It’s been said that those who stand for nothing will fall into a constant state of decline.
As an outspoken critic of “my way or the highway” approaches to education and one-size-fits-all solutions to roasting and quality control concerns in specialty coffee, it may not surprise you to learn that I take issue with the popular strategy in certain roasting circles of using a “constant state of decline” in a coffee’s so-called “rate of rise” (ROR), change rate, or (if we’re being vaguely mathematical) derivative. In some coffees and for certain roasting styles, this theory popularized by consultant Scott Rao works well. In others, it does not.
A roaster recently pointed me towards a webinar entitled “Why Rate of Rise is a bad reference point for optimizing flavour in coffee roasting,” hosted by Morten Münchow of CoffeeMind Academy. The two-hour recording is available on YouTube for you to view, and Münchow makes a number of good points.
Among these, he notes that rate of rise is already present in visible roast charts – that is, the X axis (time) and Y axis (temperature) already represent a roast’s derivative. He also notes that most live Rate of Rise readings present a lot of “noise” or interference to valuable data, and that the obsession with instantaneous ROR readings magnifies insignificant information.
Many other lucid points are made throughout the course of the webinar, but I want to pick apart a few specific items, not because I think Münchow or Rao are specifically right or wrong, but because I think they can sometimes be both at the same time. That is to say, the theories are contingent on a couple of presuppositions that aren’t always transparent.
Is Rate of Rise a Bad Reference Point?
The first and most important question to address is whether Münchow proves his point: is rate of rise a useful metric to observe?
My suggestion is that it can be. I agree with Münchow that looking at the finished curve and making judgements based on a visual representation of the rate of change over time is unlikely to be a particularly useful diagnostic tool.
However, when used to create a roast plan, or “profile,” and follow it consistently, observing incremental readings in the temperature change rate and applying them in a predictive nature can be helpful.
Münchow admits this outright at about minute twenty:
You can use [ROR] for example if you know you’d like to hit first crack at 9 minutes and you’re at 6 minutes, and you’re thirty degrees off… if you get first crack at 200, and you’re at 170, and you’ve got 12% rate of rise, and you’re rate of rise is going down, it seems like you’ll pretty much be there, right? So, you can use it to see if you’ll be… just like the speedometer on your car, you can use it to see if you’ll be early, late, or on time for your meeting. So you can use it to see if you’re going the right direction with your roast. And I’ll explain later that I’d never improvise, that I’d never make a decision based on it, and that’s why I don’t want to obsess about it.”
So, if ROR might be used occasionally, but isn’t worthy of decision-making processes, is it actually useful? Münchow’s argument is that it’s not, and his reasoning for never making a decision based on ROR is boiled down to a few simple points, which I’m about to dissect, and maybe slightly eviscerate.
The ROR visual curve is unnecessary (the data already exists on the bean temperature curve)
One of the central tenets of Münchow’s argument, and one which is repeated frequently over the course of the two-hour session, is that the ROR data is already a part of the bean curve. This is true, and you can ask a mathematician with the most basic understanding of calculus to confirm.
When I roast, I typically ignore the visual ROR curve and instead focus on the number itself. In Münchow’s example above, if I were roasting a coffee with a target first crack of 9:00 at 200C, and my current time and temperature are 6:00 at 170C, I have 3 minutes to make 30C of progress towards my goal.
Let’s use that car speedometer analogy here to help clarify. In Münchow’s analogy, ROR is the speedometer (though he later conflates it with navigation). The car itself is our current position on the roast curve, and the meeting is a destination in the future at a given distance away – our time and temperature of first crack, e.g.
If all we know is the position of our car on the road and the time of day, we have insufficient data to predict our arrival time. If we know where the car was located one minute ago, and the amount of distance travelled, we can infer our rate of travel.
Münchow’s argument seems to be that he’d rather drive without a speedometer, and constantly look back over his shoulder to calculate how far he’s traveled while checking his watch for the time that’s passed.
Put another way, can a roaster infer the trajectory from 6:00 to 9:00 in 30C? Sure. Is a ROR readout strictly necessary to predict this? No, it could be calculated mentally and can be parsed visually from the temperature curve.
But in the moment, in front of a live roast, is it useful to have that predictive number on hand to help you make decisions about your roast (rather than using precious seconds to do mental gymnastics, especially if your numbers don’t conveniently line up to full minute integers in 10C increments)?
I think so.
The “constant decline” of ROR is a natural phenomenon in the presence of a consistent heat source.
Theoretically, this is true… but there are two massive caveats in coffee.
The first is that the typical roasting system doesn’t use a consistent air temperature or heat source to craft a nicely roasted coffee. Operators manage these variables specifically to avoid the phenomenon of over-roasted coffee after first crack. If you’ve ever left your burner at 100% all the way up until first crack, this will probably ring a bell – coffee roasting becomes unmanageable, and roasts taste bad when roasted in a “set and forget” style.
The second (related) reason ROR isn’t naturally constant in coffee roasting is that the way coffee responds to heat changes significantly as it roasts, particularly at and after first crack. You can see this represented on Münchow’s own graph of a constant burner setting on the Aillio Bullet, where the ROR dips and then rises again towards the end of the roast.
