Data & Analytics
dbt v2 can skip freshness-selected models from a v1 baseline
A reported sources.json status-case mismatch can turn a state-based dbt build into a successful no-op.
November 17, 2026
A migration can preserve sources.json yet lose the build decision it was supposed to carry. dbt issue #17161 reports that v1 writes lowercase freshness statuses while v2's source_status reader accepts title case. With a v1 baseline, source_status:fresher+ selects nothing, warns and exits 0. That makes a scheduled build look successful without updating models.
Check the state file at the version boundary
Before the first v2 production run, capture the baseline sources.json, its dbt version, status values and result count. dbt's artifact guide identifies this file as the source freshness result. Do not edit a production artifact in place to change capitalization; preserve it as evidence and use a disposable copy to test compatibility.
Assert the selected model set
Run dbt ls -s 'source_status:fresher+' --state ... --resource-type model against the saved v1 artifact and compare the selected IDs to the v1 job. The selector reference describes how freshness results drive this selection. Treat an empty set as an error when fresh inputs were expected; capture warnings in the gate.
Bridge without hiding the problem
Regenerate a v2 baseline only after confirming it represents the same source state and does not skip the first required build. Alternatively keep the v1 runner for this job until the v2 reader accepts the artifact. The issue links a write-side fix that could change casing again, so test both directions when upgrading.
Adovo’s view
State files are interfaces between runs and between versions. If a pipeline trusts them, their compatibility is part of the migration plan. The first successful process exit after an upgrade tells you little until you prove the intended models were selected.
