On this page
A responsive gallery can still omit the record
Hypothetical situation: a service coordinator searches a canvas app for a known customer and receives no results. A maker tries a familiar customer and sees the expected row immediately. Both tests can be accurate descriptions of what appeared on screen. Neither establishes that the search works over the complete source.
Treat completeness as a requirement to test explicitly. Build an expected set of record identifiers with the business user, using approved test data. Include records beyond the part of the source that a small local sample would cover. The expected set must come from an authorised source query or controlled fixture, not from the gallery being tested.
The decision is whether users can rely on this particular search expression. It is not whether the application feels fast or whether the maker has seen one successful match. Keep the search term, expected records and observed records together so another reviewer can repeat the check.
Keep the original requirement visible when someone proposes changing the search to make a test pass. Prefix matching and searching anywhere in a name solve different user needs. A technically supported expression can still be the wrong product behaviour. Ask the coordinator to review representative results and record any accepted change in meaning before updating the app and repeating the completeness check.
Understand the documented query boundary
Microsoft’s January 2026 delegation overview explains that a nondelegable query retrieves the first 500 records by default and evaluates locally; the configurable limit can be increased to 2,000. Raising that limit is not a general completeness fix for a larger source. The relevant question is whether the expression can be processed by the selected connector.
The SharePoint connector documentation at its September 2025 revision lists StartsWith as delegable for text, while excluding StartsWith on Choice and Lookup subfields. That is not permission to assume every formula containing StartsWith delegates. Inspect the complete expression and the actual column type.
This article intentionally does not supply a universal search formula. An operation suitable for one connector or field may not be suitable for another. Record your connector and field definitions before choosing the expression. Keep the dated documentation alongside the test so the reviewer can see the scope of the supported behaviour.
Use deliberate test records
Create test names that distinguish prefix matching from finding text anywhere in a value. A result containing the search term in the middle should not be expected from a prefix-only requirement. Check empty input, a known match, several matches and a known non-match. Record the intended behaviour before editing the formula.
Then run the search against a controlled source large enough to expose a local-query boundary. Keep the same test identities and permissions when comparing the source and application results. A row hidden by source permissions is a different issue from a row missed by an unsuitable query expression.
Microsoft’s dated overview recommends setting the data-row limit to one during scaling tests to expose nondelegable behaviour. Use that technique in a test copy, restore the agreed setting afterwards and repeat the realistic-volume check. Inspect delegation warnings as well; the absence of a warning is not a substitute for the expected-record test.
| Case | Expected evidence | Question for the reviewer |
|---|---|---|
| Known record outside the local sample | Record identifier appears | Was the source queried as intended? |
| Term at the start versus middle | Results match the agreed semantics | Does the user expect prefix or substring search? |
| Restricted record | No disclosure beyond authorised access | Are source and app tests using equivalent identities? |
| Low-limit test copy | Nondelegable expressions become visible | Was the normal setting restored and retested? |
Make completeness part of the handover
Keep a compact set of expected-record tests with the app. Repeat them when a column type, connector, filter or sort expression changes. Include the business user who can recognise a missing operational record. A maker may understand the formula while the coordinator understands which missing case would delay work.
If the requirement cannot be met with the selected source and supported query, document that limitation and revisit the design. Do not present an increased local limit as proof that the application will scale indefinitely. The receiving team needs to know the tested dataset and any remaining boundary.
For a practical lab, give two participants the same search term and different controlled datasets. Ask them to explain why a small successful test does not establish completeness for the larger one. Finish with the expected identifiers and the test conditions written down. That is a more useful acceptance record than a screenshot showing that a gallery displayed some rows.
Sources & further reading (2)
- Microsoft: delegation overview, revision 13 January 2026
- Microsoft: SharePoint delegation support, revision 22 September 2025
Historical evidence checked . Review practices and hypothetical exercises are editorial recommendations, not product guarantees.
