[pengo]

The most reliable estimate is for work you've already done and documented. Fifteen years ago I wrote my own timesheet system where entries can be directly linked to code changes. When I have to estimate I can interrogate this to get comparable (and sometimes identical) tasks from the past.

I do have to create proposals for new clients moving into my platform but, again, I have data from previous onboarding exercises. And all estimates have the clear and repeated caveat that they're based on the stated requirements; if the requirements change, the cost and dates change too.

When I was leading a development team through the estimation process for large and complex projects, I used the Wideband Delphi process, which is essentially a process of iterative decomposition; I recommend it.