Retrospect on life#
A recording of:
Time costs
Action weights (on completed actions, when possible)
It’s not enough to retrospect on just the past week. You often don’t know whether an action you took was a good idea until years later. For example, many actions you take are justified based on the expected payoff over a course of years (e.g. going to college, getting faster feedback in some area of code). You won’t know until much later whether your investment paid off. If you need to retrospect on items like these, then you may have:
New calendar reminders or snoozed emails
Value#
More accurate weights on high-cost actions.
Cost#
Are you doing the weightiest actions? What is the scheduling function that selected the actions you took? In your case your function is the planning function you used at the start of the sprint, e.g. a sort of E[V]/E[E] on all the actions you were aware of at the start of your timebox (right now, based on a search for TODo).
Run a git grep for the TODo you removed in the last sprint (filter to e.g. one week ago). Every
completed TODo will generate a statistical error (or disturbance); see Errors and residuals -
Wikipedia applicable to your
estimation functions. If you don’t know yet whether a TODo was a good investment, then snooze an
email reminder for when you think you’ll know (or add a calendar reminder). All of these errors can
be vectors or tensors, not just scalars.
Generate Actions Error#
What new high weight actions did you discover this week? Run a git grep for the TODo you added in
the last sprint (filter to e.g. one week ago). If you had known about them at the start of the
sprint, would you have done them instead?
How did you discover them? For example:
Bugs, complaints, and errors
Drastically lower cost ways to complete a feature
Refine Actions Error#
Test Error#
Was the change in output as large as expected? Theoretically, measure E[T] - T.
Value Error#
Are you still doing high value work? Is the function from an output delta to a value delta stable? Theoretically, measure E[V] - V.
Cost Error#
Did it take you as long as predicted? Theoretically, measure E[E] - E.
Planning Error#
Did you spend or too little time planning? Adjust the heuristic you use to decide how long to plan together or separately (e.g. 5% of your time).
Count interruptive actions#
Measure (acknowledge) where time was spent so you’re aware of problems; don’t forget the fixes you do. Record your time costs as expected value (in time savings) on a new or existing task to prevent the interruption.
Did you properly handle interruptions, staying on focus? See Maintain focus.
Backprop Errors#
You should be able to backprop these errors into your estimation functions. If your estimation function is primarily based on previous experience (i.e. reference class forecasting) based on a population of training data effectively stored in your head, all you need to do is observe the difference. That is, update your priors.
Sometimes you’ll have also estimated based on first principles and need to think about what theory you got wrong.