modellerUpdated 2026-05-02

Table Auto-Analysis

What this covers

Table auto-analysis runs a heuristic engine against a source table's column names and data types to suggest a table type (fact or dimension) and a role for each column (measure, dimension, date_key, or ignore). This page explains how the heuristics work, what confidence levels mean, and how to use the suggestions.

When to use auto-analysis

Auto-analysis is useful when:

Auto-analysis does not modify the model. It returns suggestions only. You still define each dimension and measure manually, using the suggestions as a guide.

How it works

The heuristic engine examines each column's name and data type:

Table type classification. A table is suggested as fact if it has numeric measure-like columns (names matching patterns like amount, quantity, count, revenue, total, price, value) or an auto-increment integer primary key alongside non-PK numeric columns. It is suggested as dimension if the majority of columns are string or small-integer types with code-like names.

Column role classification.

Suggested roleSignal
measureNumeric column with an aggregation-friendly name (e.g. amount, qty, revenue, cost, count)
dimensionString column, or integer column with categorical naming (e.g. type, status, code, category, group)
date_keyColumn name contains date, day, ts, timestamp, _at, _on and the data type is date, timestamp, or integer
ignoreAuto-increment primary keys, internal row-hash columns, ETL watermark columns

Confidence levels.

LevelMeaning
highMultiple strong signals align. Classification is almost always correct for tables conforming to standard naming conventions.
mediumSome signals present, some absent. Review before accepting.
lowNaming is ambiguous or atypical. Manual inspection is required.

Running analysis

  1. Open the model canvas.
  2. Hover over a table node; click the auto-analysis icon (wand icon) in the table header toolbar.
  3. The Auto-Analysis dialog opens.
  4. Click Run Analysis. The engine runs synchronously; typical analysis takes under a second.
  5. Review the suggested table type, confidence level, and reasoning text.
  6. Review the column suggestions table.
  7. Close the dialog and manually create dimensions and measures using the suggestions as a guide.

To refresh the analysis after adding columns or changing the connection, click Re-run in the dialog.

Interpreting results

The reasoning text explains which patterns drove the table type classification. For example: _"Table has 4 columns matching measure name patterns (amount, qty, price, total) and an integer primary key. Classified as fact."_

The potential calendar column field, when present, identifies a date/timestamp column that could be used as the calendar anchor for time-variant measures. This is a suggestion only; it does not bind the calendar automatically.

Limits

Related