Unexpected LOD and running-total results

Understand the likely causes

• A table-wide FIXED denominator can precede an ordinary dimension filter, while context changes that population earlier. A dimension filter can remove prior marks before a table calculation, unlike a post-calculation display restriction.

Write the intended population

Define numerator, denominator, requested LOD and visible mark detail before changing the formula. North 30 is 50% of all 60 and 100% of North 30. January 10, February 30 and March 20 produce cumulative 10,40,60 when all contribute. These expected values answer distinct questions. Do not pick the result that looks familiar and then move filters until it appears. Preserve the exact source fixture, current selection and worksheet identity. The same calculation can be correct under one population and misleading under another, so metric names and supporting context must expose which question the workbook is answering.

Map grain and filter stage

List source/extract restrictions, context, ordinary dimension and later stages. For the simple documented pattern, FIXED is not the same as ignores every filter: earlier restrictions still matter. INCLUDE and EXCLUDE also have different detail/filter relationships. Identify whether the suspect sum explicitly aggregates displayed replicas, as in a table calculation across Product marks or an external sum. That operation can multiply a regional total. It is distinct from native coarser-LOD replication, for which a shelf aggregation may be ignored. Verify view grain, calculation scope and total settings rather than assuming a SUM-to-MIN change fixes the result. The record below describes expected scope decisions only. No calculation editor or worksheet is opened, and it must not be read as a universal pipeline simulator for every connector and feature combination.

North numerator30 / FIXED all60 ->50%
North context population30 / FIXED North30 ->100%
all months before running sum ->March60
March-only raw marks before running sum ->March20

Inspect addressing and partitioning

For a table calculation, record the virtual table, addressing dimensions, partition dimensions, sort and restart behavior. Adding Region to Color can change detail and therefore traversal even if RUNNING_SUM text stays identical. Decide whether a last-month display retains earlier history or recomputes over a selected window. A table-calculation filter and an upstream dimension filter have different consequences. Also inspect missing marks and first-period nulls; generating a date domain does not establish actual zero sales. Repair the smallest context mismatch rather than changing source joins or replacing nulls indiscriminately.

Verify boundaries and labels

Test ordinary/context North, added product detail, two running-total partitions, reversed sort and a last-month-only display. Compare expected contributing facts and marks, not only the final rounded percentage. Explain scope to readers through labels and reset behavior. This guide executes no Tableau calculation, filter or UI action. A future authorized review must capture actual settings and results and check dependent sheets. Local reasoning can identify the necessary correction but cannot establish engine behavior, connector translation or security enforcement; a hidden mark or filter is not an authorization boundary.

Keep it from recurring

• Keep ordinary/context and upstream/post-calculation fixtures with LOD and window measures. Document outer aggregation, sort and restart settings, and revalidate after dimensions or action filters change.