bi-validation

name	bi-validation
description	BI validator driven by ValidationHook — inspects every dashboard, chart, and dataset delivered in the run, verifies config quality (chart type, metrics, dimensions, dataset wiring) and data presence via get_chart_data when supported
tags	["bi","dashboard","publish","validation","metrics"]
version	2.0.0
user_invocable	false
disable_model_invocation	false
allowed_agents	["gen_dashboard"]
kind	validator
severity	blocking
mode	llm
targets	[]

BI Validation

Driven by ValidationHook.on_end for gen_dashboard runs. The hook passes a SessionTarget containing every DashboardTarget / ChartTarget / DatasetTarget the run delivered. Iterate them and run the checks below.

Target shape

You receive SessionTarget.targets — loop over each entry and dispatch on type:

• dashboard → inspect chart list via get_dashboard / list_charts
• chart → inspect config via get_chart; if supported, run get_chart_data
• dataset → verify fields via get_dataset

Layer A (the builtin hook) has already confirmed each resource exists and is reachable — skip existence checks and focus on config correctness and data presence.

Publish success criteria

A dashboard publish is complete when:

• get_dashboard returns the dashboard
• the expected chart ids / names are present in get_dashboard or list_charts
• every chart can be inspected with get_chart
• supported get_chart_data calls return data or a valid empty result without backend errors
• known values match expected results or tolerances

When these criteria pass, return PASS and stop. Treat additional styling or structural changes as a separate request unless a concrete failing check is present.

Core workflow

For every target in the session:

1. Dashboard targets: call get_dashboard(dashboard_id) to retrieve the chart list. For each chart on the dashboard call get_chart and inspect config (step 2). When get_chart_data is supported, validate data (step 3).
2. Chart targets (standalone or from step 1): verify chart_type, metrics, x_axis, dimensions, dataset_id against the intended design.
3. Data presence: when get_chart_data is supported, call it on every chart to confirm the chart returns data without backend errors. Compare numeric values against expected tolerances when expectations are available.