Build your first dashboard
The included Sales project is a complete example of the dashboard-as-code workflow. Make and validate a small change there before creating connections, Models, and semantic models from scratch.
Before you begin
Complete Installation, bootstrap the Olist data, and keep the development target local. Use a branch where changing the example label and note is safe.
Follow one reviewable loop:
- Confirm the unchanged sample dashboard works.
- Trace the resources that compose it.
- Change one semantic label and one dashboard note.
- Validate and plan the complete source root.
- Deploy to development and verify the rendered behavior.
flowchart LR
accTitle: Dashboard authoring lifecycle
accDescr: Authors edit project resources, validate them, review a deployment plan, deploy the candidate, and verify the active dashboard.
edit["Edit resources"] --> validate["Validate"] --> plan["Review plan"] --> deploy["Deploy"] --> verify["Verify"]
verify -. "iterate" .-> edit
Start the sample project
Prepare the Olist sample data and start the managed development server. The development task provisions a worktree-local PostgreSQL service and admits its local physical pool before publishing the sample candidate:
task bootstrap
task dev
The server writes worktree-local process state and logs beneath .tmp/. Open the URL printed by task dev, open the project resource browser, and choose Executive Sales. Confirm that the KPI cards, revenue trend, category chart, filters, and orders table load before editing files.
Trace the resources
The report is assembled from these files:
dashboards/connections/olist.yaml
dashboards/sources/olist.*.yaml
dashboards/models/*.yaml
dashboards/semantic-models/sales.yaml
dashboards/pipelines/*.yaml
dashboards/dashboards/executive-sales.yaml
dashboards/access/*.yaml
Read them from the outside in. The source-root loader discovers shared Olist inputs, Models, the sales semantic model, refresh pipelines, dashboards, and access rules into one graph. The dashboard refers to fields and metrics exposed by the sales semantic model.
Add a semantic metric
Open dashboards/semantic-models/sales.yaml. Its revenue and order_count metrics are the inputs to the existing average-order-value metric:
metrics:
aov:
type: ratio
label: Average order value
numerator: revenue
denominator: order_count
format: currency
For a first change, update only the label to Average revenue per order. This changes presentation metadata without changing the metric identity or formula. Stable IDs such as aov let dashboards and API clients continue referring to the same semantic field.
Validate the entire project, not just the edited file:
go run ./cmd/leapview validate --source-root dashboards
If validation reports a location, fix the resource before continuing. Common first-edit failures are incorrect indentation, an unknown field, or a reference to a semantic name that does not exist.
Change the dashboard
Open dashboards/dashboards/executive-sales.yaml. Find the aov_kpi visual and change its note:
aov_kpi:
type: kpi
query:
metrics:
aov:
presentation:
note: Average revenue per completed order
tone: warning
The visual owns its semantic query and typed presentation. A page component later places that reusable visual on the grid by referencing aov_kpi.
Review and publish the change
Inspect the candidate before activating it:
PLAN_JSON=$(go run ./cmd/leapview plan --source-root dashboards --format json)
PLAN_ID=$(printf '%s' "$PLAN_JSON" | jq -r .planId)
BUILD_JSON=$(go run ./cmd/leapview build "$PLAN_ID" --format json)
CANDIDATE_ID=$(printf '%s' "$BUILD_JSON" | jq -r .candidateId)
go run ./cmd/leapview publish "$CANDIDATE_ID"
Apply it to the managed development target:
task dev:publish
Validate the project
Run validation and planning once more immediately before deployment if another edit occurred after the earlier checks:
go run ./cmd/leapview validate --source-root dashboards
go run ./cmd/leapview plan --source-root dashboards
The plan should contain only the resources you intended to change. Unexpected additions or removals usually indicate a discovery-pattern or stable-ID mistake.
Verify the dashboard
Reload Executive Sales and confirm both label changes. Also change a date or state filter to verify that the KPI still participates in the shared query lifecycle.
LeapView validates the complete project graph before switching the active serving state. A rejected candidate does not replace the last valid serving state. This makes validation and plan review normal parts of authoring rather than recovery steps.
Troubleshooting
If the dashboard does not change, confirm the deployment targeted the same local instance shown in the browser and inspect task dev:logs. If validation cannot find aov, verify that the metric ID was not renamed. If the KPI changes but filtering does not, compare its semantic query and interaction scope with the original definition before editing presentation code.
Next steps
Revert or keep the example changes as appropriate for your branch, then continue with Build dashboards for the complete workflow.