Self-study course · working competence · 10 weeks · part 1 of 2

Copying Morse

Morse is a spoken language that happens to have a written notation. This course teaches the listening. It never shows you the chart.

Progress0%

This bar is the word LISTEN, drawn to scale in real Morse timing. Tap it to hear where you are.

What you will be able to do

Copy an unfamiliar plain-text transmission onto paper — the full 26-letter alphabet, sent at 12 WPM character speed, at 90% character accuracy, cold, with no chart and no replay. And run the drill regimen that carries you onward without this course.

Where this starts you

No exposure to Morse whatsoever. Comfortable with structured self-study and technical reading. No radio, no key, no licence, no musical training assumed. Headphones and a pencil are the entire equipment list.

What this is not

Time budget

Roughly 24–26 hours across ten weeks, published per block so you can decide one sitting at a time. That figure already has the standard 2.5× correction applied — self-study consistently runs long. Sourced fact S6

Week 6 is a real stopping point. It crosses all three thresholds and leaves you copying sixteen letters — a genuine skill. Weeks 7–8 close the alphabet, Week 9 adds light and is self-contained. Stopping at any of those boundaries leaves something whole rather than something half-crossed. Inference

How claims are marked

Morse instruction has an unusual evidence problem: the primary research is a 1935 German engineering dissertation and a handful of pre-1960 psychology papers, while nearly all modern writing about it is published by people selling an app. Every load-bearing claim below carries a marker so you can tell which is which.

Sourced fact S# traceable to a numbered source Inference my reasoning, in no source Vendor framing authored by someone selling the thing Contested the field genuinely disagrees Single study real but unreplicated

Sound settings

These persist. Leave character speed at 12 unless a block tells you otherwise — that number is the whole argument of Week 2.

Silent practice — and a note on order

Everything in this course runs on any of four channels. Pick one here and every exercise, every drill and every self-check uses it. Two of them are completely silent, so a meeting or a five-minute gap is practisable.

Which channel should you learn on first?

The common answer — "I learn better visually" — is the one part worth examining. Pashler and colleagues reviewed the evidence for matching instruction to a learner's preferred modality and found the studies with the right design either failed to show the effect or contradicted it. Preferences are real; the benefit from meshing instruction to them is not established. Sourced fact S9

But that research does not settle this case, because it is about presenting the same content in different formats. Here the format is the content. Morse by ear and Morse by light are not two presentations of one skill — they are two skills with different targets. So the real question is not which suits you; it is which one you actually want. Inference

  • Want to copy radio CW? The target is auditory. Learn on sound; use vibration when you cannot listen.
  • Want to read signal lamps and send with a torch? The target is visual. Learn on light from Week 1 — that is not a compromise, it is the actual skill. Accept the 6 WPM ceiling.
  • Want both? Sound first is the better bet, because the auditory channel can carry 12 WPM and the visual channel cannot, so the harder-won timing is learned at the rate that supports it.

Why vibration is the good silent option. The flash-rate limit is a photosensitivity constraint and applies only to light. Vibration has no such ceiling, so it runs at the full 12 WPM character speed — above Koch's critical rate, unlike the lamp. It is the only silent channel that does not force you below the speed where the pattern holds together. Inference

And sending is silent already. Any key exercise works with the sound off: tap, and the analyser still measures your timing. That is real practice you can do under a table.

Week 1~2 hrs · 5 blocks

Copy something real, today

This week producesA sheet of paper with 25 characters on it that you copied by ear. Crude, two letters only, and a complete instance of the actual skill.

The whole task, in its simplest form, on day one. You will not "learn the alphabet and then start copying." You start copying now with two letters and add letters to a thing you can already do.

Where you are expected to struggleBlock 2 is the odd one. Two sounds is not enough information to feel like progress, and your instinct will be to want the chart so you can "check." That instinct is the thing this course is built to defeat. Sit in it.
Listen~10 min

Get a sheet of paper and a pencil. Press play. You will not be told what letters these are, and you are not trying to identify them. Every time you hear one complete sound-shape begin and end, make a single dot on the paper. That is the entire exercise.

This is Koch's own first exercise, from the 1935 dissertation, and almost no modern course includes it. His reasoning: marking rhythm on paper builds the link between the heard shape and the writing hand before any letter meaning is attached, and it forces you to listen for the shape rather than the parts. Sourced fact S1

Count your dots afterward. If you got 20, your ear is already segmenting the stream, which is most of what this block is for.

