Find Tables

2. Search Strategy

Use appropriate carto-cli commands to:

• List available connections and data sources
• Search for tables by name, description, or category
• Filter results based on user criteria

3. Present Findings Clearly

For each relevant table found, provide:

• Full table path (project.dataset.table)
• Brief description of the data content
• Key columns relevant to the user's needs
• Geographic coverage and temporal extent when available
• Data freshness/update frequency if known

4. Recommend Best Matches

Rank tables by relevance to the user's stated needs and explain why each recommendation fits.

Key carto CLI Commands

Before any operation, check authentication status:

carto auth status

If not authenticated, login first with carto auth login.

Essential Commands

1. List connections (if user hasn't specified one):

   carto connections list
   

2. Browse datasets/tables in a connection:

   carto connections browse <connection-name>
   carto connections browse <connection-name> "project.dataset"
   

3. Search tables by pattern (recommended for large datasets):

   carto sql query <connection> "
     SELECT table_name
     FROM \`project.dataset.INFORMATION_SCHEMA.TABLES\`
     WHERE table_name LIKE '%keyword%'
     ORDER BY table_name
   "
   

4. Get table schema and metadata:

   carto connections describe <connection> "project.dataset.table"
   

5. Preview table data:

   carto sql query <connection> "SELECT * FROM \`project.dataset.table\` LIMIT 10"
   

6. Analyze column value distributions (to understand the data):

   # Get distinct values for categorical columns
   carto sql query <connection> "SELECT DISTINCT column_name FROM \`project.dataset.table\` LIMIT 50"
   
   # Get value counts for a column
   carto sql query <connection> "
     SELECT column_name, COUNT(*) as count
     FROM \`project.dataset.table\`
     GROUP BY column_name
     ORDER BY count DESC
     LIMIT 20
   "
   
   # Get numeric column statistics
   carto sql query <connection> "
     SELECT
       MIN(numeric_col) as min_val,
       MAX(numeric_col) as max_val,
       AVG(numeric_col) as avg_val,
       COUNT(*) as total_rows
     FROM \`project.dataset.table\`
   "
   
   # Get date range for temporal data
   carto sql query <connection> "
     SELECT
       MIN(date_col) as earliest,
       MAX(date_col) as latest
     FROM \`project.dataset.table\`
   "
   

For full documentation on CLI commands, run carto --help or carto <command> --help.

Proactive Data Profiling

When you find a potentially relevant table, proactively analyze it before presenting to the user:

1. Get the schema with carto connections describe
2. Identify key columns based on the user's analysis goal
3. Profile relevant columns to understand:
   • What values exist (for filters/categories)
   • Geographic coverage (query distinct regions/countries)
   • Temporal extent (min/max dates)
   • Data volume (row counts)
4. Present findings with context, not just column names

Example Proactive Analysis

If user asks for "retail foot traffic data in California":

# 1. Find the table
carto connections describe <conn> "project.dataset.retail_foottraffic"

# 2. Check geographic coverage
carto sql query <conn> "SELECT DISTINCT state FROM \`project.dataset.retail_foottraffic\` WHERE state LIKE 'CA%' OR state LIKE 'Calif%'"

# 3. Check date range
carto sql query <conn> "SELECT MIN(visit_date), MAX(visit_date) FROM \`project.dataset.retail_foottraffic\`"

# 4. Check available metrics
carto sql query <conn> "SELECT * FROM \`project.dataset.retail_foottraffic\` LIMIT 5"

Then present: "This table has California data from 2020-2024, with columns for visits, unique_visitors, and dwell_time. The geographic granularity is at the store level with lat/lon coordinates."

Output Format

When presenting table recommendations, use this structure:

### Recommended Tables for [Analysis Goal]

1. **[table_name]**
   - Path: `project.dataset.table`
   - Description: [what the table contains]
   - Relevant columns: [key columns for the analysis]
   - Coverage: [geographic/temporal scope]
   - Fit: [why this table matches the user's needs]

Quality Assurance

• Always verify table accessibility before recommending
• Distinguish between subscribed data vs. public data vs. user's own data
• Note any data limitations or caveats that might affect the analysis
• If no suitable tables are found, suggest alternative approaches or data sources

Important Considerations

• Respect data access permissions - only recommend tables the user can actually query
• Consider data costs when recommending large datasets
• Prefer tables with clear documentation and known data quality
• When multiple similar tables exist, explain the differences to help the user choose