This skimmer is useless for those don't know CW. Its whole logic is "an assistant to the CW operator's own ear", so in the case of "I don't know CW but want an QSO" the result will be, let's say, entertaining. And don't tell afterwards you was not warned.

The skimmer is not a magician - it will only "hear" what the raw IQ stream actually delivered. On top of that, the skimmer has been told to "help the operator as much as you can". And you have to be clear about this: if you force an assistant to "hear" a signal where there is none, it will try very hard, and the outcome will nil. But when the signal IS there ...

The skimmer, however, doesn't live on its own - it also has a log.

The log

The built-in log follows the logic of "I work together with the skimmer" - everything in it targeted to that goal. Take the skimmer away and the log is "so-so", doing exactly what the others: writing, reading, editing, deleting, exporting. For the Basic version it is effectively just one more log, only in a different look.

What matters for the Basic version: this log is not meant to store your QSO database from the last 50 years of operating. This log ment to be "short-term", built around the logic of "worked it, exported it, deleted it". The back end is a database good for 100,000 records, of course, but it is not designed as lonh term storage. In exchange, this log exports every QSO you worked with the skimmer into ADIF - and you can import that ADIF into the BIG LOG where you keep your beloved database, alongside with labels, awards, etc. So for Basic, a friendly piece of advice: use the skimmer's built-in log the way it is meant to be used.

The log's "power" start with the Advanced version - the contest version. That is where the "I work together with the skimmer" logic really begins.

First, the log in Advanced has a contest builder. A contester should find it intuitive (especially after reading the manual). The builder comes with several contest examples, forming the BASE together with the manual. The point of the builder is: you build it yourself, you check it yourself, you operate it yourself. Pointing a finger and saying "somebody gave me a wrong contest config" won't really work here - you eat what you've cooked.

Second, with a contest activated, the log in Advanced can compare the defined multipliers (and the rest of it) against the current state of the log, and enable internal filters that act at the skimmer level. So if the skimmer happens to hear a multiplier that is potentially interesting for the operator, it will make sure you know about it - and then it's up to the operator to decide whether it's "the one" or "not the one". The feature can be switched off, so if you don't want it.

And in the DX version all filters are tuned so that a DXpedition operator can breathe out with a relief - because the skimmer starts handing them exactly the countries, continents and prefixes the operator is working right now. So the people who can't be bothered to listen to the DX operator's instructions can go blue, because the skimmer has a good filter.

So what's under the hood?

What actually is a fun: how decoders are selected and how the ones that "don't cut it" are weeded out. It is done on live air: roughly 400 hours run per decoder. And the only "held up and did well" goes into production.

Why does DX version have four decoders, on top of that in 50/75 variants? Because each one is sharpened for the TASK. Building a "universal" decoder is possible (that is exactly what the decoder in Basic is). But it will always be "just a little" weaker than 'specialised'. And that "just a little" is precisely what decides often on live air.

What Decoder1-2-3-4 are

There are no secrets here.

• D1 is a classic magnitude decoder ... shaped and sharpened ... and tuned and tweaked ... and fully reworked :). Twice. Then again. And again. Until is passed the set benchmark of my own. And it has the advantage of EE's unique TCI IQ protocol (used wisely).
• D2 and D3 are all NN. Specialized, of sourse - NNs don't come in other way. Why two? Because they are different NNs. For different tasks.
• D4 is a mongrel. An unholy offspring of a fantastically mixed marriage between the D1 and the D2/D3 principles. Very complex and very temperamental to tune, but it surprises even me quite often.

Explaining anything beyond that to a non-specialist is practically pointless, but if the question really bothers you, then the best compromise between bone-crushing linguistic theory, mathematics, programming and "explain it simply" is this: the NN has been trained to hear "CW tune singing". And its trainer is exactly like that good old drill sergeant of yours who used a ruler to teach you to hear the tune of a character, not "a dot and three dashes".

Why can the NN sometimes end up hearing something you can't? Because the NN is a cheat - it has TCI IQ and it works with it directly. Whereas in your ear there is a converter from an analogue signal into a nerve impulse, plus imagination. Which of the two "wins" depends on the air conditions.

Why do you often see coherent text in D1's decode window, while the same thing decoded by D2/D3 is an incomprehensible mush from which it is a complete mystery how a correct callsign is ever extracted? That is exactly the nut shell principal difference which was explained above: D1 is heavily tweaked magnitude-based, and an NN is an NN. Can you express "CW tune singing" mathematically? No? So the NN can't explain in common language how it picks the callsign out of that mess. "Street magic" it is. Does it irritate you? Switch the "raw decode" window off, there is a dedicated button in the settings. Will the NN always hear correctly? Of course not. It can do mistakes exactly as you do. Can easily hear one melody as a completely different one, a Q turning into a Y, especially when the signal is QSB. But the skimmer takes care of you: right-click on the callsign field and it will helpfully offer you the "alternatives".