This approach of shredding data is risky if the schema is absent when we start. This is the same problem as ducktyping where we endup with variables with questionable types.
On a different note, all references are by the same author so it would have been nice to see a different point of view