°FAI: Which Number Is Lying? A Morning of Making the Data Honest

Our weekly press report showed setup running at 18.2 minutes per screen, over our target. That target isn’t arbitrary. It came out of the work I walked through in °FAI: Screen printing setup times, where it drops by a minute a week as long as the crew keeps holding it. So when the report says we’re over, its worrisome. On its face, 18.2 reads as a crew problem. But before telling anyone the team slipped, we went and looked.

The 18.2 was a fiction, and one job created it. A single sample-and-approval timer had been left running to 250 minutes on a two-color back print, 138 minutes per screen. Averaged across twenty-nine jobs, that one entry dragged the whole figure above target. Pull it, and the crew’s real number was 13.7 minutes per screen, headed in the right direction. Nobody slipped. The metric slipped.

I’ve seen this one before, wearing different clothes. The first version was a sleeping tablet: “The tablet went to sleep overnight, its software timers stopped with it, and the check-in popup and auto-pause slept right through, while the clock kept counting.” We fixed that. This time nothing went to sleep. Somebody just got pulled to another press and never hit Stop. Same eight-thousand-minute damage, new cause. The failure moves to whatever form you haven’t hardened yet.

The delay figure told the opposite story. The report showed 22% of logged hours lost to delay, and that number was accurate, we’d already built the delay math to discard timers left running past a full shift, so it reflected real waiting: client approvals, ink changes, screens not ready. Two numbers on one report, opposite trustworthiness, and the difference had nothing to do with the data.

Two dark-room timers had been left running seventeen and nineteen hours, overnight, into the next day, pushing its average work time to 208 minutes when a normal entry runs closer to one hour. The form had no guardrail; a forgotten timer simply kept counting. Half of the common mistakes in screen making come down to something nobody wrote down.

So we changed two things. The report now drops any setup over 60 minutes per screen as impossible and lists every excluded entry by order number, so nothing disappears quietly. And the dark room form now stops any timer that runs past four hours, the same protection the press form already carried. A forgotten timer can no longer manufacture a phantom day of work.

None of this is glamorous, and that’s the point. A number is only as trustworthy as the failure modes you’ve closed underneath it. The setup figure was lying because we hadn’t closed one; the delay figure was honest because we had; the dark room was exposed because we’d never gotten to it. This morning we closed a few more, and the averages finally say what they mean. The progression of this over time is necessary because you need to see the numbers and work with them to figure out how to build the numbers. That’s why using AI to do this is so effective.

If you’re building your own logs °FAI – Build your own time tracker put the ceiling and the exclusion list in on day one, if you can figure out what those caps are. It’s twenty minutes while you’re building and a month of bad decisions if you skip it. That matters more the faster you’re growing, because growth means more decisions made on numbers nobody has time to check. And the rule from °FAI: AI Pipeline for a Screen Print Factory still holds: “None of it replaces the printer’s judgment. It carries the judgment while you’re watching it.” A report that hides its own garbage isn’t carrying your judgment.