The short answer
A late-arriving dimension (or early-arriving fact) happens when a fact arrives before its matching dimension row exists, or the dimension update that should have applied at event time shows up late.
Fact load 09:00: order for customer_id C-99, product SKU-7
Dim load delayed: C-99 not in dim_customer yetWhy it hurts
- Orphan
- Facts temporarily map to "unknown"
- SCD2 point-in-time lookup may pick the wrong version if history is repaired late
Common handling patterns
- 1Unknown member: assign
customer_sk = -1and repair later with an update/re-key job - 2Inferred dimension row: insert a stub dim row with the and placeholder attributes; enrich when the real feed arrives
- 3Delay fact publish: wait for dim dependency (hurts latency)
- 4Reprocessing: when late dim history arrives, re-lookup for affected fact date ranges
Interview tip
> "Facts can arrive before dims. Use unknown or inferred members, then repair keys when the dimension catches up (especially careful with SCD2)."