The short answer
A junk dimension packs many small, low-cardinality flags and indicators into one instead of creating dozens of tiny dims or leaving clutter on the fact.
Examples of flags: is_gift, is_mobile_order, payment_method, coupon_used_flag, channel_type.
dim_order_flags (junk)
order_flags_sk
is_gift
is_mobile
payment_method
has_coupon
fct_orders
...
order_flags_skWhy use it
- Keeps the fact narrow (one key instead of 10 flag columns)
- Avoids a swarm of one-attribute dimensions
- Combinations of flags are usually a manageable cartesian set
Caution
If attributes are high or independently huge, junk dims explode. Keep junk for true low-cardinality misc flags.
Interview tip
> "Junk dimension = miscellaneous flags bundled into one dim to declutter the fact."