Ask the boardroom question. Get the governed answer.

Use the sample datamart like an executive would: ask a revenue or engagement question, watch the analysis run, then review the memo, caveats, and source queries.

What should we inspect?

Answers are generated only when metric definitions, grain, and source SQL are reviewable.

Analysis ready

Ask from sample data

Ready to analyze sample data Choose a question, then run a local read-only query plan.

Map question

Validate SQL

Run DuckDB

Hash results

Draft brief

Window Jan-Mar Grain month x surface Mode read-only SQL

Executive answer

The memo appears in stages as the run validates context, executes SQL, and records caveats.

Waiting for analysis Choose a sample question or click Analyze. The answer, decision lens, caveats, and SQL appendix will appear as the local query run advances.

SQL appendix will appear after read-only queries finish.

Total ad revenue

$973M

+$48M vs Feb

Growth deceleration

-$21M

change in monthly lift

Strongest surface

Shopping

+$38M in March

Primary drag

Video

-$14M in March

More details

The first screen is the executive experience. This section shows the evidence, workflow, and guardrails behind it.

Ranked findings

Each material claim is backed by executed SQL and dbt-documented context.

2 source queries

Revenue still grew, but the monthly lift fell from +$69M to +$48M.

The slowdown is a deceleration story, not a contraction story. The total reached $973M in March.

q002_revenue_growth_waterfall

Shopping and visual search added $62M combined, offset by a video decline.

The commercial-intent surfaces compounded while brand video reduced the net March gain.

q001_surface_revenue

The model can answer this at month x surface grain, but not at advertiser cohort grain.

The context file does not document cohort joins for this mart, so the MVP keeps that caveat explicit.

datamart_context.json

Surface mix

Chart data is exported from the same query results used in the brief.

JanFebMar

Shopping Visual search Brand video

How the run works

The agent narrows the question to documented metrics before producing the memo.

Trust console

The MVP records exactly what it used before writing the brief.

Run health

SQL modeRead-only

Mutating statementsRejected

ManifestLoaded

Join coverageLimited

Result hashesRecorded

Follow-up analyst

Ask a narrower question without losing provenance.

Brand video should be inspected first because it declined by $14M while the commercial-intent surfaces grew.

Sales pitch

Package the governed-agent workflow as an internal operator tool: installable skills, local data execution, and board-ready output.

Trust-first answers Every numeric claim maps to SQL, result hashes, metric docs, and caveats.

Fits local workflows Runs against local DuckDB and dbt artifacts without sending data to external services.

Marketplace-ready shape Ships as focused skills with plugin manifests for agent installation and internal distribution.