Analytical scans are not the only workload a lakehouse table serves. In many production query mixes, a large share of queries are needle-in-haystack reads: fetch one order by order_id, pull a user's profile by user_id, trace a request by uuid, list all events for one customer_id.
