Tell me about a time you led a team through a crunch without burning people out.

I maintain the quality working group on an open-source project — around twenty-two active contributors, nearly all volunteering around day jobs. We had committed to a compatibility-breaking major release with a six-week hardening window, and by the end of the first week the incoming queue was growing faster than we were closing it. My first decision was made on the group's behalf rather than my own: I cut the release scope. Three planned migrations came out and were announced as next-cycle work, with the reasoning written into the milestone. That cost us goodwill with two downstream consumers who had been waiting, and I said so in the announcement instead of pretending the cut was free. Then I restructured the load. Triage duty had been landing on the same four people every week, so I made it a rotating seat with a written handover, and a rule that nobody serves two weeks running. I capped the release checklist at what one person can finish in a two-hour sitting, because a checklist nobody completes is a checklist nobody trusts. I also stopped posting requests for volunteers late in the evening; those get picked up by whoever is least able to decline. We shipped four days past the original target, which the maintainers accepted because they had seen the risk from week two. Reopen rate on the release fell from seventeen percent to nine. What I care about more is that all twenty-two contributors were still active the next cycle, and the rotating seat is still how the group runs.

The pattern I was seeing was larger than any one release. Three sibling repositories in the same ecosystem each ran their own hardening push, and the same handful of experienced reviewers were pulled through all three in sequence. Two of them stepped back from the ecosystem entirely inside a year, which is a far more expensive outcome than a late release. So I proposed something at the ecosystem level and spent about ten months getting the maintainers of all three repositories to adopt it. Three parts. A shared release calendar so hardening windows do not overlap, which alone removed most of the double-booking. A written load contract per repository: a named rota, a limit on consecutive weeks of duty, and an explicit rule that scope is cut before effort is extended. And an agreed definition of a release-blocking bug, settled in advance so that negotiation happens when nobody is tired. The difficulty was not the design. One repository read the duty limit as a loss of autonomy. I let them run a modified version and report back rather than forcing the original, and their variant is the one the other two eventually adopted. Across the three repositories, reopen rate on release fixes settled at eleven percent against the twenty-one we started from, and the number of people willing to serve triage grew once the commitment was bounded and legible. I have since stepped away from two of those repositories, and the contract is still in the contributor docs. That is the test I hold myself to now — does it survive me leaving.