Add documentation and tests for remote training variant flags

Documentation updates for issue #27:

1. test_documentation.py:
   - Added comprehensive test for variant flag documentation
   - Verifies all long-form and short-form flags are documented
   - Tests that usage information is displayed

2. README.md:
   - Added troubleshooting section for remote_train.sh
   - Includes verification commands for variant flags
   - Documents connection and screen session debugging

All documentation tested and verified.

Note: CLAUDE.md also updated locally with detailed implementation
guidance for remote_train.sh (not committed - in .gitignore).

Related: #27

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

Author: jeremymanning

README.md

@@ -333,6 +333,50 @@ screen -r llm_training
 exit
 ```
 
+#### Troubleshooting remote_train.sh
+
+**Variant flags not working?**
+
+If variant flags (`-co`, `-fo`, `-pos`) aren't being passed correctly, you can verify:
+
+```bash
+# Check the training script on the server
+ssh username@server 'cat /tmp/llm_train.sh | grep VARIANT_ARG'
+# Should show: VARIANT_ARG='-co' (or '-fo', '-pos')
+# NOT: VARIANT_ARG=''
+
+# Check the training log
+ssh username@server 'cat ~/llm-stylometry/logs/training_*.log | grep -i variant'
+# Should show: "Training variant: content" (or "function", "pos")
+# NOT: "Training baseline models"
+
+# Check debug output at start of training script
+ssh username@server 'cat /tmp/llm_train.sh | head -5'
+# Should show:
+# RESUME_MODE='false' (or 'true')
+# VARIANT_ARG='-co' (or '-fo', '-pos', or '' for baseline)
+```
+
+**Connection issues?**
+
+```bash
+# Test SSH connection manually
+ssh username@server echo "Connection works"
+
+# If you get permission denied, check your SSH keys or use password authentication
+ssh -o PreferredAuthentications=password username@server
+```
+
+**Screen session not found?**
+
+```bash
+# List all screen sessions
+ssh username@server 'screen -ls'
+
+# If training crashed, check the log
+ssh username@server 'tail -50 ~/llm-stylometry/logs/training_*.log'
+```
+
 #### Downloading results after training completes
 
 Once training is complete, use `sync_models.sh` **from your local machine** to download the trained models and results:

tests/test_documentation.py

@@ -215,6 +215,31 @@ def test_python_api_variant_parameter():
             pass
 
 
+def test_remote_train_variant_flags_documented():
+    """Verify all variant flags are documented in remote_train.sh usage."""
+    result = subprocess.run(
+        ['./remote_train.sh'],
+        input='\n\n',  # Empty inputs to get past prompts
+        capture_output=True,
+        text=True,
+        timeout=10
+    )
+
+    # Check long-form flags
+    assert '--content-only' in result.stdout, "Missing --content-only in remote_train.sh"
+    assert '--function-only' in result.stdout, "Missing --function-only in remote_train.sh"
+    assert '--part-of-speech' in result.stdout, "Missing --part-of-speech in remote_train.sh"
+
+    # Check short-form flags
+    assert '-co' in result.stdout, "Missing -co short flag in remote_train.sh"
+    assert '-fo' in result.stdout, "Missing -fo short flag in remote_train.sh"
+    assert '-pos' in result.stdout, "Missing -pos short flag in remote_train.sh"
+
+    # Check that usage information is shown
+    assert 'Usage:' in result.stdout or 'Options:' in result.stdout, \
+        "remote_train.sh should display usage information"
+
+
 if __name__ == '__main__':
     # Run tests
     pytest.main([__file__, '-v'])