Beyond Koch: what modern learning science adds
The Koch method got the big things right in 1936. But learning research did not stop in 1936 — and almost none of what it found since has made it into Morse trainers. Most tools still run Koch's conventions unchanged, on the assumption that a ninety-year-old dissertation settled the matter. It did not. Here are four findings that change how Morse should be trained, each labeled honestly by how strong the evidence is.
1. The best error rate is not zero
Evidence: strong in models, promising in practice. A 2019 paper by Wilson and colleagues showed mathematically that for a broad class of learners, learning is fastest when training difficulty is tuned so that accuracy sits around 85 percent — the "eighty-five percent rule". The exact number comes from simplified models, so treat it as a signpost rather than a law. But its core lesson is robust and directly contradicts common practice: a learner cruising at 97 percent accuracy is not being trained, they are being parked. Difficulty should adapt — a new character, tighter spacing, more variability — whenever performance climbs too high. Koch's fixed 90 percent gate becomes a corridor to be held, not a ceiling to celebrate.
2. Variability makes recognition robust
Evidence: strong, from an adjacent field. The closest well-studied relative of Morse learning is phonetic training in second languages — teaching Japanese speakers to hear English r versus l, for example. Its clearest finding, replicated for three decades under the name high-variability training: learners trained on many voices generalize to new speakers; learners trained on one voice largely learn that voice. Morse is auditory category learning of exactly this kind. A trainer that always plays one perfect tone at one speed produces learners who recognize that tone — and struggle on a real band, where every signal differs in pitch, speed and character. Variability, introduced gradually once the basics are stable, is arguably the single biggest upgrade available over classic trainers.
3. Recall beats review, and short beats long
Evidence: the two strongest effects in the field. The testing effect — you retain what you actively retrieve — is among the most replicated findings in psychology, and it rules out passive listening with a chart as a training mode. Just as sturdy: distributed practice. In a classic study on typewriting training (a skill remarkably close to Morse), postal workers practicing in short daily sessions learned more per hour than groups in long blocks. For Morse this converts to a simple prescription: ten to fifteen minutes, every day, always answering — never just listening along.
4. Memory for characters fades on a schedule
Evidence: strong for facts, moderate for skills. Spaced-repetition research shows that reviews timed just before forgetting produce durable memory at minimal cost. Modern schedulers (the kind behind serious flashcard software) model each item's forgetting curve individually. A Morse character's sound-to-letter mapping behaves enough like such an item that a trainer should track each character separately — which ones are solid, which are fading, which are overdue — instead of treating the alphabet as one undifferentiated mass. The fine print: fluency itself still needs massed listening practice; scheduling governs what to practice, not whether to practice.
What this looks like when you build it
We built Morse Lab as the trainer these findings describe: Koch's core (full-speed characters, mastery-gated growth) plus an adaptive accuracy corridor instead of a fixed gate, per-character tracking with targeted speed drills for characters you still translate, pitch and tempo variability that arrives in stages as your foundation stabilizes, and a strict answer-first loop. Every one of these choices is documented, and where the evidence is thin, we say so — the same standard this article holds itself to.
The practical, do-this-today version is in the learning guide.