Self-Assessment Quiz: dbt

Twenty questions. Aim for 16 or more.

Question 1

dbt is best described as:

• A. A data warehouse
• B. A templating engine, a dependency graph inferred from ref(), and a test runner
• C. An orchestrator
• D. An ELT platform that extracts, loads, and transforms

Question 2

dbt runs your transformation compute:

• A. On dbt's own workers
• B. Never — it sends SQL to a warehouse, which does the work
• C. On Spark
• D. Locally, then uploads results

Question 3

A dbt model file contains:

• A. CREATE TABLE AS SELECT
• B. One SELECT
• C. A transaction
• D. Whatever DDL the materialization needs

Question 4

Replacing {{ ref('stg_orders') }} with analytics.stg_orders:

• A. Fails to compile
• B. Compiles, runs, returns correct results, and deletes a DAG edge
• C. Fails at runtime
• D. Works identically

Question 5

Which consequence of that substitution is the most insidious?

• A. It is slower
• B. Slim CI will not test the model, because it is not a descendant of anything modified
• C. The docs page is missing
• D. It uses more warehouse credits

Question 6

A view materialization is:

• A. Free
• B. Free to build, and recomputed on every query
• C. Expensive to build, cheap to query
• D. Only for staging

Question 7

An ephemeral model referenced by six models is computed:

• A. Once
• B. Six times — it is inlined as a CTE into each caller
• C. Once per run
• D. Zero times

Question 8

The right size for a dbt model is:

• A. Under 100 lines
• B. One join
• C. The largest unit you would be willing to test as a whole
• D. One CTE

Question 9

Jinja has gone too far when:

• A. You use a for loop
• B. You must compile the model to know what SQL it produces
• C. You use is_incremental()
• D. You write a macro

Question 10

A loop over a query result in a model is:

• A. Fine
• B. A build-time dependency that is invisible in the DAG
• C. Faster than a literal list
• D. Required for dynamic pivots

Question 11

dbt build differs from dbt run && dbt test in that:

• A. It is faster
• B. A failed test stops downstream models, so the mart holds yesterday's correct data
• C. It runs tests first
• D. It skips snapshots

Question 12

A dbt project with 400 tests and a 100% pass rate for eight months is:

• A. Well tested
• B. Possibly untested — for each test, ask what upstream change would make it fail
• C. Over-tested
• D. Correctly configured

Question 13

When the ingestion pipeline lands nothing overnight, dbt build:

• A. Fails
• B. Succeeds, with every model built and every test passing
• C. Warns
• D. Skips the affected models

Question 14

The only thing in dbt that catches it is:

• A. not_null
• B. dbt source freshness, which is a separate command dbt build does not run
• C. relationships
• D. dbt docs generate

Question 15

dbt build --select "tag:finance,tag:hourly" selects models with:

• A. Either tag
• B. Both tags — comma is intersection
• C. Neither
• D. The first tag only

Question 16

--defer allows CI to:

• A. Skip tests
• B. Resolve ref()s to models not in the current run against production relations
• C. Run models in parallel
• D. Cache results

Question 17

profiles.yml should:

• A. Be committed so the team shares it
• B. Be gitignored, with credentials supplied by env_var()
• C. Contain production passwords for reproducibility
• D. Live inside models/

Question 18

The unknown-member pattern (COALESCE(key, -1)):

• A. Is wrong and should be removed
• B. Is correct, and converts a loud referential-integrity failure into a quiet data quality one
• C. Makes relationships tests stricter
• D. Only applies to date dimensions

Question 19

Which assertion would have caught Case Study 1's missing day?

• A. unique on the grain
• B. not_null on every column
• C. A volume assertion with a floor well below the expected value
• D. relationships to the customer dimension

Question 20

"The pipeline is fixed" and "the data is fixed" differ because:

• A. They do not
• B. Rows written during an incident fall below an incremental model's watermark and are never revisited
• C. The pipeline runs first
• D. dbt repairs data automatically on the next run

Answer Key

1. B — §19.1. Everything else is built on those three.

2. B — §19.1. dbt's own process is small enough to run on a laptop while the query it dispatched consumes a hundred nodes.

3. B — §19.2. Separating the query from its persistence is the core idea; the boilerplate dbt replaced is where the bugs lived.

4. B — §19.3. Which is why review does not catch it: the diff looks like a correct change.

5. B — §19.3. The one place you would have caught the problem is the place the problem disables.

6. B — §19.4. Kestrel's four-view chain cost $2,845 a month more than the same models as tables.

7. B — §19.4, 📐 callout. And no cost report attributes the multiplication to it.

8. C — §19.5. If you can state one assertion meaning "this model is correct," it is the right size.

9. B — §19.6. You have written a program that writes SQL, and you now own two things.

10. B — §19.6. It runs during parsing and fails with error messages pointing at the wrong file.

11. B — §19.7, 🔎 callout. Stale beats wrong, and it is not close.

12. B — §19.7. A pass rate is being read as evidence.

13. B — §19.8. Structural: dbt transforms what is in the warehouse and cannot know what should have been.

14. B — §19.8. And Case Study 2's version was configured, correct, and never invoked.

15. B — §19.12. Space is union. A scheduled job with the wrong one builds a subset and reports success.

16. B — §19.14. Twelve models built against real production upstreams instead of ninety.

17. B — §19.13. It is the most commonly committed secret in the data world, by well-meaning people, because it looks like configuration.

18. B — Case Study 2. You still want it; the alternative silently drops rows from every aggregation. It just creates a monitoring obligation.

19. C — Case Study 1, 🔎 callout. A, B, and D all pass on an empty table — every standard dbt test describes rows that exist.

20. B — Case Study 2. The remediation everyone does makes the symptom go away; the one nobody does is the one nothing will report as missing.

Topic map