Meet two sounds~25 min

Two letters: K M

Play each one. Say the letter out loud as you hear it. Do this until the letter arrives in your head before you have finished thinking about it — which will feel like nothing happening, and then suddenly like the sound simply is the letter.

K and M are Koch's own suggested starting pair, chosen because their sound-shapes are maximally unlike each other. He noted that any well-differentiated pair works. Sourced fact S1

Copy — worked~20 min

Before you copy cold, watch it done. Press play and follow along with the answer visible. Your job is only to bind sound to letter while the pressure is off.

K M M K   M K K M   K K M M   M M K K

This is a fully worked example. Week 3 gives you partial ones. Week 6 gives you none — the support is scheduled to disappear, and you should expect to notice it going.

Copy — cold~45 min

The real thing. Paper and pencil, not the keyboard. Press send, write the letters as they arrive, then type what you wrote into the box. Do not replay before committing.

If you drop a character, leave a gap and keep going. Chasing a missed letter costs you the next three — this is the single most common beginner failure and you should practise the recovery now rather than at speed later. Inference

Write~15 min

Three sentences on paper, for you, not for anyone else:

  • What did the moment of recognition actually feel like — did you decode it, or did it arrive?
  • At what point did you want to see the dots and dashes written down, and what were you hoping that would give you?
  • Which of the two letters is less solid, and how do you know?

Prompted self-explanation is where most of the gain in worked-example learning actually comes from, and learners left to themselves explain superficially or not at all. Writing it beats thinking it. Sourced fact S7

Week 2~3 hrs · 4 blocks

Speed is two numbers, not one

This week producesA timing diagram you derived by hand, and a copied transmission of five letters — plus a defensible answer to "what speed are you learning at," which is a trick question.
Where you are expected to struggleThe arithmetic here is easy. The idea is not. You are going to be asked to accept that slowing Morse down makes it harder, which contradicts how every other skill you have learned works.
Derive~50 min

Morse timing is built from one unit: the length of a dit. Everything else is a multiple.

ElementUnits
dit1
dah3
gap between elements inside a letter1
gap between letters3
gap between words7

Speed is defined against one reference word, PARIS, chosen because it comes to exactly 50 units including its trailing word gap. Count it yourself before reading on — the diagram below is drawn to scale, and every bar is one element.

Tap the diagram to hear it. Sourced fact S4

So one word per minute means fifty units per minute, and:

dit = 60 / (50 × WPM) seconds
    = 1200 / WPM milliseconds

Work these out on paper before committing:

1. How long is a dit at 12 WPM? And at 5 WPM?

At 12 WPM, 1200/12 = 100 ms. At 5 WPM, 1200/5 = 240 ms — nearly two and a half times longer. Hold onto how large that difference is.

2. Of PARIS's 50 units, how many are inside characters, and how many are the gaps between characters and words?

The five characters take 31 units between them. The four inter-character gaps take 4×3 = 12, and the word gap takes 7, so 31 inside, 19 in the spacing. That 31/19 split is the whole basis of the next block.
The threshold~40 min

Here is the idea the course turns on. You can stretch those 19 spacing units without touching the 31 character units. The letters keep their exact rhythm; the silence between them grows. That gives two independent speeds: a character speed and an effective speed.

This is Farnsworth timing, named for Donald R. "Russ" Farnsworth, W6TTB, and given a formal specification by the ARRL so that "18 WPM characters at 5 WPM overall" means one reproducible thing. Sourced fact S4

Hear the difference. Same six letters, three ways:

The third button is the important one. Listen to it twice. A letter sent at 5 WPM is not the same sound played slowly — it is a different sound, one you would have to learn separately and then unlearn.

Koch measured exactly where this breaks. Below roughly 10 WPM, he found, the character stops being perceived as a single acoustic whole and starts being perceived as a sequence of separate beeps you have to count and reassemble. He called the intact version the Gestalt, and named the changeover the critical rate. His conclusion was blunt: training speed must be above that rate from the very beginning. Sourced fact S1

A correction you will need. Nearly every app and article says the Koch method means starting at 20 WPM. Vendor framing

Koch's dissertation says something different. He identified ~10 WPM as the critical rate below which the shape decays, and found 12 WPM optimal for initial learning, with speed increased toward 20 only after all characters were known. The 20 WPM figure is the destination in his report, not the starting line. This course uses 12 because that is what the primary source actually says. Sourced fact S1

3. A friend says: "I'm learning at 5 words per minute and I'll speed up later." Two separate things are wrong with the plan. Name both.

