Row coherence

Faker-style generation produces rows that are individually plausible but mutually absurd — a created_at after updated_at, a first name that doesn't match the email, a US city with a German postcode. django-data-seed makes each row internally consistent so aggregate data passes the sniff test.

Coherence runs after the raw values are generated and before foreign keys are attached, and it never overrides a value you supplied via an override or a unique column.

Ordered lifecycle dates

Date and datetime fields on the same row that describe a lifecycle are generated as an ordered set, inferred from their names, with realistic gaps:

created_at  ≤  updated_at  ≤  shipped_at  ≤  delivered_at

So an order's delivered_at is always after its created_at, not a random moment that might predate it.

One persona per row

Name-like fields on a row are driven by a single person, so they agree:

first_name = "Malik"
last_name  = "Bolton"
full_name  = "Malik Bolton"     # not "Priya Nakamura"

One address per row

Location fields (city, state, postal_code, country, …) are drawn from a single locale record, so a row's city and postcode belong to the same place rather than being three unrelated countries stitched together.

Turning it off

Coherence is part of the smart realism layer. Run with --realism uniform to generate every field independently (faster, but incoherent) when you don't care about how the fields relate.