Structuring tables sensibly.
WHAT NORMALISATION IS
Organising data so each fact is stored once.
WHY IT MATTERS
Duplicated facts diverge, and updating them all is error-prone.
WHAT THE PRACTICAL RULES ARE
Each column holds one value Each column depends on the whole key Nothing depends on anything but the key
WHAT THAT MEANS IN PRACTICE
Repeating groups become separate tables Descriptions of one thing do not live in another thing's table
WHAT AN EXAMPLE OF THE PROBLEM IS
A customer's address stored on every order.
WHY IT IS A PROBLEM
Changing the address means changing every order, or accepting inconsistency.
WHEN THAT IS ACTUALLY CORRECT
When the historical value matters.
WHY
An invoice must show the address at the time, not the current one.
WHAT THAT ILLUSTRATES
Normalisation is about facts, and a historical record is a different fact.
WHAT DENORMALISATION IS
Deliberately duplicating data for speed.
WHEN IT IS JUSTIFIED
When a join is genuinely too expensive, and you have measured it.
WHAT IT COSTS
Keeping copies consistent, forever.
WHAT TO DO FIRST
Normalise, then denormalise only where evidence demands.
WHAT TO NEVER DO
Denormalise pre-emptively.
