Estimate work you have never done before

I've been asked how long this will take: {{describe the work}}. What I know: {{stack, team size, what already exists}}. What I've never done before: {{the unfamiliar part}}.

Break it into tasks small enough that I could tell you within a day whether each one is going badly. For each: a best case, a realistic case, and what specifically would make it the bad case.

Then separate the estimate into two piles — work I understand and can size, and work that's genuinely unknown. For the unknowns, don't estimate them: propose a timeboxed spike that would turn each one into an estimate, and say what the spike must produce to count as finished.

List the assumptions the whole number rests on. Flag the one that, if wrong, doubles the estimate.

Finally, write the two-sentence version I can say out loud, including the range and the biggest unknown.