EXECUTABLE PRODUCT DEMO

Follow a question.
Run the evidence.

An AI-assisted SQL investigation workbench for data engineers. Try its SQL investigation pattern here with fictional data.

Synthetic data · SQLite in your browser

Actual SQL execution. Authored guided explanations. No live Claude, database login, local service, or paid API endpoint.

Use PostgreSQL + Claude in the local app →

BUSINESS QUESTION

Why does this shipment JOIN show 32 ordered units when we ordered only 22?

SQL / SQLITE

Query editor

Loading engine

SQL query

Execution results

Run the initial query to inspect its rows.

Then load and run a diagnostic. Nothing executes until you choose Run SQL.

GUIDED CONTENT · NO CLAUDE CALLS

Investigate, then correct.

Compare actual executions with versioned fixture expectations. Each check explains what it can establish.

Check shipment key cardinality
Find non-unique join keys. Two valid partial shipments are not necessarily duplicate business events.
Awaiting execution
Count and total the base lines
Establish the base order-line grain and measure before joining.
Awaiting execution
Measure the joined total
Compare the same measure after the join; a repeated left-side quantity changes this total.
Awaiting execution
Check unmatched shipments
Find shipment rows outside the order-line match. This is a separate coverage problem, not the cause of left-side inflation.
Awaiting execution

Compare the correction

Return one row per order line by bringing shipments to the same grain before the join.

AUTHORED FIXTURE EXPLANATION

What caused it?

LINE-001 has two valid partial shipments. The join repeats its 10 ordered units twice. The joined measure is 32 rather than 22; pre-aggregating shipments by line restores three order-line rows. Do not delete legitimate shipment events just to make a join unique.

0/6 expected query results verified by your executions

This explanation is written for the synthetic fixture. Execution links above and the exported report contain the actual rows you ran.

REMAINING UNCERTAINTY

The correction assumes an order-line report. A shipment-level report legitimately has several rows per line. The unmatched SHIP-005 is a separate issue; these synthetic checks do not establish production reconciliation.

Continue with a real connection

The local application supports PostgreSQL, explicit SQL approvals and optional Claude assistance. Claude is unverified live for this release. BigQuery is implemented but not live-validated.

Read setup and verification details →