UX vs. User Results
The original version of this app had no CSS. Raw HTML, dark mode enforced by a browser plugin. No animations. No encouragement. No streak counter. No owl. At the end of every session, the screen displayed a picture of Rocky Balboa getting punched in the face — bleeding, absorbing the hit — while the hold music from a Vanguard customer service line played in the background.
The message was deliberate: you are making an investment, and it is supposed to hurt.
That was not a joke. It was the honest statement of a design philosophy that the rest of the EdTech industry has spent fifteen years actively contradicting.
What the Industry Optimizes For
Duolingo is a magnificent piece of software. The engineering is serious, the gamification is sophisticated, and the retention numbers are extraordinary. Tens of millions of people open it every day.
Almost none of them reach fluency.
This is not a coincidence. It is the direct consequence of what the product is optimized for. Duolingo's business model requires daily active users. Daily active users require that the app feels good to open. An app that feels good to open is an app that rewards you for showing up, keeps difficulty just below frustration, and celebrates small wins with animation and sound. The product experience is the product. The language learning is the justification.
Streaks. Hearts. Confetti. Encouragement after wrong answers. Difficulty tuned to keep you comfortable and coming back.
A diff. Character by character. What you typed versus what was said. No consolation. Next sentence.
These are not two points on the same spectrum. They are opposite design goals. An app that makes you feel good about the gap between where you are and where you need to be is an app that has decided your comfort matters more than your progress.
Stupid, Sad, and Lonely
"You should feel stupid, sad, and lonely. That's how you know you're making progress."
— Notes, written during the learning period, 2021–2022
This is not motivational rhetoric. It is a description of what serious skill acquisition feels like from the inside, and a diagnostic tool.
Stupid — because the gap between native speed and your current ability is made visible on every attempt. The diff does not soften this. You typed something wrong, and the exact shape of your error is displayed in red. If you do not feel slightly humiliated, the material is too easy.
Sad — because progress is slow and the corpus is large. In March 2021, the cold-start success rate was 39%. More than half of every new sentence was wrong on the first attempt. There is no reframing that makes that feel good. There is only the next session.
Lonely — because this method does not have a community, a leaderboard, or social features. There is no one to compare yourself to. There is no external validation. There is the audio, the transcript, and the diff. That is the work.
Comfort is the signal that you are not being challenged.
Challenge is the only thing that moves the number.
The Rocky Principle
Rocky does not win the fight in the montage. He wins it because of what the montage cost him. The montage is not the point — the accumulation of effort that the montage represents is the point. You do not watch Rocky getting punched and think the punches were bad. You understand that absorbing the punches is what makes him capable of landing them.
The hold music is the other half of the message. Vanguard does not apologize for making you wait. They put you on hold because the investment process takes time and cannot be rushed without destroying the outcome. The music is not there to entertain you. It is there to mark the passage of time that the investment requires.
Language acquisition at the level of C1 fluency is not a product feature. It is the result of a sustained process that feels, from the inside, like absorbing punches while waiting on hold. The data from 14 months of Romanian practice confirms this. The cold-start rate went from 39% to 74%. That journey did not feel like confetti.
The Discipline Parallel
People who fast, climb mountains, and do cold plunges already understand this tradeoff intuitively. They are not confused about why the thing that is hard produces better results than the thing that is comfortable. They have made their peace with the discomfort as the mechanism, not the obstacle.
Language learning has the same structure. The discomfort of failing on new audio — repeatedly, visibly, with no encouragement — is not a flaw in the method. It is the method. Removing it would remove the result.
"Fluent or fail? If you're not fluent, then you need feedback. 'My success rate is low, what can I do to improve it?' More transactions — at a low success rate."
"There is no in-between: fluent or fail."
— Notes.21, entry 459 and annotation, 2021–2022
More transactions at a low success rate. Not fewer transactions because the low rate feels bad. The low rate is the information. Act on it by continuing.
What This Means for the App
SiteDictation has a dark theme now, and the typography is clean. Rocky no longer appears at the end of a session. These are concessions to legibility, not to comfort — the underlying experience remains the same. You hear a sentence. You type what you heard. You see exactly where you were wrong. You continue.
There is no feature on the roadmap to make this feel better. A feature that makes it feel better would make it work worse. That is the design principle, and it is not going to change.
Ready to feel stupid, sad, and lonely?
Romanian shareware — free, offline, macOS. No encouragement included.
Download Free