Lee Sun-woo

Review the transitions between screens

Pulse briefly told readers that there was nothing to show before it had even started loading. The completed screen looked correct. The problem existed in the short interval between opening the feed and starting the request.

That was one of three corrections I made in September 2026. Together, they changed what I ask to see when reviewing an interface built with AI: the sequence of states a person encounters, including a return to content that is already available.

The empty state appeared too early

Loading was set inside an asynchronous task. Before the task ran, the interface saw an empty list and a loading flag that was still off. It rendered an empty state from a combination that existed only because of the order of execution.

I moved the loading update to the start of the request, before the asynchronous work. An empty state should describe a successful response with no items. A request still in progress and a failed request need different explanations.

Conceptual states. Wait for the response before deciding that a result is empty.

Those distinctions are part of the interaction design. If the screen says “No results,” a person may change the query. If it says loading failed, retrying may be the appropriate next action. The message needs to correspond to what actually happened.

An available image looked as if it were loading again

The image cache held content that the view did not use immediately. When a card returned to the screen, it began without an image and checked the cache later. That produced another placeholder and fade for an image already available in memory.

I initialized the view with the cached image when one existed. The first load still needed a loading treatment; returning to available content did not need to replay it. This was a state-initialization correction, not a reason to make every animation faster.

The placeholder promised the wrong height

The skeleton’s text bars were based on font sizes rather than line heights, and its metadata area did not reserve the share action’s full height. The loaded card therefore occupied different space. I adjusted the placeholder slots to match the final row.

Samhita Tankala’s guide to skeleton screens describes placeholders as a preview of the page’s structure and cautions against flashing them on quick loads. In Pulse, that structural promise was the useful part to correct before polishing the shimmer.

These fixes suggest a compact review sequence: open with an empty cache, inspect a successful empty response and a failure, then scroll a loaded card out of view and back. At each step, ask what the screen says has happened. A state transition is successful when that explanation remains accurate through the change, not only at its endpoints.