The classroom still had that fresh-paint smell underneath the dust we'd tracked in, twenty desks waiting for a September that felt further away than it had three weeks earlier. I'd brought the training manual up from storage myself, the one I'd drafted and redrafted for a decade, and set it open on the instructor's table next to the routing sheet Inez had found.
"Walk me through it," Inez said. "The way you taught it, not the way it reads now."
I'd taught the exception classification a hundred times, maybe more, always with the same example on the board: a late report, an access log with a gap that had an obvious innocent cause, a conversation and a refresher course instead of a formal inquiry. The manual was clear about the mechanism. Section four, the part I'd written myself in 1989 and barely touched since. An exception classification was temporary. It closed a concern provisionally, pending independent follow-up within ninety days, confirmation from someone who hadn't been the one to raise the original flag.
"Ninety days," I said, reading it aloud like it might sound different spoken than it had in my own head for fifteen years. "That's what I wrote. That's what I taught."
"And what happened to the ninety days?"
I didn't answer right away. I turned the routing sheet over, looking for a second signature, a follow-up stamp, anything that would show someone had come back to Della's 1991 concern within the window the manual required. There was nothing. Just the original flag, the closure code, and a date that sat alone on the page like it had never expected company.
"Somewhere along the way the follow-up requirement stopped being enforced," I said. "I don't know exactly when. I know I never checked."
"You never checked whether your own follow-up rule was being followed."
It wasn't an accusation, not from Inez, not in her flat record-comparison voice, but it landed like one anyway. I thought about every instructor review I'd signed over the years, every annual assessment where I'd written some version of the same sentence — exception process functioning as designed, recommend continuation without modification — based on nothing but the absence of complaints reaching my desk. I had never once pulled a sample of closed exceptions and asked whether the ninety days had actually happened.
"I praised a process I never tested," I said. "Not once. Not in fifteen years."
Inez didn't soften it, and I didn't want her to. "That's not the same as approving Della's closure specifically."
"No. But it's the reason a closure like that could sit unexamined for thirteen years without anyone thinking twice. I built a door and told everyone it locked itself. I never once tried the handle."
I sat with that for a while, the manual open to the page I'd written in my own handwriting before it was typed and distributed and taught to two decades of examiners who trusted it because I'd trusted it, or at least acted like I had. The difference between a system with a real safeguard and a system with a safeguard nobody checked was, on paper, nothing at all. In practice it was the entire thirteen years between 1991 and the handcuffs.
"I want this in the review," I said. "Not as background. As a finding. The follow-up requirement existed and wasn't enforced, and I'm the person who should have been checking."
"That'll draw a line straight back to you."
"It already runs straight back to me. I'm just the one saying so before someone else has to."
I closed the manual and set it on top of the routing sheet, the two documents finally in the same stack where they should have lived for thirteen years, and thought about how many other closures carried that same code, how many other ninety-day windows had simply never happened, and how long I had let myself believe that silence from the field meant the system was working rather than meaning nobody had ever gone looking.