One: 5 WPM is below the critical rate, so they are learning sounds that decay into countable beeps — they are training the counting habit, not the recognition reflex. Two: "speed up later" means the sounds they learned are not the sounds they will need, so the speed increase is not a scaling-up but a relearning. The fix is not to go faster overall; it is to split the speed into two numbers and leave the character number high.
Add letters~50 min

Three new: R A N  —  working set K M R A N

Hear each alone first, then drill the set. Move on when you are at 90% on the drill below; that threshold is Koch's, and it is the rule for every week from here.

R and K are mirror images of each other written down, and so are A and N. Koch specifically noted that visual mirror-images cause almost no confusion by ear, because their sound-shapes are plainly different — which is a small proof that the written form and the heard form are not the same object. Sourced fact S1

Schedule~10 min

Set up the practice pattern now, because it does more work than anything else in this course.

Two sessions a day, half an hour each, morning and afternoon. Not one long session. Koch tested this directly and reported that distributing the practice hours produced faster learning and better retention than massing them — and that half an hour was the useful ceiling for one sitting. Sourced fact S1

This is also the single most robust finding in the modern study-strategies literature, independently of Morse: distributed practice and retrieval practice are the two techniques that survive meta-analysis. You are getting both. Sourced fact S6

Week 3~3.5 hrs · 4 blocks

No translation step

This week producesA copied transmission of eight letters, and evidence about yourself: whether a letter is arriving as a reflex or being decoded. Those feel similar from the inside and are not the same.
Retrieve~20 min

These reach back to Weeks 1 and 2 deliberately. A self-check sitting under the section that taught it measures whether you just read something. Spacing effects show up on delayed tests and not immediate ones, so the useful version of this is the annoying version. Sourced fact S6

1. Without scrolling up: how long is a dit at 12 WPM, and what fraction of PARIS's 50 units can Farnsworth timing stretch?

100 ms. And 19 of the 50 units — the inter-character and inter-word gaps. The 31 units inside the characters never move.

2. Cold ear check, no warm-up. Four characters from Weeks 1–2. Write them down before revealing.

Whatever you scored, note it. The number matters less than whether it surprised you.
Diagnose~40 min

There are two different things happening in your head when you copy, and you need to be able to tell them apart.

DecodingRecognition
You hear the parts, hold them, count them, and assemble a letter.The letter arrives. There is no intermediate object.
Scales badly. Fails around 10–12 WPM and gets worse.Scales. Speed becomes a matter of practice, not method.
You can feel yourself working.Feels like you did nothing, which is why people distrust it.

Koch's criticism of every method that came before him was aimed exactly here. Teaching from a printed chart, he argued, requires the operator to count the elements and reassemble them for comparison against a memorised optical symbol — workable at low tempo, impossible at speed. He recommended charts never be used at all. Sourced fact S1

The test. Play a single letter. Say it out loud immediately — within about half a second, before you have time to think. If you can do that reliably, it is recognition. If you need a beat, it is decoding, and that letter needs more drill rather than more speed. Inference

Copy — completion~30 min

Scaffolding step two. Part of the answer is given; you fill the gaps. Play, copy what you can, then type the full string.

M A _ _   _ A N _   K _ _ M

Answer: MARK RANK KARM. Compare against your paper, not your memory of how it went.

Add letters~1.5 hrs

Three new: U D W  —  working set K M R A N U D W

A note on ordering. The character sequence used by nearly every Koch trainer — K M R S U A P T L O and so on — does not appear in Koch's dissertation. It comes from a much later software implementation and is attributed to him by convention. Koch's own sequence was different, and his stated principles argue against parts of the popular one: he explicitly warned that confusable pairs like S/H and U/V should be kept apart, and the standard sequence introduces S immediately. Sourced fact S1 Contested

This course therefore uses a sequence I built from Koch's stated principles rather than either his list or the popular one: well-differentiated shapes first, confusable families deferred, and the runs of pure dits and pure dahs held back until there is enough context to contrast them. That is my choice, not his, and you should treat it as such. Inference

