For quite a long time I thought the Gene Cafe CBR-301 would simply stay at the centre of this project. It was the machine I was learning on, the machine most of the early roasting notes were built around, and the machine that forced me to pay attention because it does not give you very much telemetry to hide behind.

By the time I was seriously thinking about moving on, I had completed 215 roasts and had put close to 43.9 kg of coffee through it. I had also replaced heaters, changed the way I recorded roasts, added power monitoring, experimented with audio, built tools around the machine and spent an unreasonable amount of time looking at JSON files that most normal people would probably have ignored.

I am not abandoning the Gene Cafe, and I am certainly not going to pretend those roasts no longer matter. But I do think the roasting part of the Gene Cafe story has reached a natural end. The next stage of Coffee | One Roast at a Time will be centred on the Kaleido M2 and on learning what becomes possible when I can see more of what the coffee and the machine are doing.

This is probably the right point to bring the Gene Cafe roasting story up to date and close the loop.

Where the last roast notes left me

The later Gene Cafe roasts were very different from the early ones.

At the beginning I was trying to understand basic things such as how quickly the machine responded, where first crack might happen, whether I should change temperature earlier or later, and what a sensible total roast time looked like. A lot of the learning was fairly crude because the information available from the machine was crude. I had chamber temperature, time, what I could see through the glass, what I could hear, the smell coming from the exhaust and eventually the roasted coffee itself.

Over time the process became much more controlled. Most of my normal batches settled around 250 g, and I became much more interested in repeatability than in finding a supposedly perfect profile. I wanted to know whether I could start from a known machine state, use the same sequence of temperature changes and end up somewhere recognisably similar.

One of the later Honduras roasts is a reasonable example of where that process had reached. I charged 250 g and finished with 206 g, a weight loss of 17.6%, with a total roast time of about 13 minutes 30 seconds. The point of recording numbers like that was never that 17.6% was somehow the correct answer. It was that I could compare it with earlier roasts, look at what I had changed and then see whether the cup gave me a reason to keep going in that direction.

That became a much more useful way for me to think about roasting.

Repeatability became more interesting than the profile

One thing the Gene Cafe taught me quite well was that repeating a roast is not the same as repeating a list of settings.

The state of the machine mattered. A cold first roast behaved differently from a machine that had already been running. Ambient conditions moved around. The coffee changed. The heater aged. And because I was measuring more as time went on, I could see differences that I probably would not have noticed during the first few months.

That gradually changed the question I was asking.

Instead of asking whether a roast curve looked right, I became more interested in whether the machine behaved in the way I expected it to behave. If I made a change at a particular point, did I see the consequence when I expected it? If I repeated the same coffee, did total roast time, weight loss, colour and the cup remain reasonably close? If they did not, could I explain why?

I still made plenty of mistakes, but the process became much less random.

There is also something useful about learning this on a machine that does not hand you every answer. I had to learn to watch the coffee, listen to it and build a record around it. I suspect I will appreciate the extra telemetry on the M2 more because I spent so long without it.

The heater problem changed things

The part I cannot ignore is reliability.

I had heater failures more than once. At one stage I ended up with the wrong voltage replacement heater before eventually getting the correct 230 V unit. Later there were further failures and replacements, including another new heater in 2026 and spare units kept on hand because by then I had lost confidence that the problem was completely behind me.

Every heater change created another problem for the learning process. Even when the machine was roasting again, I could not assume it was behaving exactly as it had before. A profile that had made sense with one heater could not automatically be treated as a fixed reference after the hardware changed.

That became increasingly frustrating because the project had moved on.

I was no longer only trying to make enjoyable coffee. I wanted to understand why one roast behaved differently from another, and I wanted the machine to be stable enough that the comparison meant something.

I do not think that makes the Gene Cafe a bad roaster. My own machine has simply had a reliability history that eventually became part of the decision to move on.

I ended up building instrumentation around the roaster

Because the Gene Cafe does not expose bean temperature in the way I wanted, I gradually started measuring everything else I reasonably could.

RoastAid grew out of that. I used BLE capture from the Gene Cafe, added electrical monitoring through a Shelly device, recorded audio using a UMIK-1 and experimented with identifying candidate crack activity from the sound. The roast data could then be lined up with operator actions and analysed afterwards.

That became a project in its own right.

The important word with the audio work is candidate. I never reached a point where I was prepared to say that an acoustic algorithm had somehow solved first crack. It could identify activity that looked interesting and it could help me review the roast, but I still treated first crack as something that needed human judgement or another authoritative signal.

