This is an experiment.
You would be the subject.

Experiment №1: the founder. n=1.

March 2021 → April 2022. The founder went from beginner to C1 Romanian. One listener, roughly 25 minutes at a time, most days, every rep and every miss recorded. His Romanian ledger — more than 126,000 recorded attempts in that one language — is this product's founding dataset, not a testimonial.

He is probably not a unique snowflake. Probably. He has a good memory, but it's not photographic. That uncertainty is exactly what an experiment is for.

Experiment №2: the replication.

The goal is to replicate his result across volunteers who suspect they are not unique snowflakes either. The method is dictation: you listen to real volunteers speaking their language at real speed, you type what you hear, and the machine grades you character by character. Your misses come back on a schedule.

It works while you sleep. Spaced repetition returns each sentence just before you would forget it. The consolidation happens between sessions, not during them — you get as much from sleeping on your reps as from making them. Difficulty bands deal you clips your ear has earned.

There is no trying it once.

One session proves nothing — one day at the gym never built a body; one week of a diet never changed one. That is why we cannot honestly sell you the retail version of this. A week of it would only buy you a verdict formed too early. What's for sale is wholesale, industrial — the whole two years, through the service entrance. Not 5 GB for your photos; enterprise storage, for the company that is your ear. The mercy of it: each document is one sentence, three to five seconds of audio, and more meaning fits in it than you'd think. Daily dollar-cost averaging in your linguistic investment: the schedule does the compounding, and missed months cannot be made up with a lump sum.

This is a bet. It is also a chance — and no other method works anyway.

Two years from now.

Two years from now you are somewhere. Either you are effortlessly transcribing native speech — having better conversations with the people you love and the people you work with — or you are exactly where you are today.

The problem feels minor, which is what makes it permanent: nodding along at dinner, asking people to repeat themselves, smiling through the parts you missed. Tolerable — and tolerable is how it survives. Minor problems of this kind are usually the visible tip of something larger, and the people who resolve them tend to find the resolution reaches further than the problem did.

As for the price: a tutor runs €30 an hour, and the protocol is hundreds of hours of ear training. Six years of gamified apps is also a price — paid in time, at the end of which you are, by our observation, roughly where you started. Compare against either. Then note that the schedule sits with you at six in the morning, and a tutor does not.

Your part. Our part.

Your part: keep your foot on the gas. About 25 minutes a day, most days, for as long as your term runs.

Our part: keep the software available, measure your progress honestly, and never go hunting for bright spots on your behalf. You get your numbers as they are — that is the whole point of a measured experiment.

Terms of participation.

Experiment №2 takes 100 participants. When the places are taken, enrollment closes until №3 opens.

The refund: money back after six months of honest reps if you judge there was no benefit. Honest reps are read from your own ledger — practice on enough distinct days and the refund is yours for the asking, by email: refunds@sitedictation.com. No streak required; a day off is part of the method.

Milgram paid his volunteers $4 to take part in an experiment — Yale sponsored the science. We have no Yale, so the $4 flows the other way: you pay to participate. His volunteers were paid to deliver shocks that weren't real. Ours are real but mild: transcripts marked wrong by a machine, instantly and in private. A machine's correction stings less than a human's.

Informed consent.