Tableau Output Contract

TableauOutputs.from(DataAiResult) returns a TableauOutputBundle. The adapter does not write these DataFrames. Persist only the outputs needed by the customer and use governed table/view names if they differ from the defaults.

Standard outputs

Bundle method	Default table constant	Grain	Tableau use
`cleanRows()`	`dataai_clean_rows`	One accepted source record	Governed analysis and record drill-through
`rejectedRows()`	`dataai_rejected_rows`	One rejected source record	Remediation workflow
`qualityFindings()`	`dataai_quality_findings`	One rule finding	Severity/rule/field analysis
`fieldProfiles()`	`dataai_field_profiles`	One field profile for the current run	Completeness and distribution analysis
`pipelineRuns()`	`dataai_pipeline_runs`	One pipeline execution	Audit and run history
`dashboardMetrics()`	`dataai_dashboard_metrics`	One pipeline execution	Fast Tableau overview

The default names are constants in TableauOutputNames. They do not force a catalog, schema, storage format, or write mode.

`dataai_dashboard_metrics`

Field	Spark type	Nullable	Meaning
`run_id`	string	No	DataAI pipeline run identifier
`completed_at`	timestamp	No	Pipeline completion time
`status`	string	No	Pipeline status
`rows_read`	long	No	Input records evaluated
`rows_accepted`	long	No	Records without findings
`rows_rejected`	long	No	Records with at least one finding
`quality_score`	double	No	Accepted records / read records * 100
`critical_findings`	long	No	CRITICAL finding count
`error_findings`	long	No	ERROR finding count
`warning_findings`	long	No	WARNING finding count
`info_findings`	long	No	INFO finding count
`fields_profiled`	long	No	Profile rows produced
`total_null_values`	long	No	Sum of profile `null_count` values

`dataai_pipeline_runs`

Field	Spark type	Nullable
`run_id`	string	No
`started_at`	timestamp	No
`completed_at`	timestamp	No
`status`	string	No
`rows_read`	long	No
`rows_accepted`	long	No
`rows_rejected`	long	No
`quality_score`	double	No
`library_version`	string	No

`dataai_quality_findings`

The base DataAI findings schema is:

record_key, rule_id, field_name, severity, finding_code, message, original_value, normalized_value.

For multi-run Tableau history, add the current run_id before appending:

Dataset<Row> findingsForHistory = tableau.qualityFindings()
        .withColumn("run_id", lit(result.summary().runId()));

`dataai_field_profiles`

The base profile schema is:

field_name, source_type, record_count, null_count, distinct_count, minimum_value, maximum_value, mean_value, standard_deviation.

For multi-run Tableau history, add run_id in the same way before append.

Recommended Tableau model

Use dataai_dashboard_metrics alone for the overview Accelerator.
Relate dataai_pipeline_runs (one) to dataai_quality_findings (many) by run_id for finding details.
Relate dataai_pipeline_runs (one) to dataai_field_profiles (many) by run_id for completeness details.
Relate findings to clean/rejected source rows by record_key only when the customer authorizes record-level BI access.
Keep row-level security and data-source credentials in Tableau or the underlying platform; DataAI does not manage Tableau identities.