Power measurement turned out to be useful too. Looking at the energy used by a roast added another piece of evidence when I was comparing sessions. Some later roasts were using around 300 Wh from start to cooling, which gave me another way of spotting whether two apparently similar roasts had actually been driven in a similar way.

None of this turned the Gene Cafe into a different machine. It did, however, teach me a lot about evidence.

What I still like about the Gene Cafe

After all of that, I still have a lot of affection for it.

It is approachable. You can see the coffee. You can hear it. You can make meaningful changes without needing to understand a complicated control system before the first roast. For somebody starting at home, that simplicity has real value.

It also made me roast.

That sounds obvious, but it matters. I did not spend the first six months designing an elaborate roasting laboratory and never putting beans into it. I roasted coffee, made mistakes, drank the results, changed something and tried again.

215 roasts later, I know considerably more about coffee than I did when the Gene Cafe arrived. Most of that came from doing the work rather than from the specification of the machine.

I would also separate my heater experience from the basic roasting capability of the CBR-301. My reliability problems are real and they are part of my ownership story, but they are still one ownership story. I do not have enough evidence to turn that into a universal claim about every Gene Cafe.

Where I reached the limit

Eventually I wanted information the machine could not give me directly.

The biggest missing piece was bean temperature. I could see what the machine was doing, but I could not directly see the temperature history of the coffee itself in the way that modern roast logging systems allow. I also wanted a clearer view of environmental temperature and rate of rise.

The controls themselves were another limitation for the experiments I wanted to run. On the Gene Cafe I was effectively steering through target temperature changes and the behaviour built into the machine. That was enough to learn a lot, but it made it difficult to separate heat input, airflow and drum behaviour as independent variables.

I increasingly wanted to make one deliberate change and then watch what happened.

That is a very different requirement from simply wanting a machine that can roast good coffee.

And it is probably the main reason I ended up looking seriously at the Kaleido M2.

Why the M2 is the next step

The incoming machine is a Kaleido M2-C in the Artisan configuration, rated at 1800 W. On paper it gives me the things I have been trying to get closer to for quite a while, including bean temperature, environmental temperature, rate of rise, more explicit heater and airflow control, and a proper connection into Artisan.

I am deliberately saying on paper because I have not roasted on it yet.

I do not know how responsive the controls will feel in practice. I do not know what the probes will tell me during the first few roasts. I do not yet know how repeatable it will be at the batch sizes I normally use, or whether the small batch capability will turn out to be genuinely useful for sample roasting.

Those are questions I now get to investigate rather than assumptions I want to make before the machine arrives.

The plan is also different from the way I approached the Gene Cafe. I want to start with a fairly boring baseline roast, learn what the signals look like, repeat it and then begin changing one thing at a time. Heat, airflow and machine state need to become things I understand through evidence rather than controls I move because the graph looks untidy.

That will probably be slower initially, but I think it will teach me more.

The Gene Cafe material is staying

Moving to the M2 does not mean I am removing the Gene Cafe material from the site.

Quite the opposite.

The Gene Cafe section now represents a fairly substantial record of learning on one home roaster. The roast notes, the longer ownership pieces, the buying guide, the owner’s guide and the instrumentation work can stay as a reference for anybody who arrives here because they own a CBR-301 or are thinking about buying one.

I will also keep the distinction between those different pieces.

The article about whether I would still recommend the Gene Cafe after 100 roasts answers a different question from the buying guide. The owner’s guide is there to help somebody use the machine. The individual notes preserve what I was seeing at the time. This article is not meant to replace any of them.

It is just where the roasting journey changes direction.

What I am taking into the next machine

The most useful thing I am carrying forward is probably not a profile.

It is the habit of recording what happened.

Weight loss matters. Machine state matters. Total roast time matters. The cup matters. What I thought happened during the roast matters too, particularly when the data later suggests something different.

I have also become much more suspicious of conclusions based on one roast. A good result is useful, but if I cannot repeat it or explain what changed, I have not learned very much from it.

The M2 will give me more measurements, but more measurements will not automatically give me more understanding. If anything, there will be more opportunities to convince myself that a graph explains something when it does not.

So the next phase is going to start quite simply.

Learn the machine. Establish a baseline. Repeat it. Change one thing. Watch what happens. Cup the coffee. Then decide what question comes next.

The Gene Cafe got me far enough to know why I want to do that.

Browse all articles →