Talent search returned stale rankings

Root cause

Cron worker was OOM-killed mid-cycle; the next scheduled run started cleanly but the gap was not detected until a recruiter flagged a known-stale profile.

Customer impact

Two recruiter seats viewed top-25 lists for two distinct SkillQuery records that excluded a profile updated within the prior 90 minutes.

Remediation

• Doubled the indexer pod memory limit and set GOMEMLIMIT to match.
• Added freshness telemetry — alert if the last-completed-cycle age > 75 minutes.
• Documented the recovery playbook step for catching up missed indexer cycles.