Skip to content

Query Results

Explainability

Tursio helps users understand how their question was processed and how the answer was generated. It adds valuable explainability to the results, so users know exactly what data was used and how the system interpreted the question.

Screenshot

Which Data Inputs Were Used in the Query?

This section lists the data elements involved in answering the question:

  • Dimensions: One or more key categories or attributes (e.g., city, product, date) that were part of the question.
  • Columns: All columns inferred from the query, including any that were ambiguous. When a column name might refer to multiple possible data fields, Tursio reports all the options it considered for clarity.

How Was the Data Filtered?

Details on how the question limited or refined the data, including:

  • All filters detected in the question (for example, date ranges or product categories).
  • If input values were close matches but not exact, nearby values considered by Tursio are also shown to help explain what was included.

How Was the Data Ordered?

Information about how the results were sorted before being presented:

  • What limit (e.g., top 10) was applied.
  • Which columns or measures the data was ordered by.
  • The sorting direction (ascending or descending).

Was the Data Aggregated?

This section explains how the data was grouped and summarized:

  • Grouping columns used to organize the data (e.g., grouping sales by month).
  • Details of measures calculated — which columns were aggregated and by which functions (SUM, AVG, COUNT, etc.).

What Kind of Analysis Was Performed?

Tursio supports several types of analysis that may be detected in the question, including but not limited to:

  • Trend analysis
  • Over-time patterns
  • Forecasting questions
  • Hourly analysis
  • Change patterns
  • Patient forecasts
  • Targets analysis
  • Data quality checks
  • Delinquency analysis
  • Weekly, monthly, yearly comparisons
  • Ordering and planning tasks

This helps users understand the analytical approach taken to answer the question.

Which Parts of the Question Were Relevant but Could Not Be Interpreted?

Sometimes, parts of a question may be relevant but not fully understood by Tursio. This could be due to ambiguity, unsupported terms, or missing context. These parts are highlighted so users know what to adjust.

Tips to Improve the Question

  • Ambiguous columns: If a column name matches multiple fields in the dataset, be more specific to clarify which data you mean.
  • Trend questions require a measure: When asking about trends or patterns over time, include a measure (like sales, count, or revenue). Without a measure, Tursio may not understand what needs to be analyzed.
  • Try different aggregate functions: If the question is not working as expected, experiment with common aggregate functions:
    • total (SUM)
    • average (AVG)
    • minimum (MIN)
    • maximum (MAX)
    • deviation of (STD)
    • variance of (VAR)
  • If results are unclear, consider rephrasing the question or adding more detail about what you want to analyze.

By reviewing the Query Awareness section, users can gain insight into how Tursio processed the question and learn how to adjust it for better answers.

Text Response

Tursio generates descriptive natural language responses to user questions. The type and length of the response depend on:

  • The type of question (e.g., listing, group analysis, comparative, trend).
  • The framing of the question (e.g., requesting a description vs. analysis vs. enumeration).

Users can refine the response by providing further specific instructions in the follow-up.

Tabular Result

Tabular results allow quick viewing and exploration of the returned data. Displaying query results in this format makes it easy to interact with, review, and act on data efficiently.

Screenshot

The table presents each entry as a row, with individual columns showing key attributes related to the query such as names, identifiers, contact information, or other relevant fields. A summary above the table shows high-level insights — counts, aggregate statistics, or recommended next steps — depending on the data and analysis context.

Key Table Features

  • Interactive Sorting: Click any column header to sort the data in ascending or descending order, helping you find the highest, lowest, or most recent values quickly.
  • Search Bar: Instantly filter results by typing keywords or phrases, narrowing the focus to records that matter most.
  • Column Filtering: Use the gear icon to select which columns to display and remove others from the table.
  • Pagination Controls: Navigate through large datasets with left/right arrows and an entries-per-page selector to adjust how much is shown at once.
  • Result Summary: Above the table, a summary may describe total counts, unique values, or recommended next steps for deeper analysis.
  • Export Functionality: Download the displayed results for offline analysis or reporting with a single click. Options include downloading the full table or a pruned (row- and column-filtered) view.

How to Use

  • Scan and analyze the table for patterns, insights, or specific data points.
  • Use search and sort to rapidly isolate entries of interest or check specific criteria.
  • The summary at the top may highlight opportunities for new insights, such as analyzing trends or performance metrics. This can be used to guide deeper analysis.
  • Download the dataset for further investigation.

Visualization

Tursio automatically generates charts based on available data columns. Depending on the data, one or more of the following chart types may be created.

