MaKoMo wrote:I wrote a short post "How close?" on how the numeric similarity measure in Artisan works.

Thanks...another rabbit hole!

I hear what you are saying in the last graph about RoR vs BT curve, but the discrepancy is due to the beginning of the roast not being aligned. Instead of aligning at drop, what if we align at charge? This would get the profiles off to the same start.

I, for one, currently believe the heat applied to the very beginning of a roast is critical to the subsequent development of that roast. Hitting DE at 3:00 or 3:30 makes a difference in how that roast will develop. In other words, a roast where DE occurs at 3:00 might be fully developed by 7:00 with only 1:10 DTR, whereas a roast that doesn't dry until 3:30 might be underdeveloped with 8:00 total roast time and 1:30 DTR.

I'm just throwing numbers out there, but I am definitely finding a link between high initial RoR/fast dry time and completely developed fast, light roasts.

All that said, I need to play around with this feature for a while.

I assume that charge temperature and total length of two roasts to be compared are about equal. Those are easy to measure and compare and would need to be added to the overall equation. I forgot to mention this in the post as I thought it is obvious. I also should have selected two profiles that are closer together in that respect, but wanted to have the reader spot that those are two different ones. The background profile here is some 15sec longer and thus I expect that those two roasts indeed taste slightly different despite the good CM results.

When comparing the two approaches, align by CHARGE vs align by DROP, under this precondition the later is better. This is mostly because the data we record with our bean temperature probes is too far from the actual bean temperature at the begin of the roast. Further, I assume that a discrepancy in the second half of the roast is more significant to the result then one in the first half. Note that if both profiles have the same length, aligning by CHARGE vs DROP does not make any difference. The CM measure is quantifying really small discrepancies under the above pre-conditions.

I've matched* roasts and achieved different flavor profiles precisely by varying the heat transfer characteristics of my roaster. In other words, the same BT profile but different drum speeds, airflows, and gasflows will certainly mess with the flavor of the beans. I've actually fantasizing about creating a variable-walled drum to give me one more bit of control (i.e. a drum that can act double walled at times and single walled at other times).

*my definition is 1 degree F or 5-10 seconds. Generally inter-bean characteristics (i.e. cupping consistency) seem to have a greater effect once i get to this range on my roaster.

I've wrapped my limited brain around your post, Marko, and I think I'm getting it. The CM tool is really neat. If this catches on, it might need a new name.

What you say makes sense, the RoR is clearly a poor indication of replicated roast. This can be seen, as you illustrated, by pulling up a roast and background roast concurrently and, using the shift tool in Roast> Background, shift one profile up or down 10-15 degrees. The BT curve moves, but the RoR does not because it is not "actual" temperature dependent, only "change in temperature" dependent.

Here are two curves. The first is unmolested:


The second has the BT raised 42*. Note the CM change.



Note that the RoR does not move.

But I think where Scott comes from, with his focus on RoR, is treating RoR like fine tuning a radio signal. He assumes we are already on the correct station. Once dialed in to that station, the RoR gives us far more detail about the data that is hidden in a straight BT curve. It's like looking at the BT under a microscope.

Another cool feature: by moving the background curve up and down by 1* you can dial in he absolute lowest CM possible given the two curves. In this case it is 4.7*C. Again, measuring only post DE, yes?



In the last graph I also aligned the curves by DROP instead of CHARGE, so it's easier to see what Artisan is actually measuring.

Now I have to set about determining my CM tolerance number for replicating a roast. Can I live with 2.8*, or do I need to be under 1*?

FWIW, these are two complete different coffees and roasts. I was just trying out the features.

I think SR was giving too much credit to the reader (as usual). One could nit pick and say "oh well I wouldn't replicate a roast of a bean from season to season", I mean you could and I have and with little tuning the chord was struck. But let's say you aren't the person who makes everyone wait because you don't know what to order (they will always get pumpkin spice) batch to batch needs to be matched with such little deviation in both ROR and temps to be considered by SR as well done. You could hold the bar lower but that is what he says is proper reproduction. Almico, I think you are getting confused between replicating a roast, perhaps from roaster to roaster/season to season vs reproduction of the same bean on the same day for the same client (maybe a different client).

In my opinion color times are more than possible, but a stretch from accurate and a long shot from precise. Could it fool my palate, probably. I have rarely to never tried to copy a profile based on color markers alone, but I have (lots of times) looked at ROR as a measure of heat applied (as strength) to mimic a roast and used that as once piece of a large puzzle.

In a side note I'm gona be roasting an 84 Guatemalan (super fresh), I'm gona see if I can bump it up to an 87.

Almico wrote:
But I think where Scott comes from, with his focus on RoR, is treating RoR like fine tuning a radio signal. He assumes we are already on the correct station. Once dialed in to that station, the RoR gives us far more detail about the data that is hidden in a straight BT curve. It's like looking at the BT under a microscope.

It was not Scott that suggested to compare RoR instead, but some of his followers. It is true that to some sense the RoR zooms into some details, however, to compare two profiles a numeric argument over the underlying temperature curve is (mathematically) stronger than one over the derived RoR data (as you observed yourself by moving the background curve up and down; a value calculated over the RoR would have been left unchanged here).

Almico wrote:
In the last graph I also aligned the curves by DROP instead of CHARGE, so it's easier to see what Artisan is actually measuring.

Note that you can change the lower limit of the second phase from the DRY end temperature or those defaults 150C to anything lower to have more of the curve covered by the criteria (up to the TP). Still I think the data before DRY is not reliable enough to support this. Note that I still agree that everything that happens before DRY can have a significant impact on the outcome, even what happens before CHARGE. Just that we cannot easily extend this numeric criteria to that part of the profile without computing nonsense.

@Makomo I think if you are going to look at a single mathematical approach it would be the area under the curve.

Sr. uses (I believe) ROR as a marker for how much heat the beans are bathed in, and secondly and indicator for moisture loss progression. Which I think makes sense. So using those to things to triangulate for the accuracy of a roast is, I dare say, intelligent.

But how do you extrapolate a sample roaster curve to a larger roaster? And create a profiling (data) approach? As I look at other profiles and try and understand the link my guess so far, is that ROR is a good place to start + AUC (with relation to bean mass and density), the temps can be "calibrated" the same but that is taking in consideration of the probe reading the energy the same (as an equal relation, e.g. My probe reads 480 yours 440 but we both hear 2C). All linking back to AUC as the base for calibration/coorelation/triangulation.

Not easy:



getting better



Even better



Artisan calculates the discrepancy by aligning DROPs, but I find setting it to align at DE helps me match the roast better. This way if I miss DE time by a few seconds, the BT curves still line up and I can run the rest of the roasts keeping them that way.

It is still very difficult to get drop time and drop temp perfectly matched.

.