This is the phase many learners have been waiting for — you're finally going to start speaking. But you might be surprised: after hundreds of hours of input, your first words will come out more naturally than you expect. You already know thousands of words, you understand native speakers, and you've internalized many of the patterns of the language. The main challenge now is building the "brain to mouth" connection.
There's an important gap you'll notice immediately: the difference between what you can understand and what you can produce. You'll hear your own mistakes, your awkward phrasing, your pronunciation issues — because your ear is far more advanced than your mouth. This is completely normal. It's actually a good thing, because it means you have a built-in error detector that will help you improve. The Input-Output Gap
The Refold approach to speaking follows a specific model: Comfort first, then Accuracy. You need to be able to get your thoughts out before you can worry about getting them out correctly. Trying to be accurate before you're comfortable leads to overthinking, hesitation, and a worse experience for everyone. The CARA Model for Output
By the end of Phase 4, you'll be able to have a real conversation with a native speaker. You'll still make mistakes, but you'll be communicating. Roughly equivalent to B1.
Check out Ben's video he made about speaking Czech for the first time after 1000 hours of learning:
| Cousin | Similar | Neutral | Distant | |
|---|---|---|---|---|
| This phase | 90 | 150 | 200 | 300 |
| Cumulative | 530 | 900 | 1100 | 1800 |
4A — Preparing for Output: Build the physical and mental foundation for speaking through chorusing, reading aloud, and monologuing.
4B — Speaking Comfort: Get comfortable speaking through lots of conversation practice. Mistakes don't matter yet — volume does.
4X (Optional) — Situational Speaking: Practice using the language in specific real-life situations (travel, work, daily life). Can be done at any point or skipped entirely.
The input-output gap described here is supported by Swain's (1985) Output Hypothesis, which argues that production serves distinct functions beyond comprehension — it forces learners to notice gaps in their knowledge, test hypotheses about the language, and engage in deeper syntactic processing than comprehension alone requires.
The "Comfort first, then Accuracy" sequencing reflects the Refold CARA model. The rationale draws on multiple research strands. However, attempting accuracy before you're comfortable speaking at all leads to overthinking and hesitation, as learners try to monitor their output for correctness while simultaneously retrieving words and constructing sentences. By delaying accuracy-focused work until Phase 5, the roadmap ensures that learners first build fluency and procedural knowledge through volume and practice, then layer on monitoring and correction once the basic speaking skill is in place. As long as accuracy does become a focus, bad habits and frequent errors will not fossilize.