Live Stress Testing

Artwork is an image of Billie Eilish for “A Generation-Defining Collaboration” by Re-Edition Magazine

Algorithm stress testing is a form of black-box auditing (Stanford University HCI) used to study how digital systems react to provocative, sensitive, or nonconforming content. Rather than taking a platform at face value, the method enters diverse inputs and patterns of use to observe what the system boosts or buries. In platform investigations, this usually means interacting with ranking and recommendation systems in an abstract way so their habits begin to show. Researchers have long used strange queries, simulated users, comparative testing, and other audit methods to infer how digital systems sort and distribute information.

I learned the logic of this approach at The New School, where I was introduced to stress testing and interface analysis, and then continued studying it through independent research of my own. What interested me was not only the academic version of the practice, but what might happen if it were applied from inside a live platform environment with a real organic following using posting itself as the probe. Instagram was the obvious site because it is not just a place where images and captions sit beside each other. It is a major visibility engine that ranks, sorts, and distributes, then leaves users to mistake the outcome for truth.

My hypothesis was that certain forms of critical content, especially criticism directed at Instagram itself or the growing authority of AI, would run into subtle forms of suppression that might be hard to prove one post at a time but easier to recognize across a longer stretch of time. I identify as a conceptual artist (like my gal Sophie Calle) so I suspected that deeply abstract, dense, or weird material would be treated differently from content that was instantly classifiable and easy to digest.

Platforms tend to reward material that arrives already simplified, so I wanted to know what happened when it did not. I also learned something much darker in the process, which was that the platform responded far more favorably to provocative images of me, including visible breasts and sultry facial expressions. That made something embarrassingly clear, which is that Instagram rewards sex appeal, which makes the whole environment feel less like a creative venue and more like a sexualized attention economy where desirability travels faster than critique and the body is treated as a more valuable currency than thought.

To test that, I used my own account as a case study. I created digitally collaged posts organized around larger ideas, tensions, and comedic critiques, including themes I believed the platform would be less eager to reward, and then I watched what happened to their circulation, reception, and interpretive afterlife. My apologies, but this was not random digital activity dressed up as theory afterward like some might suspect, as the content evolution was the sport and the crazy amount of pressure was born from curiosity that was bound to confuse.

What emerged was not only a vivid picture of algorithmic behavior, but also the social world that forms around it. A platform like Instagram shapes the terms under which content gets read. What circulates begins to look valid. What stalls begins to look suspicious, and what disappears looks like personal failure, instability, irrelevance, or poor judgment, as though the platform had nothing to do with it and everyone had somehow reached the same tiny wrong conclusion all by themselves. This is one of the system’s more sleazy moves. It governs visibility and then lets the crowd deliver the sentence.

That is why my following’s response was not separate from the test. Projection (more than anything) and confusion, selective reinforcement, and silence all became evidence of how platforms determine interpretation. My followers often misread my digital activity, as I suspected they would, and that misreading became one of the most important findings. The project did not only test what the algorithm would do with difficult material. It also tested what people would do with it once the platform had already thinned out the context around it. People are remarkably sure of themselves when very little information is available. This is not always their most lovable trait, and in my case, this characteristic of being human included mild cancellation.

Using a personal account made the project much riskier, but also way more revealing. A personal account carries a rich history, expectations, accumulated context, ready-made social framing, and, eventually, a soiled online identity. That meant I was not working in a sterile environment with burner profiles and an entirely experimental setup. I was working from inside a digital life with a heartbeat, where the outcome is messy and therefore much more useful. The findings were not limited to whether certain themes seemed to travel less effectively than others. They also included the speed with which problematic online cues harden into false narratives about a person’s motives and emotional state.

This is what algorithm stress testing can make visible, though I wouldn’t recommend it. The approach offers a disciplined way to investigate systems that exert enormous influence while revealing very little about how they function. In my own work, it meant using conceptual digital collage and excessive posting as a method for studying the boundaries of visibility, suppression, distortion, and platform power. The account became the live probe, the feed became the testing ground, and the reactions became part of the evidence. What looked from the outside like inconsistency was, from the inside, a way of questioning what happens when a human being pushes against a platform built to present a corrosive lie as reality.