• One project, one pipeline, three stages.
• Development (the sandbox you build in) has no branch of its own.
• Staging deploys to the staging sandbox from dcstaging1.
• Production deploys to the production org from main.
• Activate the pipeline before you create work items. Otherwise moving one to In Progress fails with An unexpected error occurred.

• A File Upload data stream from a CSV, which creates its DLO.
• Map the DLO to a DMO.
• Build it in the default data space, which exists in every org.

The example: 50 branch offices, CSV stream → branch_offices_seedcsv__dll → custom DMO Branch_Office__dlm.

• Data kit type: DevOps, never Standard. Only a DevOps data kit can be built in a sandbox, moves by Metadata API, and carries its connector information. You can't switch types later.
• Add the data stream bundle. Its DLO, DMO and mappings come with it.
• Check the Publishing Sequence: connection, then bundle.

A "You haven't defined the publishing sequence" banner may stay after you save it. It's cosmetic: the order still travels and is used in every target.

• In the Hub: Lifecycle Center → Work Items → New. Pick the project, and the dev sandbox as the Development Environment.
• Click Mark as In Progress. Lifecycle Center creates the branch WI-00000N from dcstaging1.
• Note the work item's name; the script needs it.

DX Inspector stages every component, then the commit fails at finalize. Retrying doesn't help. It fails the same way for any data kit, large or small.

• Committing the data kit's components yourself (next step) is the only step that changes.
• Ordinary metadata (flows, permission sets, fields) still commits through DX Inspector. Put it in a separate work item.
• No failed DX Inspector attempt is needed before running the script.

• In DX Inspector in the dev sandbox, on the work item, click Create Review. Lifecycle Center opens a pull request from WI-00000N to dcstaging1.
• The work item moves to In Review. Review the pull request as your process requires.
• Click Mark as Ready to Promote. Don't merge the pull request on GitHub.

• Open the pipeline. The work item is under Approved Work Items.
• Select it and click Promote Selected.
• Options: Only changes missing from Staging, No tests (a data kit has no Apex).
• Lifecycle Center merges into dcstaging1 and deploys to the staging sandbox.

Data Cloud Setup → Other Connectors → UploadedFiles (File Upload). An inactive connection makes the component deploy fail.

• Staging: it was inactive; activate it.
• Production: it was already active, and the data kit left it alone. Skip the step.

Check the connection's stateCopysf data query -o afd360-dcstaging1 --query "SELECT DeveloperName, ConnectionState FROM MktDataConnection WHERE DeveloperName = 'UploadedFiles'"

• Data Streams → the stream → Update File.
• Keep Full Refresh, upload the CSV, wait for File Upload Complete, click Deploy.
• The stream turns Active and runs once. Rows reach the DLO and the DMO.

Needed once per org, because a data kit carries the stream's configuration but never the uploaded file. Later component deploys keep the stream and its data.

• In the Staging column, click Promote Stage. It promotes every work item in Staging together, by merging dcstaging1 into main.
• Keep the defaults: Promote Now, Only changes missing from Production, No tests.

1. Create the calculated insight in the dev sandbox (the UI, or the example script below).
2. Data Kits → the data kit → add the insight. Check the Publishing Sequence lists it after the bundle.
3. New work item, In Progress, then the same commit script.

Example insight, dev sandbox onlyCopy./scripts/create-calculated-insight.sh -o afd360-dcdev1 --run

The commit carried only the change: 3 files (the insight template, its data kit membership, and the data kit definition's new deploy order). The same data kit is updated; no new data kit is needed.