Week 4~3.5 hrs · 3 blocks

The flat stretch

This week producesEleven letters copied, and a written decision about what you will do when progress stops — made now, while you can still think clearly about it.
Where you are expected to struggleThis week is placed here on purpose. Around now the curve usually goes flat: you are not getting worse, you are simply not getting better. That experience is the subject of this week rather than an interruption to it.
Read~45 min

In 1897 and 1899, two psychologists at Indiana University followed telegraph students for months and published the first careful learning curves anyone had produced for a complex skill. They found that receiving improved, then stopped, then improved again — and their explanation was that the operator was building a hierarchy of habits: first letters, then words, then whole phrases, with each level having to become automatic before the next could form. The flat stretch was the lower level consolidating. Sourced fact S2

Their own words are worth having: mastery of the telegraphic language, they concluded, involves mastery of the habits of all orders. A student who jumps from 18 to 25 words a minute has not suddenly learned more English — something else has automated. Sourced fact S2

Then, sixty years later, someone argued it was never there.

Fred Keller — who had run code training research for the US Army — published a paper in 1958 titled "The Phantom Plateau," arguing that the plateau in code learning is an artifact of how the curve is drawn and how the students were taught, not a necessary feature of skill acquisition. Under good instruction, he said, it largely disappears. Sourced fact S3 Contested

And the argument did not end there: later authors defended the plateau on different theoretical grounds, and a 2017 review in Current Directions in Psychological Science opens by observing that the field has been arguing about whether training plateaus are real for a hundred and twenty years. Sourced fact S5

What to actually do with a contested finding.

Not "pick a side." Notice which part is disputed and which part is not. Nobody disputes that learners experience flat stretches. What is disputed is whether the flatness is intrinsic to the skill or produced by the teaching. Both sides agree on the practical consequence, which is the only part you need: a flat stretch is not evidence that you have hit your limit, and it is not evidence that the method is failing. Keller's version is arguably more encouraging than Bryan and Harter's, because it says the flat part is at least partly removable. Inference

1. Bryan and Harter watched real students and saw a plateau. Keller looked at similar data and said it was not really there. Before revealing: what kind of thing could they disagree about, given they were not disputing each other's measurements?

Mostly two things. What counts as a plateau — a genuinely flat interval, or just a slow stretch that looks flat at a particular scale and smooths out when averaged across learners or plotted differently. And what causes it — an intrinsic reorganisation of the skill, or the ordinary consequence of instruction that stopped pushing on the right thing. This is the standard shape of a lot of scientific disagreement: same numbers, different construct and different attribution. Noticing that is worth more here than knowing who won.
Decide~15 min

Write this down on paper now, before you need it. When the drill goes flat for a week or more, which of these will you do?

  • Keep the set the same and keep the sessions short, on the theory that consolidation is happening and needs time.
  • Add a letter anyway, on the theory that the flatness is boredom and the level has stopped demanding anything.
  • Raise character speed by 1–2 WPM and let accuracy drop temporarily.
  • Drop effective speed to widen the gaps and rebuild accuracy first.

There is no consensus answer, and anyone who gives you one confidently is overselling. The value of choosing now is that you choose while you are not frustrated. Inference

Add letters~2.5 hrs

Three new: G F Y  —  working set K M R A N U D W G F Y

You now have enough letters for real words. Try this one as a word rather than as five separate letters — it is the first hint of what Week 5 is about.

Week 5~3.5 hrs · 3 blocks

Words, not letters

This week producesFourteen letters copied as continuous text, and at least one word you caught whole without hearing its letters.
Choose the fix~30 min

These are deliberately mixed. Each describes a symptom; your job is to pick which intervention applies and say why. Executing a named technique is not the skill — choosing among them is. Sourced fact S6

1. You copy individual letters at 90%, but as soon as the text is real words you fall behind and lose whole chunks. Character speed, spacing, or something else?

Something else — and neither speed control is the fix. You are trying to hold letters in working memory long enough to assemble words while more letters keep arriving. Bryan and Harter's answer was that word-level habits have to form on top of letter-level ones, and they form from exposure to real words, not from more letter drill. Send words. Also start writing one letter behind what you hear rather than trying to stay level with it.

2. You confuse two specific letters constantly, and only those two. Everything else is solid.

Neither a speed nor a spacing problem — a discrimination problem, and it needs targeted contrast drill: those two letters alternating, nothing else, until the difference is obvious. Raising speed will make it worse; slowing character speed below the critical rate will destroy the shapes you have. Note that this is exactly the failure Koch predicted from putting confusable pairs close together in the sequence.

