Super Mario 64's Blurry Plaque Gets Its Closest Read Yet

A programmer working as Mochi64 has put forward "Eternal Star" on the top line and "We hope so" on the bottom as the closest computational match for the illegible plaque in the castle courtyard in Super Mario 64, according to Engadget.

The plaque is a small, blurry texture that never sharpens, even up close. Players have debated its text for over 30 years.

Mochi64 tested it with open-source analysis software, without using graphical upscaling. The software compared millions of word-and-font combinations against the plaque's pixels and scored each one for fit.

That search returned "Eternal Star" and "We hope so" as the best fit. The reading is offered as the closest match, not a final decipherment. That distinction matters.

The uncertainty kept alive a long-running rumor that the plaque reads "L is Real 2401," taken as a hint that Luigi is hidden in the game. IGN offered a $100 bounty to anyone who could work out how to unlock Luigi in Super Mario 64.

The competing readings were already circulating years earlier. In August 2019, Nintendo Life reported the text was debated as reading either "Eternal Star" or "L Is Real 2401".

The broader context here is the difference between enhancement and hypothesis testing. Upscalers, which sharpen images using patterns learned from other images, fill in missing detail and can invent legible text that was never in the source. Direct comparison works the other way. It creates candidate letters, shrinks them down through the same kind of blur and compression, then measures the pixel-by-pixel difference.

In my view, that choice fits the problem. Low-resolution text loses information in a way that cannot be undone. Once letters are turned into pixels, filtered and compressed, several different strings can leave behind very similar blocks. Searching a limited set of words and fonts does not recover the original. It ranks what is most likely, which is closer to decoding a noisy signal than to ordinary text recognition.

Looking at what this means for preservation and reverse engineering work, the method is more portable than the result. Community analysts often deal with baked-in textures, pre-rendered signs and compressed menu graphics where the original art is gone. A search that renders fonts explicitly keeps its assumptions visible in glyph shapes, layout, sampling and error scores. Its failure modes are also clear. A wrong font assumption or an incomplete word list produces a wrong winner.

The broader context here also includes why "L is Real 2401" lasted so long. Blurry pixels invite pareidolia, the tendency to see a familiar pattern that is not really there, especially when the reward would be a hidden character. A bounty focuses attention. It does not add resolution. The new result does not settle the case by proof. It shifts the weight of evidence toward a plain dedication-style inscription and away from a secret message.

In my view, that is a useful outcome even without certainty. It gives modders, archivists and curious players a baseline that can be tested. Other strings can now be rendered and scored against the same target, rather than argued over from screenshots. The long arc of this kind of tooling points toward fewer permanent mysteries and more reproducible guesses that can be updated.