Row context and filter context

A DAX expression never runs "in a vacuum": the result depends on the context in which it is evaluated. There are two kinds, and mixing them up is the source of most measures that "don't add up".

The filter context is the set of active filters over the tables of the model: rows and columns of the visual, slicers, page filters and the filters propagated through relationships. The measure is evaluated once per cell of the visual, each one with its own filter.

The row context is the "current row" of a table. It exists in calculated columns and inside iterators such as SUMX. It does not filter anything: it only says which row to read the values from.

-- Calculated column in Sales: there is a row context
Line Cost = Sales[Quantity] * RELATED ( Product[Cost] )

-- Measure: the filter context comes from the visual
Revenue = SUM ( Sales[Amount] )

RELATED reads, in the current row, a value from the table on the one side of the relationship (here, the cost of the product of that sale). In a visual with Product[Category] on the rows, the Revenue measure for each row is the sum of only the Sales rows whose product belongs to that category. The total of the visual has no category filter and sums everything.

Pitfall

Using a raw column inside a measure, outside an iterator:

Double Revenue = Sales[Amount] * 2

Measures have no current row, so this expression raises an error: DAX does not know which of the many rows of Sales[Amount] to use. Write SUM ( Sales[Amount] ) * 2 or, if the calculation has to be done row by row, SUMX ( Sales, Sales[Amount] * 2 ). When in doubt, ask yourself: "is there a current row here?". If not, the column needs an aggregation.