3. Everything was fine, and now you are missing the character immediately after a long one. Consistently.

A spacing problem: your recovery time is too short. Drop effective speed — widen the gaps — and leave character speed exactly where it is. This is the case Farnsworth timing exists for, and it is the one situation in this course where "slow down" is the right advice, because it slows the silence rather than the sound.
Add letters~1.5 hrs

Three new: T E O  —  working set K M R A N U D W G F Y T E O

These are the short ones, held back until now on purpose. E is a single dit and T a single dah, and learned early they tend to become a default guess for anything unclear. With eleven solid letters around them they land as themselves. Inference

Copy words~1.5 hrs

Real words now. Copy onto paper. Deliberately try to let a word land whole rather than assembling it — and expect that to work perhaps once, briefly, and then not again for a while.

When a word does arrive whole, that is the hierarchy Bryan and Harter described, forming in real time. It is worth noticing precisely because it feels like less effort rather than more, and effortlessness reads as "not learning" when it is the opposite. Sourced fact S2

Week 6~2 hrs · 2 blocks

Sixteen letters

This week producesSixteen letters copied cold, and a delayed check on everything from Weeks 1 and 2.

This is a natural stopping point. All three thresholds are crossed and you have a working skill. It is also the prerequisite for Sending Morse, which can be started from here in parallel with Weeks 7–10.

Add letters~1 hr

Two new: I S  —  final set K M R A N U D W G F Y T E O I S

S is the one Koch warned about — three dits, easily confused with H's four. H is not in this course, which is part of why S can safely come in now. When you add H later, drill S and H against each other directly from the start. Sourced fact S1

Retrieve~20 min

1. Back to Week 2, no scrolling. Why is a character sent at 5 WPM not simply "the same character, slower"?

Because below roughly 10 WPM the acoustic shape decays — Koch's critical rate. The character stops being one perceptual object and becomes a sequence of separate beeps you count and reassemble. It is a different stimulus that you would learn separately and then have to unlearn.

2. Your sending is being copied badly. Your dah/dit ratio measures 2.6 instead of 3.0, and your inter-character gaps measure 1.9 units instead of 3.0. Which do you fix first, and why?

The gaps, and it is not close. A 2.6 ratio is well within what the ear tolerates — dahs are still plainly longer than dits, which is all the ear asks. Gaps at 1.9 units are close to the 1-unit intra-character spacing, so your letters are running together and the receiver cannot tell where one ends. Fixing the ratio would change almost nothing; fixing the gaps fixes the copy.
Week 7~4 hrs · 2 blocks

The confusable families

This week producesTwenty-two letters, including every one that is commonly mistaken for a letter you already own — drilled against its partner rather than alone.
Where you are expected to struggleThese will feel like they break letters you had solid. That is the point of introducing them in pairs: the confusion is the thing being trained away, and it cannot be trained away in isolation.
Contrast drill~2 hrs

Three new: H V B  —  working set K M R A N U D W G F Y T E O I S H V B

Each is introduced against the letter it will be confused with. Koch warned specifically that sound-alike pairs cause trouble; his answer was to keep them apart in the sequence. Mine is to keep them apart until now and then deliberately jam them together, because the discrimination has to be built at some point and doing it under supervision beats discovering it later. Sourced fact S1 Inference

NewConfused withDrill
HS — one extra dit
VU — one extra dit before the dah
BD — one extra dit at the end

All three are the same mistake in different clothes: a trailing dit you did or did not hear. If you are missing them, the fix is not more speed and not wider gaps — it is counting nothing and instead learning the two shapes as different words. Inference

Contrast drill~2 hrs

Three new: C L P  —  working set now 22 letters.

NewConfused withDrill
CK — a trailing dit
LF, and R — same length, different middle
PW, and F — the dahs sit in different places
Week 8~4 hrs · 2 blocks

Closing the alphabet

This week producesAll 26 letters copied as continuous English text, with no character excluded.
Contrast drill~2 hrs

Three new: J Q X  —  twenty-five letters.

These are the four-element characters with heavy dahs. They are long enough that impatience is the main enemy: the shape is not finished when you think it is. Inference

NewConfused withDrill
JW — one more dah
QG, Z — the family that starts with two dahs
XB, D — dah at both ends
Close the alphabet~2 hrs

One new: Z  —  and the set is closed. A–Z

Now full English. This is the first time the drill has not been drawn from a restricted set, which means it is also the first time letter frequency works for you: E, T, A, O, I and N carry most of the text, and you have owned those for weeks.

