What Has to Survive for Work to Continue

A relay baton with a tag taped to it: now, Postgres; next, yours. No replay of the race, just the baton and the note. Where things stand, not how they got here. A transcript reads: use Postgres, no, SQLite, fine, Postgres after all. Which one still stands? Continuing work needs current truth and an owner for the next step.

Three questions: what happened, which is the record and evidence for looking back; where things stand, current truth for going on; and who owns the next step, which is the handoff. One file can serve all three, but they are different jobs.

A transcript shows a choice decided, reversed and decided again, and doesn't say which one governs. A current-state card does: database Postgres, decided at 14:15, next step yours. Length isn't the defect; ambiguity about what is current is.

An instruction, such as approval is required before publishing, is true for every piece of work. State, such as this change is waiting on approval, is true this afternoon and false tomorrow. Mixing them buries rules in status notes, or writes live status into the rules.

This repository's handoff document, current.md, lists thirty-five key files touched, and thirty-four of them no longer exist at the commit the article cites. They were true when written. Make a handoff's position checkable: pin the version, date it, and name the owner.

The next person needs what's done, what's true now, the next step, the constraints still active, and who owns it; then they can act. A longer record isn't the fix. Separate how the work got here from where it currently stands.