We spent a year filtering for the title "SRE" and hiring mediocre operators. The best reliability engineer we found had "support technician" on their résumé.

Our first SRE hiring loop had a keyword filter for "site reliability engineer," "SRE," or "platform engineer" in the résumé's title field. It filtered out a candidate who'd spent four years as the sole "support technician" keeping a regional bank's ATM network online, and who turned out to be the strongest debugger we've ever interviewed.

The title is a bad proxy

SRE as a job title is maybe a decade old at most companies. The skill it names — staying calm while a system degrades, forming and testing hypotheses under time pressure, communicating clearly to people who are panicking — shows up in ER nurses, air traffic controllers, restaurant kitchen leads, and yes, ATM support technicians, long before it shows up in a résumé's title field. We removed the keyword filter and replaced it with a structured work-sample interview instead.

interview_stage: incident_simulation
duration_min: 45
setup: |
  Candidate is given a dashboard showing a degrading but not-yet-down
  service and a 10-minute-old, incomplete alert. No prior context.
scored_on:
  - hypothesis_formation   # do they generate testable theories, not guesses
  - prioritization_under_uncertainty
  - communication_while_investigating  # do they narrate, or go silent
  - grace_under_a_dead_end  # what happens when their first theory is wrong
not_scored_on:
  - familiarity_with_our_specific_stack
  - years_of_experience
  - job_title_history

What changed in the offer pool

Removing the title filter and adding the simulation nearly doubled our qualified candidate pool in six months, and our offer-acceptance rate went from 61% to 84% — because candidates who'd been screened out everywhere else for lacking the right pedigree were thrilled to be evaluated on what they could actually do.

"Nobody had ever asked me to just think out loud while something broke. Every other interview asked me to recite Kubernetes trivia I'd have looked up anyway."

What we still get wrong

We're not claiming this is solved. Our incident simulation still implicitly rewards people comfortable talking while thinking, which correlates with confidence more than competence for some candidates, particularly non-native English speakers. We're now piloting a written-narration option for the same simulation, scored identically, to see if it closes that gap.

  • Job titles are a weak, biased proxy for reliability skill — screen on simulated behavior instead.
  • A structured incident simulation, scored on hypothesis formation and communication, transfers across industries far better than stack-specific trivia.
  • Removing a filter that felt safe can widen your pool more than any sourcing campaign.