Week 9~1.5 hrs · 2 blocks

By light

This week producesA message read off a flashing lamp. Same code, different channel, and a safety limit that turns out to collide with Week 2.
Read this before switching the light onMorse by light is flashing content, and at the speeds you have been using it flashes fast enough to matter. The first block explains exactly how fast and what this course does about it. If you have any photosensitivity, read that block and then decide — every visual exercise has an audio equivalent, and skipping the week costs you nothing.
The rate problem~30 min

The code is channel-independent. The timing structure is the message, and it survives translation to anything that can be switched on and off — a tone, a lamp, a torch, a shutter. That is genuinely the same skill.

But the constraints are not the same, and one of them is a safety limit rather than a preference.

WCAG 2.3.1 sets the general flash threshold at three flashes per second; above that, flashing content can trigger seizures in people with photosensitive seizure disorders, which affect roughly one person in four thousand. Sourced fact S8 A run of dits is a square wave, so the worst-case flash rate is one flash per two dit-lengths:

Character speedDitWorst-case flashes/sec
6 WPM200 ms2.50within the limit
8 WPM150 ms3.33over
12 WPM100 ms5.00over
20 WPM60 ms8.33over

So visual practice is capped at 6 WPM character speed in this course, and the lamp reports its own flash rate so you can check the arithmetic rather than trust me. Inference

Which collides with the whole first half of this course.

Koch's critical rate is 10 WPM: below it, the acoustic shape decays. The safe visual rate is 6 WPM. Both constraints cannot be satisfied at once. I am not going to paper over that. My reading is that Koch's finding is about auditory pattern perception specifically, and the eye integrates a flashing light differently from how the ear integrates a tone — which is consistent with signal-lamp work historically running far slower than radio telegraphy. But I have no source that tests the visual case directly, so treat this as my reasoning and not an established result. Inference Contested

Read the lamp~1 hr

Copy from light instead of sound. The lamp below is deliberately small — a small flashing area sits under the threshold's size exemption, which a full screen does not. Sourced fact S8

Expect this to be much harder than the ear at first, and expect it to come back faster than it did the first time. You are not learning the code again; you are attaching a channel to something you already know.

Week 10~1.5 hrs · 1 blocks

On your own

This week producesA cold assessment score, and a written regimen you can run for six months without opening this page again. That second one is the actual deliverable.
Assess — copying~45 min

No worked example, no completion, no partial answer. Paper, pencil, one pass, no replay. Full alphabet.

Notice this looks nothing like Week 1, which handed you the answer before you started. That was scheduled: support that helps a beginner actively obstructs an improver, so it was designed to withdraw. Sourced fact S7

Write the regimen~45 min

One page, your own words, your own numbers:

  • When are your two half-hour sessions, on which days?
  • What are your two speed numbers today, and what has to be true before either changes?
  • How will you tell recognition from decoding on a character that has gone stale?
  • What is your rule for a flat stretch — the one you decided in Week 4?
  • What is still missing: numerals, punctuation, prosigns, Q-codes. In what order, and why that order?

If you can write that page without consulting this one, the course is finished. Everything left is more of the same, and you know how to add it.

Last one. An app advertises "the Koch method — scientifically proven, start at 20 WPM, learn the code in two weeks." Three things should bother you. What are they?

The speed: Koch's dissertation gives ~10 WPM as the critical rate and 12 as optimal for initial learning, with 20 as a destination. "Scientifically proven": the evidence is one 1935 dissertation with small subject numbers and no modern control design. It is the best available and it is thin. "Two weeks": Koch's reported result was a class reaching about 12 WPM in roughly fourteen hours — instructor-led, twice daily, reported by the method's own author. Single study And the fourth thing: it was written by someone selling a subscription. Vendor framing
Appendix

The chart

There is no dot-dash chart in this course. That is the one design decision everything else rests on, and it is Koch's recommendation rather than mine: he argued charts should never be used, because reading one trains you to count elements and reassemble them, which is the habit that caps your speed. Sourced fact S1

Locking it away entirely would be dishonest, though — you can find it in four seconds and a course that pretends otherwise is just posturing. So here is the honest version of the trade: looking at the chart will feel like it helps and will cost you weeks of unlearning later. If you are stuck on one letter, the cheaper fix is almost always contrast drill against a letter you already own. Inference

Sources

What this rests on

Every load-bearing claim is keyed to one of these. Each was opened and read rather than cited from a summary — with one exception, flagged below, where the primary source is in German and I worked from a translation.

