Explore and Analyze Data

For: Analytics Readers and Writers

Use this page to discover available schemas, tables, views, and columns and to answer read-only questions in an Analytics Project. The complete workflow runs through your AI Agent and Culsma PAWS MCP.

No Analytics Web/App interface is available in the current pilot. The SQL, sources, row counts, truncation status, and audit ID are the evidence for the result.

Before you start

You need:

• the exact Analytics Project identified;
• lab.analytics.read or lab.analytics.write on the MCP connection;
• an active Reader or Writer grant for that Analytics Project; and
• a question with enough detail to define the population and measure.

Publisher access alone does not authorize read-only queries. Ask for a Reader or Writer grant if you also need to analyze the data.

Start in your AI Agent

For discovery:

Discover available analytics data

In Culsma PAWS Analytics Project [PROJECT NAME OR ID], list the schemas, tables, and views I can access. For each relevant table, show its columns and data types and explain the likely row meaning from available metadata. Do not change any data.

For a question:

Run an audited read-only query

In Analytics Project [PROJECT NAME OR ID], answer: [QUESTION]. Confirm the authenticated project, inspect the accessible tables and columns, clarify any definition that materially affects the result, dry-run the SQL, and execute it only in read mode with a conservative result limit. Return the exact SQL, source tables, assumptions, row count, truncation status, environment, and query audit ID.

The AI Agent should:

1. state the selected environment and authenticated Analytics Project ID;
2. discover only the schemas and tables available through your project role;
3. inspect table and column metadata before drafting SQL;
4. clarify ambiguous measures, dates, groupings, and exclusions;
5. dry-run every query;
6. execute only in read mode with a conservative limit; and
7. return the result and complete reproducibility evidence.

Read-only analytics queries do not need a write approval.

Review the evidence

Evidence	What to check
Environment and project	Correct selected environment and canonical Analytics Project ID.
Source objects	Exact schemas, tables, views, and columns used.
Metric	Counted entity, distinctness, aggregation, grouping, and exclusions.
Time logic	Date or timestamp column, time zone, range, and grouping interval.
SQL	One read-only query that matches the stated question.
Dry run	Query was checked before execution.
Result completeness	Returned row count, maximum-row limit, and truncation status.
Audit	Query audit ID is present beside the selected environment.
Interpretation	Observed results are separated from causal or scientific conclusions.

If a result is truncated, ask for a more focused aggregation or an explicitly controlled continuation. Do not describe a partial result as exhaustive.

Visualizations and follow-up questions

The AI Agent may create a chart from the returned result when a visual makes the relationship easier to understand. The chart is a presentation of the query result, not additional evidence.

For a follow-up question, preserve the prior SQL and assumptions, then state exactly what changes. A materially different population, metric, grouping, or source should produce a new dry run, query, and audit ID.

More questions to try

Count measurements by month and measurement name for 2026. State which timestamp defines the month and show zero-count groups when supported by the source model.

Which sample identifiers occur more than once in [SCHEMA.TABLE]? Define the duplicate rule and show the number of occurrences.

Compare the missing-value rate for each required measurement field. Show the numerator, denominator, and missingness rule.

Show the distribution of numeric results by unit. Stop and report inconsistent units rather than combining them.

If the query cannot continue

What you see	What to do
The Analytics Project is ambiguous	Resolve the canonical project before schema discovery or query execution.
Reader or Writer access is missing	Ask the Workspace Owner for the minimum appropriate grant.
A required table or column is unavailable	Confirm project ownership, role grants, and object names; do not substitute unrelated data.
A metric has several meanings	Choose the distinct key, date, population, and exclusion rule explicitly.
The dry run fails	Review the reported SQL or schema error and revise before execution.
The result is truncated	Aggregate further, narrow the question, or continue with an explicit limit.
Units or categories conflict	Separate incompatible values or define an approved normalization.
No audit ID is returned	Treat the answer as incomplete and request audited query evidence.
The answer implies causation	Reword it as an observed association unless supported scientific analysis exists.

You have completed the analysis when the question, SQL, sources, assumptions, result, completeness information, environment, and query audit ID form one reproducible answer.

To load or correct reusable data, continue to Maintain Analytics Data.