Curious how you handle feed evolution over time. When an RSS source changes structure (fields added/removed, summaries truncated, etc.), do you normalize to a fixed schema or store the raw payload alongside a best-effort normalized version? Longitudinal datasets tend to get tricky there.