Tell me about a time you managed someone whose performance wasn't meeting the bar.

I ran a four-person team funded to work full time on an open-source data-quality project. About four months in, one engineer was not meeting the bar for his level. His batch work was solid, but every streaming task either stalled or came back with correctness problems — one fix he merged was supposed to close out a known defect and left the nightly null-key rate sitting at 3.7%. I wrote a one-page description of what meeting the bar meant at his level: the scope he should carry without help, the quality of review he should give, roughly how long a task of that size should take. Then I walked him through where his last six tasks landed against it. I said directly that this was a gap I needed to see close, not a stylistic note. I looped in my own manager and our HR partner at that point rather than later, so anything that followed matched how we handle these situations rather than my improvisation. The diagnosis was capability in one specific area, not effort. We ran eight weeks with weekly written checkpoints, a paired reviewer on every streaming change, and one question at the end: does he own that path or not. He did not get there on streaming. The batch validation path he took over instead went from 3.7% to 0.4% null keys, and he became the person we trusted with it. I moved the streaming scope permanently and wrote the change down. I would start the written bar in week three now, not month four.

Across the three teams contributing to the project I kept meeting the same failure, and it was not about individuals. People were handed whole areas with no written definition of shipping, work stalled quietly for most of a quarter, and the first real conversation happened at review time — by which point it was a shock to the person and half the evidence had gone cold. Two managers escalated cases to me that were, honestly, our process failing before anybody failed. So I built what was missing. Every area got a named owner and a one-paragraph charter stating what done means there. Contributors declared the hours they could give per cycle. Anything unmoved for fourteen days reassigned automatically with no blame attached, which removed the awkwardness that had made people wait. And I gave managers a three-step frame — name the gap, offer specific support, state the date you decide — plus a rehearsal with me before their first such conversation. Over four release cycles, milestone issues stalled longer than twenty-one days fell from 31% to 6%, and the reference-corpus null-key rate settled below 1% because the validation areas finally had owners. Two situations still ended with someone leaving the project. I believe both were fair, and both resolved in week five rather than month five. If I keep one thing from that, it is the rehearsal. Managers rarely avoid these conversations for lack of a policy. They avoid them because they have never said the sentence out loud.