Immersion learning is the attempt to reproduce that soaking on purpose. It has fifty years of research behind it, some of it genuinely good, and a lot of folklore piled on top that the research doesn't support. This site separates the two, then shows you how to build the thing in a house in the suburbs, around a job.
Not by studying how it works. Krashen's input hypothesis says comprehension comes first and grammar tags along as a side effect, which is why listening and reading do most of the work and speaking shows up late, on its own schedule. The catch people miss: the input has to be understood. A Chinese news broadcast at week one is not input, it's weather.
Nick Ellis's usage-based work frames acquisition as your brain running statistics over everything it hears. More samples, from more speakers and contexts, produce a better estimate of the language. That's the whole argument for hours-per-day over hours-per-week, and for mixing your sources instead of living inside one podcast.
The Canadian immersion programs proved this the expensive way: kids with years of all-French schooling comprehended like natives and wrote and spoke with stubborn grammatical holes. Swain's output hypothesis explains why. Producing language forces processing that comprehension lets you skip. So immersion needs a speaking component eventually, or it builds a very fluent listener.
The original: go live where the language is spoken. It works, and the Canadian data from 1965 onward is the proof, but it also comes with the catch that data found. Even total immersion produced students who couldn't speak accurately, because their classrooms were teacher-centred and nobody ever pushed them to produce anything. The country helps. It isn't magic.
The AJATT and Refold communities spent two decades working out how to reproduce the mechanisms without a plane ticket: native media for input, tutors over video call for the negotiation of meaning, phone in Mandarin for ambient exposure, Anki for spacing. The internet made the access problem trivial. What's left is the attention problem, which is harder and much less discussed.
Nine sections: Krashen's five hypotheses, Swain's rebuttal, Long on interaction, Ellis on frequency, Nation's vocabulary arithmetic, spacing, the critical period measured on 670k people, and thirty years of Canadian data.
The build, in five layers: flip your environment, pick media you can actually understand, split active from passive listening, read and mine sentences, then get humans on a video call.
About two hours a day assembled out of scraps of a normal day, with a fifteen-minute floor for the shifts that wreck you, and a twelve-month picture of what the hours buy.
What each tool in the stack is actually for, then the whole thing filled in for Mandarin: Pleco, Du Chinese, Mandarin Corner, Teatime Chinese, and where your pinyin chart fits.
Eight pieces of folklore checked against the literature (you're too old; subtitles are fine; apps count as immersion), plus the four ways home immersion actually dies.
Every paper behind the claims, with links and a line on why each one earned its place. Krashen's books are free PDFs on his own site, if you want the primary source.
"The best methods might also be the most pleasant, and, strange as it seems, language acquisition occurs when language is used for what it was designed for, communication." Stephen Krashen, Principles and Practice in Second Language Acquisition, 1982, p.7