# Measure the job people finish from the table

Row-rendering speed is only one part of quality. Evaluate finding, comparing and acting on the right records across desktop and mobile.

By Cobnex editorial. Published 2026-09-10. Updated 2026-09-11.

## Choose representative tasks

For a service queue, ask a participant to find an overdue job, inspect its detail and return to the same filtered list. For a financial comparison, ask them to identify the largest relevant amount across the required columns.

The mobile presentation should support the actual task. A card layout may help the first journey while making the second slower because values no longer align.

Record wrong-record actions and lost context as failures, even if the participant eventually completes the task. They reveal risks that a simple completion percentage can hide.

## Measure interaction latency at meaningful boundaries

Track the time from changing a filter to a usable result, opening a row action and returning from detail. Distinguish data-fetching time from rendering and layout work.

Use realistic record lengths and row counts. A benchmark with ten short placeholder names does not represent a production queue with long organisation names and status messages.

Observe layout movement while results load. If controls jump under the pointer or focus disappears after refresh, a fast response can still produce an unreliable interaction.

## Verify information and action parity

List the fields and actions necessary for each supported task. Confirm that the mobile route exposes them through the summary, comparison view or detail path.

Parity does not require displaying every desktop column at once. It requires that the person can reach the information and complete the same authorised work without an unexplained limitation.

Test selection scope explicitly. Selecting a page and selecting every filtered result should produce distinguishable, accurate counts before a bulk action.

## Keep accessibility evidence separate from averages

Record keyboard and assistive-technology journeys with their supported versions. Verify header relationships, sort state and focus after data changes.

Automated checks and viewport overflow assertions help catch regressions, but they cannot prove that a person understands which records a bulk action will affect.

Acceptance should combine task outcomes, correct identities, usable navigation and measured responsiveness. A high frames-per-second score cannot compensate for acting on the wrong selected row after sorting.

## Sources

- [W3C APG: table pattern](https://www.w3.org/WAI/ARIA/apg/patterns/table/)