Desk Research

Desk research answers a question with existing data, rather than new data (surveys, interviews, experiments) or a shipped feature. Deliver a defensible answer and a checkable trail — not production code or an unsupported claim.

Philosophy

1. Keep the exploratory trail. Unlike production work, the versioned artifact is the research process. Preserve tried approaches and dead ends as evidence; a mid-investigation wip commit is acceptable. For production history, eg-change-documentation-rigor still says to keep exploration out.

2. Search free parameters explicitly; don't assert a value. Any technique with a tunable parameter (cluster count, threshold, window size) needs the search shown, not a hardcoded choice presented as obviously correct. "I tried k=3 through k=12 and picked the elbow at k=6" is checkable; "grouped into 6 clusters" is not.