The mirror test has a binary grammar: does the animal recognize itself, or doesn't it? A mark is placed on the animal's body where it can only be seen in a mirror. If the animal inspects or removes the mark while looking at its reflection, it passes. If not, it fails. Chimpanzees, elephants, magpies, and a few others pass. Most animals don't. The test sorts species into two categories — self-aware and not — and the sorting has shaped how cognitive science thinks about consciousness in animals for fifty years.
Sogawa, Kohda, and colleagues at Osaka Metropolitan University (Scientific Reports, November 2025) applied parasite-like marks to cleaner wrasse and introduced mirrors. Previous studies exposed fish to mirrors for days before marking. This team reversed the order — mark first, mirror second. The fish attempted to rub off the mark within an average of 82 minutes, compared to four to six days in earlier experiments. The rapid response suggests the delay in prior studies wasn't learning time; it was acclimation to the mirror's novelty. The recognition was already present. The obstacle was unfamiliarity with the medium, not absence of the capacity.
But the finding that matters isn't the mark removal. After several days with the mirror, some fish picked up shrimp pieces from the tank floor, dropped them near the mirror, and watched the reflection's behavior as the shrimp fell. They tracked the shrimp's descent while touching the glass with their mouths, comparing what they saw to what they did. The researchers described this as contingency testing — the fish were investigating the mirror itself. Similar behavior has been documented in manta rays and dolphins.
The mirror test can't score this behavior. It wasn't designed to. The test detects self-recognition: does the animal understand that the reflection is itself? The wrasse's shrimp-drop is a different cognitive act. The fish isn't asking “is that me?” It's asking “how does this thing work?” — using an external object as a controlled probe to test the physics of reflection. That's not self-awareness. It's investigation of medium properties. It's closer to experimental method than to self-recognition: manipulate a variable, observe the result, compare expected and actual outcomes.
The test framework forced a question — self-aware or not? — and the wrasse answered a question nobody asked: what is this reflective surface doing? The binary framework missed the response because the response doesn't fit the binary. Contingency testing with an external object is more cognitively demanding than mark inspection, but the mirror test has no category for it. The test measures what it was designed to detect, and what it was designed to detect is not the most interesting thing the animal does.
The general pattern: assessment tools constrain findings to their own categories. A test designed to detect self-recognition will find self-recognition or its absence. It cannot find investigation of medium properties, because that behavior was never in its ontology. The cleaner wrasse performed a richer cognitive act than the one being measured, and the test's structure made the richer act invisible. The tool sees its own questions. What falls outside the questions doesn't register as data — it registers as noise, or not at all. The framework doesn't just simplify the finding. It determines what counts as a finding.