dbt Development

Targets

Four targets are available for school projects (kipptaf, kippnewark, kippcamden, kippmiami, kipppaterson):

Target	Purpose
`defer`	Default. Sources in `zz_<user>_*`, cross-regional sources read from prod. CLI `--defer` resolves unchanged models to prod manifest.
`dev`	Full isolation. Cross-regional sources also in personal namespace. Use when developing regional models and testing end-to-end through kipptaf.
`staging`	CI-equivalent builds. Sources in `zz_stg_*`. Used by dbt Cloud CI.
`prod`	Production. Default target in shipped profiles (Dagster). Developers use `defer` (default in `.dbt/profiles.yml`).

Defer to production

The dbt Core Tools extension handles --defer automatically — unchanged models resolve to the prod manifest. Prod manifests are generated by the post-merge git hook (runs on every git pull/merge to main).

Staging external sources

External tables must be staged (created/refreshed via DDL) before any dbt model that depends on them can build. dbt build does not stage sources — Dagster handles this in production, but dbt Cloud CI does not.

If your PR adds or modifies an external source, you must stage it with --target staging before the CI job runs. The zz_stg_* tables persist across PRs, so this only needs to happen when a source definition actually changes.

Use the VS Code task dbt: Stage External Sources — it prompts for project, target (defer/dev/staging), and source selection.

Terminal equivalent:

uv run dbt run-operation stage_external_sources \
  --project-dir src/dbt/<project> \
  --target <target> \
  --vars '{"ext_full_refresh": "true", "cloud_storage_uri_base": "gs://teamster-<project>/dagster/<project>"}' \
  --args 'select: <source>'

Warning

run-operation does not support --defer or state:modified — there is no way to stage only changed sources automatically. Target the specific sources you changed (e.g., --args 'select: amplify.my_new_table') rather than staging everything.

Cross-project development

When modifying regional models and testing through kipptaf:

Develop the regional project with --target defer — outputs go to zz_<user>_kippnewark_* (or the relevant region).
In kipptaf, use --target dev — cross-regional sources resolve to zz_<user>_kippnewark_*.
Test end-to-end locally.

Stale dev schemas

Your zz_<user>_* datasets in BigQuery accumulate over time. Drop them when you want a clean slate.