Question numbers match their folders: folder 01 contains 01-01, 01-02 and so on. The starter dashboard uses 01-05, 01-08 and 01-02; process filter values come from 01-20. OBP identities and SQL filenames remain stable. Screenshots show the earlier numbering; use the current question numbers in the text when navigating the updated bank.
Use this reference when a number looks unexpected or you need an adapted question. You need permission to view the relevant source question and database. The package uses Trino SQL over Next Plus data; available types and permissions must be checked in your own environment.
Follow the source chain
Source tables → a native SQL question in Question sources → a normal query-builder question in the English reporting examples collection → a dashboard card. The SQL source defines inclusion and calculation; the main question lets you filter its output; the dashboard applies the mappings you configured.
In the main question editor, Data shows the source question. In the question view, its link also appears below the title.
In the source question under Question sources, Open Editor opens the SQL. The example shows question 01-20. Reading it requires no conversion or saved change.
Open the main question and note its number.
Open Question sources and find the source with the same number and name.
Use Open Editor to inspect SQL and the selected database. Ask the administrator if access is missing.
For changes, create a new question in an authorized collection. Do not overwrite a shared source used by other cards.
Understand the SQL
WHERE: selects source records before aggregation. A result filter does not override these conditions.
GROUP BY: defines result grain. Adding SKU to a process summary changes each row to process × SKU.
COUNT versus SUM: records and quantities are different. COUNT(*) includes a record even when a quantity is NULL; SUM ignores NULL values.
CASE and COALESCE: category precedence and missing-value fallbacks can affect comparisons. A fallback label does not validate the underlying record.
Approximate medians and P90: check measured populations, dates and sample sizes. They are not specification or control limits.
Time units: timelogreport.duration is milliseconds; workflowsessionitem.nettimespent is seconds. Output hours divide appropriately. Never infer attendance from overlapping reports.
Date fields: creation, release, work start, due date and completion answer different questions. Confirm plant timezone before shift or daily analysis.
Common comparison checks
01-02 versus 01-05: frozen-order inclusion.
02-01 versus 02-02: missing status classified by CASE versus excluded by NOT IN.
02-13 versus 02-05/02-07: finished status alone versus required dates.
01-16 versus 01-10: missing-date orders remain in detail but not the measurable denominator.
02-08 versus 06-05: date-valid elapsed population versus all finished runs with net-time coverage.
Before joining or adapting
Decide what one row represents. Check identifier type, NULLs, soft deletion, uniqueness and actual relationship multiplicity before JOIN. One order joined to many reports can repeat its quantity and inflate a total. Use Use the table and relationship map for documented candidates, not assumed foreign-key guarantees.
Read or copy the source SQL
SQL definitions were inspected or authored on 8 November 2026. Replace the documentation marker {{schema}} with your authorized schema and run one SELECT at a time. The marker is not an installed Metabase parameter. Saved questions may have changed since that snapshot; compare the current editor when troubleshooting.