7.1 The Tableau Order of Operations: From Extract to Display Filters

7.1 Tableau Order of Operations & Filter Scope

Filters that appear together on a shelf do not necessarily evaluate together. Tableau's order of operations explains why a Top N list can return fewer members than expected, why a FIXED expression can ignore a normal dimension filter, and why filtering to one month can remove the prior month needed by LOOKUP. Use the ordering to express the requested population, not as a promise about a database's exact physical implementation.

Core filter sequence

For ordinary view filters, the stable high-level sequence is:

1. Extract filters restrict rows stored when an extract is created or refreshed.
2. Data source filters restrict data available through that Tableau data source.
3. Context filters establish the population for dependent filters and calculations on a worksheet.
4. Dimension filters retain dimension members; Top and Condition rules have special placement relative to normal member selection.
5. Measure filters evaluate aggregated measures for marks at the view's level of detail.

The full Tableau order diagram also locates sets, Top N and conditional filters, FIXED LOD expressions, INCLUDE and EXCLUDE expressions, blending, forecasting, table calculations, trend lines, and reference lines. Rather than memorizing one numbered list and applying it blindly, learn the dependencies that solve common scenarios.

Extract and data source filters

An extract filter determines which rows enter a .hyper extract. Excluded rows are unavailable until the extract definition changes and the extract is refreshed. A data source filter applies to worksheets using that data source, for either live or extract connections. Both occur before worksheet context, dimension, and measure filters.

These early filters reduce the population available to later calculations. Use them for a genuine source-wide requirement. Do not use a narrow extract filter if later authors will need the excluded history.

Context filters

Add a dimension filter to context when another operation must compute within that filter's retained population. A context pill is gray. Tableau may issue different queries or materialize intermediate results depending on the connector and plan, so context is not guaranteed to improve performance. Its reliable purpose is semantic: later dependent operations see only the context population.

Example: the requirement is Top 10 Customers by SUM(Sales) within the selected Region. If Region is a normal dimension filter, the Top rule can identify the enterprise-wide Top 10 first and the Region selection can then remove some, leaving fewer than ten. Add Region to context so the Top rule evaluates among customers in that region.

Avoid unnecessary context filters. Each change can require dependent results to be recomputed, and a context that barely reduces data may add overhead. Test representative interactions.

FIXED versus dimension filters

A FIXED LOD expression evaluates after context filters but before ordinary dimension filters. Therefore:

{ FIXED [Customer ID] : MIN([Order Date]) }

retains a customer's all-time first date when Year is a normal dimension filter. Add Year to context only if the requirement truly is first date within the selected years. Doing so can redefine historical cohorts, so label that result honestly.

INCLUDE and EXCLUDE LOD expressions respond to ordinary dimension filters because they are evaluated with the view's dimensional population later than FIXED. Measure filters occur after those LOD results are formed. A filter's type, not merely the field's original data type, determines its behavior.

Dimension-filter modes

A discrete dimension filter offers General member selection, Wildcard matching for text, Condition rules, and Top rules. Top and Bottom specify a count and an aggregation, such as Top 10 Customer Name by SUM(Sales). Bottom is an explicit option; do not reverse a measure unless the requested logic actually requires it.

A continuous dimension range retains values in an interval. When a numeric measure is filtered as All Values, Tableau treats the row-level values differently from filtering an aggregation such as SUM. An aggregated measure filter acts on the marks after aggregation.

Table calculations and display filters

Table calculations operate after the view's query results have been aggregated. A normal date dimension filter can remove November before a December LOOKUP with offset -1 runs, producing null. To preserve November for calculation but show only December, create a table-calculation-based display filter such as LAST() = 0 on a chronologically ascending partition or LOOKUP(MIN([Order Date]), 0) and select the desired date.

Because the display filter is itself a table calculation, the comparison calculation can evaluate over the full partition before unwanted marks are hidden. Verify Compute Using for both fields. A late filter can preserve the denominator of percent of total, but it is not a security control because the records remained in the table-calculation population.

Apply a filter to worksheets

The Apply to Worksheets menu controls reach independently of evaluation order:

• Selected Worksheets targets an explicit list.
• All Using This Data Source targets sheets that use the same source.
• All Using Related Data Sources can propagate through recognized related fields in other sources.

Confirm the field relationship and each selected sheet. A visible dashboard filter card does not automatically control every chart on the dashboard. Conversely, a filter can affect a sheet even when its card is hidden.

Forecasts and analytical overlays

Forecasting, trend lines, reference lines, totals, and table calculations have their own positions in Tableau's full pipeline and can respond differently to late filters. The practical test is to identify which marks each feature receives. A trend line built after data is filtered models the retained marks. A reference line scoped to a pane uses the marks in that pane. A forecast requires supported date or ordered fields, measures, and sufficient data.

Troubleshooting checklist

1. Identify whether the filter is extract, data source, context, dimension, measure, or table-calculation based.
2. State the population that a calculation should see.
3. Move a filter to context only when a dependent Top, condition, set, or FIXED operation must see that population.
4. Use a late table-calculation filter when marks must remain for LOOKUP, running totals, or percent-of-total denominators.
5. Verify scope across worksheets separately from evaluation order.
6. Test boundary cases, including nulls, fewer than N members, first periods, and empty selections.

Most order-of-operations problems are population problems: an early filter removed too much, or a later filter failed to constrain an earlier calculation. Describe the desired population first, then select the filter type and scope.