Default Histogram

When grouped or stacked charts are not suitable, the system generates a default histogram. How it works:

  • Prefers a distinct histogram column, selecting the one with the highest cardinality.
  • If no distinct column is available, falls back to a non-distinct histogram column.
  • When using a non-distinct column, only measure columns starting with sum, min, or max are considered.
  • These measures are aggregated if a non-distinct histogram is used.
  • The chart is generated using the best available category column.

Stacked Histogram

A stacked histogram is generated when the required conditions are met. How it works:

  • Selects a measure column with "sum" to ensure valid aggregation.
  • Chooses a color column with the lowest distinct values for better readability.
  • If the chosen color column exceeds the maximum distinctness threshold, stacking is skipped.
  • Determines the category (X-axis) column by ranking candidates in ascending order of distinctness and descending order of cardinality.
  • Groups data by the selected X-axis and color columns, aggregating measure values.
  • If a valid distribution is found, the system generates a stacked histogram chart.

Grouped Bar Chart

A grouped bar chart is created when suitable histogram candidates are available. How it works:

  • If more than five measure columns are present, grouping is skipped and no chart is generated.
  • Prefers distinct histogram columns, selecting the one with the highest cardinality.
  • If no distinct histogram exists, falls back to a non-distinct histogram column.
  • When using a non-distinct column, only measure columns starting with sum, min, or max are considered.
  • Data is grouped and aggregated when a non-distinct column is used.
  • The chart is generated using the best available category column.

Pie Chart

A pie chart is generated when valid category and measure columns are available. How it works:

  • Created only if all values in the selected measure column are positive and the chosen category column contains no duplicate values.
  • If no valid candidates exist, no chart is generated.
  • Candidate columns are sorted by cardinality (descending), and the highest is selected.
  • If multiple measure columns exist, the first one with all positive values is chosen.
  • The chart is created using the top category–measure pair.

Time Series Chart

A time series chart is generated when a valid time-based column and one or more measure columns are available. How it works:

  • Created only if at least one column can serve as a clean, well-defined time axis (e.g., year, quarter, month, date, timestamp).
  • If multiple time columns exist, the cleanest and most granular one is chosen.
  • If no suitable time column exists, no chart is generated.
  • All valid measure columns may be plotted together as a composite set.
  • The chart uses the selected time column as the X-axis and the chosen measure(s) as the Y-axis.
  • If aggregation is required (e.g., grouping by year, quarter, month), the time column is aggregated automatically.

Data Model

The data model defines the structure of the data that Tursio queries, including the tables, columns, and relationships available to answer questions.

When users submit a query, Tursio translates it into a structured SQL statement based on this data model. The generated SQL specifies exactly which tables and columns are involved and what operations are performed. This section provides an overview of the SQL generated from the configured data model, allowing users to view and validate the queries that are executed.

Key Highlights

  • Database Name: The SQL query is displayed along with the corresponding database name for clarity and traceability.
  • Generated Query: Users can preview the exact SQL that will be executed, ensuring transparency between the data model design and the actual query.
  • Model Context: The SQL reflects the tables, joins, dimensions, measures, and filters defined in the data model.
  • Validation: This view helps validate whether the SQL aligns with the intended data logic and business rules.
  • Debugging Aid: In case of unexpected results, the SQL overview serves as a debugging tool to trace back query logic.

Use Cases

  • Reviewing the SQL to confirm correct column selection, join conditions, and aggregations.
  • Ensuring the query points to the correct database instance.
  • Sharing SQL with data teams for performance checks or optimization.
  • Troubleshooting mismatches between expected results and actual outputs.

Friendly Explanation

To help users better understand, Tursio explains the generated SQL in natural language, showing how the system interpreted the question into a database query.

Why the Data Model Matters

  • Clarity: Knowing which tables and columns are queried helps validate that the results are relevant and accurate.
  • Customization: Understanding the data model allows users to tailor questions more precisely and leverage advanced query features like joins, filters, and aggregations.

Query History

Query history allows users to quickly revisit and reuse previously asked questions, saving time and avoiding repetitive typing.

How It Works

  • On the Home Page search bar, click the History icon.
  • A list of recent questions appears in descending time order (latest first).
  • Each entry represents a previously asked question.

Screenshot

Reusing a Question

  • Browse the list and click the question you want to reuse.
  • The selected question automatically appears in the search bar.
  • Press Search to run the question again.

Benefits

  • Easily re-ask frequent queries without retyping.
  • Maintain context of what you have already explored.
  • Quickly compare results for recurring questions.