Database normalization puts each fact in one appropriate place and lets related records reference it. It reduces update anomalies, but reads may require more joins. section 7.4
Use the schema to ask where each fact lives and why exceptions exist:
- Give repeating relationships their own tables: Booking connects to Attendee rather than storing multiple guests in one crowded column. section 7.4
- Recognize denormalization as a deliberate read optimization: a flattened routing-form copy avoids repeated JSON parsing and joins. section 7.4
- Treat synchronization mechanisms as part of the design: a PostgreSQL trigger keeps the flattened copy current, but can drift if defective. section 7.4
Table 7.3 A mostly normalized schema, with one denormalized table to speed up a high-volume analytics query.
|
Posture
|
Where in
Cal.com
|
What it
buys
|
What it
costs
|
|
Normalized
(one fact, one place)
|
Booking → Attendee, Booking → Payment, User → Schedule
|
No update
anomalies; a fact changes in exactly one row
|
Every read
pays for joins, and a wide read touches many tables
|
|
Read-denormalized
(copy, kept fresh by trigger)
|
RoutingFormResponseDenormalized,
RoutingFormResponseField
|
Analytics
and filter queries skip the JSON parse and the joins
|
A trigger
you have to trust; the copy can drift if the trigger has a bug
|
|
JSON escape
hatch (no shape at all)
|
Booking.responses, Booking.metadata, EventType.locations
|
Ship a new
field with zero migration
|
Unqueryable
without a GIN index; the schema stops describing the data
|
Compare normalized tables, read-denormalized copies, and JSON escape hatches by their benefits and operational costs. section 7.4



