What "vocabulary size" actually counts
The unit is the word family, not the dictionary entry. "Know", "knows", "knowing", "knew" and "known" are one family, because a learner who knows one form can usually work out the rest. Counting surface forms instead would multiply every figure by roughly three and make comparison with published research impossible.
For rough calibration: an adult native speaker of English knows somewhere between 15,000 and 20,000 word families, and reaches about 10,000 by the end of secondary school. A learner at B2 typically holds 4,000 to 5,000; at C1, around 8,000. The first 2,000 families cover roughly 80% of the words in everyday text, the first 3,000 about 90%, and the first 5,000 about 95%.
Those coverage figures are the practical point. Comprehension research puts the threshold for following a text at around 95% word coverage, and for comfortable independent reading at 98%. The difference between 4,000 and 8,000 families is the difference between working through an article and reading it.
Why the pseudowords are the whole design
A yes/no test measures self-report, and self-report drifts. Some people are cautious and say no to words they half-know; others recognise a shape and say yes. Without a control, the test would measure confidence as much as knowledge.
Pseudowords supply that control. Because nobody knows them, every yes on a pseudoword is a measurable error, and the rate of those errors estimates how often the same person said yes to a real word they did not actually know. The correction applied here is the standard one: the corrected proportion for a band is your hit rate minus the false-alarm rate, divided by one minus the false-alarm rate, floored at zero.
A consequence worth knowing: if your false-alarm rate goes above 40%, the correction is dividing by a small number and the estimate becomes unstable. The tool tells you this rather than quietly reporting a figure it does not believe.
How to grow the number
Vocabulary growth is dominated by encounter count, not by study time. Most estimates put the number of meetings needed to learn a word from context at somewhere between eight and twenty, which is why reading volume beats word lists for anything past the first couple of thousand families.
The efficient place to read is just above your current level, where roughly 95-98% of the words are already known. Below that and you learn nothing new; above it and you spend the session in a dictionary instead of reading. This is the reason level-matched reading material matters more than any particular study technique.
Spaced repetition is worth adding for the words you meet but do not retain, particularly in the 3,000-8,000 band where encounters are naturally rare. It is a supplement to reading volume, not a substitute for it: a word learned from a flashcard and never met in context tends not to survive.