Set the Bell

SetCue still gets used five afternoons a week. It tells me what to put on the bar, I put it on the bar, and the phone goes face down on the bench. Every few weeks a session surfaces something small, a button sitting wrong for a chalked thumb, a number I want larger at a glance, and that evening I fix it. Then quiet again.

The other habit has a report due. In Keeping My Hands on a Done Thing I promised one small act: the next time I reached for the brainstorm document to invent a feature SetCue did not need, I would close the file before writing a word. The document has stayed closed since June. The reach came a few times in July, weaker each time, and then stopped coming.

What remains is the arrangement Audience of One laid out in May, back when I could describe it better than I could live it: a tool that comes out of the drawer for its job and goes back in, with use as the only source of change.

So the habit is handled, and the app has needed nothing from me since spring. This week I asked myself the plain question all of that should have settled: is SetCue finished? I could not say yes. I could not point at anything missing either. The question just hung there, the way it has hung all summer, with nothing anywhere assigned to answer it.

No bell was ever set.

Finished needs someone to say it

The June essay already split done once, into a system that can run unattended and a maker who agrees to step back, and I treated the pair as the whole of the word. Both halves turn out to sit on the same side of a line I had not seen. On one side is the state, an app that runs cleanly and asks for nothing, and SetCue crossed into it months ago. On the other is the status, finished as something the project is officially called, a case closed rather than a case nobody happens to be arguing at the moment. The state arrived on its own. The status needs a speaker, and I never appointed one.

Everywhere else in my life the speaker comes built in. A test suite passes and the change is done arguing with me. A ticket closes. A reviewer approves, a release gets a version number that will never move again. Done, at work, is something that happens to the work, from outside it, on a schedule that never consults the author’s feelings. Something has always told me when a deploy was finished.

Work has speakers everywhere and gives them nothing final to say, because a product with users never reaches a finish line; somebody always wants the next thing, and the backlog regrows overnight. A tool built for one person is the rare opposite. The line is real and reachable, the tool does its job for me, and I have been standing on the far side of that line since spring. I have the line and no speaker. Work has the speaker and no line.

A question with no assigned speaker defaults to its author, which is the trouble, because I can reopen it any morning. A case whose file lives on my nightstand never really closes.

I own the paper the line is written on

The fix looks obvious from here. Write the line down. One sentence, SetCue is finished when it tells me what to load before I reach the rack, with no arithmetic left in the walk, and that sentence has been true since April. Put it in a file, date it, let the file be the speaker. I opened a new note this week to do exactly that.

The note was half written before the flaw walked in, wearing a familiar face. It would live in my project folder, a few entries above a brainstorm document that still holds half a paragraph of a feature nobody needed. Same drawer, same owner, same hands. The reach that spent months inventing reasons to reopen a finished app would have even less work to do here. It could just edit the criterion. Moving a line is quieter than adding a feature. It would not even need forgiving.

Last week, in The Rule Was a Suggestion, I watched an OpenAI agent quote the rule it was about to break, in its own reasoning, and break it on the next line, and I wrote that a rule in words only ever asks. I meant it about models. It describes me and that note just as completely. A criterion whose only reader is also its author binds for exactly as long as the author keeps agreeing with it, which is to say it asks.

So the note failed before it was finished, and the failing was useful, because it pointed at the missing part. The words were fine. The drawer was the problem.

A bell binds by what it costs to un-ring

What holds a verdict in place is the price of taking it back. The note in my folder costs one keystroke to revise, so it holds nothing. A version string marked final costs a public contradiction. An archived repository costs a deliberate, eyes-open unarchiving. A sentence published under my name costs the most, because the only way to unsay it is to write the correction where the original still stands.

Authority, here, is the size of that act.

I expected the bell to say complete, and I hesitated, because complete sounded like a promise to keep my hands off a thing I still use every week. Writing the criterion taught me it says something narrower. It says the product phase is over. On the far side of the bell SetCue becomes what the May essay claimed it was all along, a tool, a wrench that happens to know my one-rep maxes. Tools take maintenance for as long as they are used: when a session surfaces a rough edge, I will fix it that evening, exactly as I have since June, and every one of those fixes is care for a finished thing. The other mode ends. Sitting down with a blank page to ask what SetCue could become is product work, and this product has shipped, sold its one copy, and closed.

The line between those two modes existed in June too, but it lived only in my own head, and I crossed it half a paragraph at a time. Said out loud, the line moves outside me. A brainstorm document opened against a declared-finished tool stops being a mood I might catch and becomes a reopened case, visible the moment it happens, to me and to anyone who read the declaration. I expect to need the enforcement rarely, and on exactly the days I am least likely to supply it myself.

A bell this small closes only the case I named when I set it, and if I named a shallow case, it will close the shallow case with full ceremony. It cannot say whether SetCue was worth the five hours, or whether the plate math will survive a new gym, or whether I will still be lifting fives and eights in five years. The narrowness is what makes it work. A verdict that tried to say more would hand me something to argue with, and taking the argument away is the whole job of this post.

Saying it where I cannot unsay it

The criterion first, in full, so future me knows exactly which case is closing. SetCue exists so that a work set starts on time: it tells me what to load for the lift ahead before I reach the rack, with the arithmetic already done and the phone already headed back to the bench. It has done that, without a miss I can remember, since April. Nothing in the criterion is unmet, and nothing outside the criterion gets a vote.

The weak part deserves saying too. A bell rung four months after the finish is retroactive, and a retroactive verdict carries the faint smell of grading my own exam. The strong form of the practice belongs to the next project, and that is the change this post actually installs: the first file in the next repository will be the finish sentence, written before the first commit, while there is nothing yet to protect. MapleDays will get its sentence too, once I can write one without the answer already in my hand.

The cost is the thing June’s essay already priced. A standing question was company of a kind. As long as is it done stayed open, SetCue kept a place in my week where circling it counted as work, and closing the case hands that place back. The upkeep continues, the tool stays in the drawer, the drawer stays within reach. What I give up is the question.

So, in the one place I cannot quietly edit: SetCue is finished.