Aggregated odds are only as good as their weakest source. A scanner that treats a 90-second-old price as live will hallucinate arbs and edges against a book that has simply gone stale - so production systems check per-book freshness before acting, and drop books that fall out of tolerance.
ParlayAPI publishes per-book freshness live at /v1/meta/source-quality, with the per-book thresholds it holds itself to at /v1/meta/per-book-sla. Wiring these into your pipeline - skip books currently out of SLA - is the single highest-value robustness check for any odds consumer.