This question touches on many topics, and this answer doesn't hope to be comprehensive.
Research on language didactics generally focusses on institutional settings (with an instructor), or, at the very least, on the learner following a defined method, such as a widely used set of learning materials composed by didactic experts. Self-learners who don't follow any definite method are therefore pretty much on their own; their "methods" and "results" are difficult to identify, to aggregate, to measure, and to study in any holistic way.
"Learning" and "acquisition" are terms for physical or mental processes leading to an increase in a skill. "Learning" is a conscious process. "Acquisition" is a subconscious process. People are born without any language skill, but with extensive and structured in-born ability to acquire their native language(s). Language learning skills (such as the most basic linguistic terminology and the ability to consciously analyze and synthesize language units) are themselves skills that have to be learned or acquired.
There are obvious differences between how humans "acquire" a language before (approximately) the age of 5 from their family members or other available fluent speakers, and how they "learn" languages subsequently. The young brain is more plastic, the older brain is a more experienced learner. As a consequence, toddlers acquire language mostly subconsciously (while the occasional conscious learning event might take the form of receiving or even requesting explicit corrections from others, or learning a new word when trying to express oneself), and adults may invest so heavily into institutionalized learning environments, methods and habits, that they may feel or assume that their language progress equates their conscious progress within that method; but Krashen would rather say (and I'd agree) that any success of a teaching method depends on a healthy proportion between conscious learning and subconscious acquisition.
Krashen would even go as far as assuming that learning has to proceed in lockstep with acquisition, that acquisition (an inborn ability) proceeds in a largely pre-determined order of complexity, and that the teaching method has to be organized in a way so that the learners are, at every stage of the language training, only exposed to content of a language difficulty level which they are ready to acquire, and that this exposure and the resulting acquisition is crucial for productivity of the language learning to be built on it.
It is well documented that the same people speak the same language differently depending on whether the circumstances encourage them to speak "casually" or "diligently". (Random example.) Krashen explains this phenomenon through a monitor function, i.e., through self-monitoring and potentially self-correction of speakers when producing a langugage (both native or learned). He goes on to postulate that people differ on how much they monitor what they are saying, sometimes depending on circumstances (including social circumstances), but also depending on their individuality. The latter leads to a categorization of people as "under-users", "optimal users", and "over-users" of the monitoring function.
Krashen's own work in language didactics (e.g., composition of curricula and of learning materials) was focusing on how to enable language learners to receive the optimal balance of learning and acquisition, and how to become optimal users of their own monitoring function. It would be a caricature of his work to suggest that Krashen would discourage language students from any language production until they are fluent in the language. It might perhaps be fair to say that Krashen would find it pointless or even detrimental to rush the conscious learning effort too far ahead of the level of language complexity to which the student has been amply exposed before. A self-learner without a solid feedback loop can shoot themselves into the foot and misunderstand some aspect of a language rather deeply and then work further to reinforce those poor habits; the same can happen with the "help" of an incompetent or crackpot instructor.
We can admire the plastic brains and the loving parents and the abundance of time and energy that the small children have at their disposal when they are learning to navigate the world. But trying to learn anything big the same way small children do it would be a cargo cult practice.
Adults have learning skills and resources that toddlers don't have. If the goal is the speed of reaching a certain level of fluency within a language per effort expended, it's best if the adults take advantage of all the learning skills and resources that they have.