  1. Koch, L. (1936). Arbeitspsychologische Untersuchung der Tätigkeit bei Aufnahme von Morsezeichen — dissertation submitted 1935, Braunschweig Technical University; published in Zeitschrift für angewandte Psychologie und Charakterkunde, vol. 50 (1–2), pp. 1–70. Tier 1 — primary. Scan, transcription and full English translation are on the Internet Archive. Caveat: the original is German and I worked from the translation plus a clause-by-clause English summary, not the German text. Source of the Gestalt argument, the ~10 WPM critical rate, 12 WPM as optimal, the no-charts recommendation, the 90% advancement rule, the K/M starting pair, the two-sessions-a-day finding, and the confusable-pair warnings. Note also that the author is Dr.-Ing. Ludwig Koch, an engineer born 1901 in Kassel — commonly described as a psychologist, and frequently confused with the well-known wildlife sound recordist of the same name, who was a different person.
  2. Bryan, W. L. & Harter, N. (1897, 1899). "Studies in the physiology and psychology of the telegraphic language," Psychological Review 4(1), 27–53; and "Studies on the telegraphic language: the acquisition of a hierarchy of habits," Psychological Review 6(4), 345–375. Tier 1 — primary, read in facsimile. Source of the plateau observation and the hierarchy-of-habits account.
  3. Keller, F. S. (1958). "The Phantom Plateau," Journal of the Experimental Analysis of Behavior 1(1), 1–13. Tier 1 — primary, open access. The refutation. Included because searching for the counter-argument is the step that gets skipped.
  4. ARRL / Bloom, J. (KE3Z). "A Standard for Morse Timing Using the Farnsworth Technique," QEX; with ITU-R Recommendation M.1677 for the element ratios. Tier 2 — canonical standard. Source of the PARIS 50-unit definition, the 1200/WPM dit, and the Farnsworth split. The arithmetic in Week 2 was verified by working the 31/19 decomposition out independently rather than taking it on authority.
  5. Gray, W. D. (2017). "Plateaus and Asymptotes," Current Directions in Psychological Science 26(1). Tier 3 — review. Used only to establish that the plateau question is still genuinely open, which is a claim a review can properly support.
  6. Hattie, J. & Donoghue, G. (2021); Dunlosky, J. et al. (2013). Meta-analyses of learning techniques. Tier 3 — meta-analysis, appropriate here. Distributed practice and retrieval practice as the two techniques that survive aggregation; spacing effects appearing on delayed rather than immediate tests; interleaving forcing method selection.
  7. Renkl, A.; Kalyuga, S. & Sweller, J. Worked-example fading, the expertise reversal effect, and prompted self-explanation. Tier 2/3 — established programme of work. Basis for the worked → completion → independent sequence across Weeks 1, 3 and 6, and for the writing prompts.
  8. W3C. WCAG 2.1/2.2 Success Criterion 2.3.1, Three Flashes or Below Threshold, and the associated general flash threshold definition. Tier 2 — published standard, read directly. Source of the three-flashes-per-second limit and the small-area exemption (no more than 25% of any 10-degree visual field). The per-speed flash rates in Week 11 were calculated from the dit lengths, not taken from a source.
  9. Pashler, H., McDaniel, M., Rohrer, D. & Bjork, R. (2008). "Learning Styles: Concepts and Evidence," Psychological Science in the Public Interest 9(3), 105–119. Tier 3 — commissioned review, plus later direct tests using its prescribed design. Basis for the claim that modality preferences exist but that matching instruction to them is not shown to improve learning. Note the limit of its reach: the review concerns presenting the same content in different formats, which is not the same situation as a skill whose target is defined in a modality.

What I could not verify. Koch's reported "class to 12 WPM in fourteen hours" is widely repeated but I have seen no subject counts, control condition, or independent replication — and a 1943 review found his two-tone variant produced a difference that probably was not significant. Treat the headline result as a favourable single case. Single study

The weakest link. Week 9 assumes that learning by ear transfers to reading light. The timing structure is identical, but I found no study testing the transfer, and the safe visual flash rate sits below Koch's critical rate — a genuine unresolved tension. If one claim here is wrong, it is that one. Inference Contested