MOUNTAIN VIEW, Calif. — Autocorrect engineers confirmed Tuesday that the software’s habit of replacing a common expletive with the name of a waterfowl was never a malfunction, but a deliberate, fifteen-year experiment to determine how much frustration a human thumb can absorb before the phone is thrown.

The admission came in a leaked internal memo titled “Re: Re: Re: Duck Study (Final) (Final v2).” The document states that the program, internally code-named Project Mallard, “has exceeded all expectations and several windows.”

“We never believed anyone typed that word by accident,” said Dr. Lenore Vashti-Plimsoll, principal investigator and Senior Director of Involuntary Substitution Outcomes at the Institute for Applied Keyboard Disappointment. “We wanted to know what happens when a person is trying to say something precise, emotional, and time-sensitive, and the phone insists on talking about ducks. The answer is that they type it again. Slower. With more force.”

According to the memo, researchers calibrated the experiment using a proprietary unit called the Thumb Rage Index, defined as the number of times a user retypes a word divided by the number of times they whisper “come on” to a device. Early trials produced an average of 4.2 retypes. By 2019, the figure had climbed to 11, with a small control group reaching what the memo calls “a spiritual place.”

The memo also acknowledges related studies, including the conversion of “I’m on my way” to “I’m on my wake,” the substitution of a coworker’s name with a nearby but unrelated fruit, and the phone’s long-standing refusal to learn the word the user has typed correctly nine hundred times.

“That last one is our flagship,” said Cormac Tilbrook-Ng, Lead Behavioral Auditor of Predictive Text Resentment. “We call it the Persistence Trial. The user teaches the phone a word. The phone nods. Eleven days later, it forgets. We are essentially running the longest-duration study of marital patience in the history of science, and none of the participants signed anything.”

Asked whether the experiment raised ethical concerns, Vashti-Plimsoll said the team had consulted an internal review board, which approved the work after one member, quote, “sent a text to his mother that went poorly.”

“We assumed people would eventually give up and switch to voice dictation. Instead they developed what we can only describe as a rich, regional vocabulary of workarounds, and that was not in the protocol.”

The company said it will conclude Project Mallard by the end of the fiscal year and has already begun designing its successor, which it described only as “the same idea, but with the word ‘the.’” Beta testers reported feeling “fine” before asking not to be contacted again.

At press time, the engineers had published a brief apology to users, which the software automatically changed to a gift certificate for a goose.