Skip to content
LakeBench
ProblemsCommunityPricing
Sign inStart practicing
Back
  1. Home
  2. Interview prep
  3. Data reconciliation

Data quality · Extra High-Value

Data reconciliation

Mediumquality-18
reconciliationverificationrow countsfinance

Question

What is data reconciliation in a data platform?

Solution

Reconciliation compares two datasets that should agree and quantifies differences. It is the practical form of verification: proving curated data matches a source of truth within tolerance.

Common reconciliations

  • Row counts: source day vs warehouse day
  • Financial sums: Stripe charges vs fct_payments.amount
  • Key sets: IDs in source minus IDs in warehouse (missing/extra)
  • Cross-system: OLTP total vs lakehouse total vs BI extract
-- conceptual day-level money reconcile
with src as (
  select sum(amount) as amt from raw_stripe_charges where charge_date = '2026-09-05'
),
wh as (
  select sum(amount) as amt from fct_payments where payment_date = '2026-09-05'
)
select src.amt as source_amt, wh.amt as warehouse_amt,
       src.amt - wh.amt as delta
from src cross join wh;

Tolerances

Exact equality is not always possible (timing, FX, pending states). Define absolute or percent thresholds and known exception codes.

Interview tip: "Reconciliation = compare aggregates/keys to a trusted source." Give count + sum examples and mention tolerances.

PreviousNext