Outcome: the user has looked at everything long-horizon they carry, said which one or two could be great, decided what dies or parks, and written that down as a dated review entry. The name comes from the essay's closing point: awards and acceptances are samples; the user owns the distribution, and this is where the distribution is tended.
Scope: long-horizon efforts only, per the repository's AGENTS.md. Near-term deliverables are out of the review even if they take most of the week.
Inputs
- The list of long-horizon efforts in flight or parked, from the user.
- The taste journal, read with entries older than three months first. Predictions without a matching hindsight entry are the first things to raise.
The review, one question at a time
- Which one or two of these could be great? The essay expects one or two a year; the rest are practice. Say which is which and why.
- For each effort: is the core still working, does the best-case conclusion still matter, and has anything more important appeared since the last review? Kill or park what fails, and say what is salvaged.
- For each surviving effort: how many months ahead of the next person is it now, compared to the last guess in the journal? Has the area's importance moved?
- Anything rejected or null since last time: what did the reviewers reject, the premise or the argument? What in the distribution would you change, and what is just the sample?
- Is the user's main effort where their skill and the area's current importance meet, or in a corner the field has left (episode E10 in the taste references)?
No scores. Reasons in the user's words.
Output
Append one dated review entry to the journal, written as the user would write it: the great candidates, what was killed or parked and why, the months-ahead guesses, and the one change to the distribution for next month. Then one line naming the first action for the coming week on the top candidate.