nu.validator.htmlparser.impl.ElementName \
+    /path/to/nsHTMLTagList.h \
+    /path/to/SVGTagList.h
+```
+
+**For Java-only builds** (not Gecko), you can use empty dummy files:
+
+```bash
+mkdir -p /tmp/tagfiles
+touch /tmp/tagfiles/nsHTMLTagList.h /tmp/tagfiles/SVGTagList.h
+java -cp target/classes nu.validator.htmlparser.impl.ElementName \
+    /tmp/tagfiles/nsHTMLTagList.h \
+    /tmp/tagfiles/SVGTagList.h
+```
+
+> [!NOTE]
+> Using empty files means the `CPPONLY` comments will all show `NS_NewHTMLUnknownElement`. For Gecko builds, use the actual files from moz-central:
+> - `parser/htmlparser/nsHTMLTagList.h`
+> - `dom/svg/SVGTagList.h`
+
+### Step 5: Update the generated arrays
+
+The program outputs:
+1. All element constant definitions (with updated `CPPONLY` comments if using real Gecko tag files)
+2. The `ELEMENT_NAMES` array in level-order binary search tree order
+3. The `ELEMENT_HASHES` array with corresponding hash values
+
+Replace the existing `ELEMENT_NAMES` and `ELEMENT_HASHES` arrays in the file with the generated output. The arrays must stay in sync—element at position N in `ELEMENT_NAMES` must have its hash at position N in `ELEMENT_HASHES`.
+
+### Step 6: Re-comment the code generation sections
+
+After regeneration, comment out the sections you uncommented in Step 2 to restore the file to its normal state.
+
+### Step 7: Run tests
+
+Verify your changes work correctly:
+
+```bash
+mvn test
+```
+
+### Technical Details
+
+The hash function (`bufToHash`) creates a unique integer for each element name using the element's length and specific character positions. The arrays are organized as a level-order binary search tree for O(log n) lookup performance.
+
+If you encounter a hash collision (two elements with the same hash), the regeneration will report an error. That would require modifying the hash function, which has not been necessary historically.
diff --git a/test-src/test/resources/html5lib-tests b/test-src/test/resources/html5lib-tests
index 6ddcf58b..8f43b7ec 160000
--- a/test-src/test/resources/html5lib-tests
+++ b/test-src/test/resources/html5lib-tests
@@ -1 +1 @@
-Subproject commit 6ddcf58bea5a01e616911050c173622f84297211
+Subproject commit 8f43b7ec8c9d02179f5f38e0ea08cb5000fb9c9e