No evidence that TDD differs from a TLD

That only works if there are many types.

I work with molecules, expressed as molecular graphs. The algorithm development I do is mostly graph manipulation, with a small number of types. For example, given N molecules, find the largest subgraph (in terms of the number of edges) which are in at least k of them.

Or, think about computing an FFT, or clustering a data set.

Types add little to how to think about these problems, so TFD provides little structure to this sort of algorithm development process.