Flatland vs.
dbt.

-- A dbt model declares a transformation
-- with typed columns, declared sources,
-- and tests as first-class assertions.

{{ config(materialized='table')  }}

select
  order_id::uuid,
  customer_id::uuid,
  order_total::numeric,
  order_status
from {{ ref('stg_orders') }}
where order_status != 'cancelled'

# schema.yml
tests:
  - not_null:
      column_name: order_id
  - unique:
      column_name: order_id

dbt's graph is data-flow: source tables → staging models → marts. Tests are first-class assertions. The compiler runs against your warehouse.

# Flatland declares typed drivers
# with declared dependencies and
# assertions as first-class guardrails.

driver arr {
  category: "computed",
  type:     Currency(USD),
  formula:  "arpu * customers * 12",
  depends_on: [arpu, customers]
}

assertion burn_mult_ok {
  expression: "burn_multiple < 1.5",
  severity:   "structural"
}

✓ Currency / Count → Currency
✓ dependency graph resolved
✓ assertion checked at compile

Flatland's graph is reasoning-flow: assumptions → computed drivers → outputs. Assertions are first-class guardrails. The compiler runs in memory against the typed IR; no warehouse needed.