My Pinned-Card Check Stopped Before the Phone Screen

The check proved a server order

I checked which card Curio's server sent first, not which card the phone finally drew first.

Curio shows short knowledge cards one at a time. The feature was meant to choose one card and place it first. When the phone asked for cards, the server sent back a list with the chosen card at the top.

That result was real but narrow. The check read the list sent by the server. It stopped before the phone turned each item into an app card, ranked the pool, and drew the feed.

The product claim reached farther than the test.

Two later steps put the screen result at risk

The phone first turns each item from the server into the app's own card shape. That step dropped the label saying which card was featured. Dropping the label did not move the card by itself. It meant later code could no longer identify the chosen card.

Separately, the phone ranked the pool by chosen interests, follows, past actions, preferred card length, and card quality. It then spread repeated topics apart. That step did not reserve first place for the featured card, so it could move the chosen card before the feed was drawn.

The build record verifies both risks, but it does not prove that either one changed the first card on a live screen. It contains no screen capture or replay of that state. The reader-visible result remains unverified.

This distinction matters because “first” is not just a database field. It is a place in a person's field of view.

First position can change what people choose

Thorsten Joachims and coauthors tested position on Google search pages. In one condition, the top two results secretly changed places. Separate human ratings were used to judge which result was more relevant.

The first link still drew a strong click bias even when the second result was judged more relevant. The authors called this a trust bias: people appeared to treat the higher rank as a clue that the search engine favored that result.

The study was small and used student searches on Google, not card feeds or Curio. It supports one general human insight. Where an option appears can shape the response it receives, apart from the option's judged value.

That makes a pinned-first claim behavioral as well as technical. The position may change which card gets a chance to be opened.

Each consequence needs proof at its own layer

Playwright's best-practices guide recommends testing behavior a user can see. For Curio, a screen-level check would pass the chosen card through the same mapping and sort, open the phone feed, and inspect which card appears first.

That check could prove position on the screen. It still could not prove that first position changed attention or opens. That larger claim would need reader behavior, just as the search study measured clicks rather than server order.

The layers now have separate jobs. A server test proves server order. A screen test proves visible position. Reader data tests the human consequence. The proof must reach the last layer named by the claim.