Fellowship Recruitment Post-Mortem · October 2026
6,629 applications. 30 spots.
How we rebuilt our Fellowship review mid-cycle: what we got right, what we got wrong, and what we fixed along the way.
*By submitting your email address, you agree to receive the Climate Cardinals monthly newsletter. You can unsubscribe at any time. You can also read the report without your email here.
The report should have opened in a new tab. If it didn't, open it here.
& Lea Obermüller
The cycle in numbers
Our learnings aren't just for us.
We're sharing what worked and what broke, so other youth-led organizations and nonprofits can build fairer, more effective recruitment.
See the open-source code on GitHub →Build a margin for error into any cutoff
Scores, AI or human, are estimates. Set thresholds that account for that, and say why.
Never let one automated score decide
AI managed the volume. Blind human review made every decision that mattered.
Test before you trust — then keep testing
We spot-checked against human scoring before going live, and caught failures with automated checks.
Be transparent about what didn't work
A long-response scoring error and a duplicate-detection gap — found, fixed, and documented in the report.
Read the full post-mortem.
Seven sections. Every number, every bug, every fix.