Data Exploration — DBX Studio

Exploration Workflow

Phase 1: Schema Discovery

Start with read_schema to list all tables, then describe_table for each table of interest.

1. read_schema(schema_name: "public")
2. describe_table(table_name: "<each table>")
3. get_table_stats(table_name: "<table>")

Phase 2: Table Profiling

For each table, gather:

• Row count
• Column names and types
• Sample data via get_table_data
• Null counts and distributions

Phase 3: Relationship Discovery

Look for foreign key patterns:

• Columns named *_id linking to other tables
• Common join patterns: users.id → orders.user_id

Quality Scoring

Score	Completeness
Green	> 95% populated
Yellow	80–95% populated
Orange	50–80% populated
Red	< 50% populated

Common Exploration Queries

Row count

SELECT COUNT(*) AS row_count FROM "public"."table_name";

Column null rates

SELECT
  COUNT(*) AS total,
  COUNT(column_name) AS non_null,
  ROUND(100.0 * COUNT(column_name) / COUNT(*), 2) AS pct_filled
FROM "public"."table_name";

Distinct values

SELECT column_name, COUNT(*) AS frequency
FROM "public"."table_name"
GROUP BY 1
ORDER BY 2 DESC
LIMIT 20;

Date range

SELECT MIN(created_at), MAX(created_at) FROM "public"."table_name";

Output Format

After exploration, present a structured summary:

• Tables: list with row counts
• Key relationships: how tables connect
• Data quality flags: any columns with high null rates
• Suggested next queries: what the user might want to know next