From the Usability Testing Document: https://docs.google.com/document/d/1Gf26KoEob7ypjofzuHcFQLYhhbGYXv59DOmUVGvgttE/edit?tab=t.0
Where: /recommend vs /performance — how users adjust workload settings
Problem: The two pages use completely different patterns for editing workload parameters:
• Recommend — params (ISL/OSL/TTFT) are near the top, visible via "Adjust?" in the info strip. Flat grid layout.
• Performance — params are at the bottom under "Want to change assumptions?" in 5 nested accordion sections (Workload, Precision & memory, Hardware, Parallelism, vLLM config). Accordion rows show inline summaries when collapsed.
A user who learns one page has to relearn the other. The interaction model for adjusting parameters should feel consistent across both pages, even if the available parameters differ.
Recommendation: Converge on a shared pattern for parameter editing. Both pages should use the same component, layout location, and interaction model — even if Performance exposes more sections than Recommend.
Severity: Medium — inconsistency increases learning cost and creates the impression of two unrelated tools rather than one product.
From the Usability Testing Document: https://docs.google.com/document/d/1Gf26KoEob7ypjofzuHcFQLYhhbGYXv59DOmUVGvgttE/edit?tab=t.0
Where: /recommend vs /performance — how users adjust workload settings
Problem: The two pages use completely different patterns for editing workload parameters:
• Recommend — params (ISL/OSL/TTFT) are near the top, visible via "Adjust?" in the info strip. Flat grid layout.
• Performance — params are at the bottom under "Want to change assumptions?" in 5 nested accordion sections (Workload, Precision & memory, Hardware, Parallelism, vLLM config). Accordion rows show inline summaries when collapsed.
A user who learns one page has to relearn the other. The interaction model for adjusting parameters should feel consistent across both pages, even if the available parameters differ.
Recommendation: Converge on a shared pattern for parameter editing. Both pages should use the same component, layout location, and interaction model — even if Performance exposes more sections than Recommend.
Severity: Medium — inconsistency increases learning cost and creates the impression of two unrelated tools rather than one product.