The phenomenon is repeatable on machines capable of producing constant air temperature, like the new Sivetz, where a computer-controlled inlet temperature produces the same results (below)
Many roasters will be familiar with the phenomenon of bean probe and exhaust temperatures “swapping” under certain roast conditions, where coffee appears to essentially “roast itself” and absorb heat more rapidly as it loses structural stability, releases moisture, and expands at and after first crack.
So, no, absent the presence of a skilled operator, rate of rise will not simply take care of itself.
Roasters should use bean temperature probes as “artifacts” to build “intuition” about when their coffees will reach first crack
The most cringe-worthy eye-roll revelation of the webinar is that Münchow’s advice for ignoring ROR is to rely on intuition with a bean probe.
The idea that the bean temperature measurement is an artifice is mostly true – no two probes or machines measure it exactly the same. And yes, you must build this into your roast plans.
But, for someone so obsessed with the importance of science to understand flavor control, to simply advise roasters to “figure it out” based on intuition is utterly jaw-dropping.
There’s a tool to do that for you. It’s called rate of rise. It’s more reliable than intuition, particularly in training new roasters.
The ROR is full of visual “noise,” i.e., peaks and valleys that give momentarily inaccurate data
Münchow’s argument is that a snapshot of ROR is also not very useful because it could be too “noisy” or that it magnifies something out of proportion. He seems particularly obsessed with debunking the “constant decline” theory based on some visual “static,” which I think misses the point.
This problem (if it exists) of overemphasizing momentary blips in ROR can be easily sidestepped by monitoring ROR over time during an approach to a target.
Münchow discourages “jiggling the gas… as if it’s going to make a difference” and the point here is well taken. A nicely planned roast profile should probably include a couple of decisive burner adjustments rather than a complicated heat profile that’s difficult to reproduce with precision.
That said, using the full data at our disposal, including the air temperature, bean temperature and the rate of rise can help us make useful predictions about our roasting success.
Only 5% of Roast Flavor Determined by Pre-First Crack Development
Münchow cites excellent research intended to prove that roughly 80% of a given roast’s flavor is determined by end color, and only 20% of the “overall timing of roasting.” Of that, the study claims, only about 5% is related to the amount of time it takes to reach first crack.
In roasts where color variation is a factor, this principle is absolutely true. Anyone used to cupping a table full of sample roasts will quickly note “too dark” or “too light” as grounds for reroasting and reevaluating to give the green coffee a fair shot at quality evaluation.
However, as Münchow admits, if the end color of two roasts is the same, then roast style has everything to do with what makes a difference in flavor during roasting. The point I think he’s trying to make is that to have a scientifically relevant comparison between roasts requires precision in color goals, but the unintended consequence of this observation, and the way it’s conveyed, is to minimize the role of the roaster.
If your roasting brand (as is true with many specialty roasters) focuses on light roasts within a relatively narrow spectrum of end color, then you better damn well pay attention to how you’re getting there because the quality of your roasts depends on it.
If all you care about is whether or not a customer can tell if your coffee is light or dark, well, then by all means follow Münchow’s advice and ignore everything that happens before first crack.
Overall, I enjoyed the webinar, and I particularly thought the sensory comparison of underdeveloped, scorched, and baked roasts was very relevant. Cuppers, myself included, are rarely as good as we think we are, and it’s critical to acknowledge that the average customer is less concerned with the nuances of roast styles than we often are. However, if some really exciting stuff happens in the little window of Maillard Reactions before first crack, I’m more than comfortable operating within that 5% flavor window to make good coffee great. Stay tuned for more on this topic in a future “Technically Speaking” Column in an upcoming issue of Roast Magazine.
Obsession with a single variable rarely results in good practice, and I appreciate the light Münchow shines on this. During writing, I asked our roaster Doris Garrido to look over this work. Her response was less about what I’d written and more of a meditative reflection. I’ll leave you with it:
I believe that what I see in a curve is just a shadow, like the man in the cave. What I believe is happening inside is not the whole truth, just a reading of temperature around the probes. I would never follow the visual rate of rise (which can be named more accurately “average rate of change” because is not a straight line). I’ll just check the number as like an speedometer. I picture every coffee as a different road and I would drive in different ways.
I suspect a lot of Morten’s views of RoR come from his “plan it and execute it” style of roasting where you set a heat, drum, fan plan in advance and then execute it without trying to steer the roast if the plan goes awry. His rationale is that by the time you see a problem, it is too late to fix it, at least on larger drum machines. This is not true on a Bullet or a small fluid bed air roaster. My Artisan 3e is PID’s to precisely match a roast curve in Artisan scope (as long as I set PID params for a given batch size and mind the loft). On the Bullet I’ve been able to use power to manually steer the roast to match a background curve very closely. When steering a roast, I find RoR useful. I also find it useful when diagnosing a roast that didn’t hit my desired curve or key time points. I’m still not convinced about the need for a linearly declining RoR or whether a crash and flick makes a difference if you control for end color.
Thanks for the thoughtful response, and for taking time to read. I agree with much of what you’ve said. I think with a nimble enough machine, or with a steady hand, even large production batches can benefit from a measured use of ROR control; I found it helpful for end roast strategy especially in the 90kg Gothots I used to work on. And, I agree with your final sentence wholeheartedly. Cheers!
I appreciate your stance on ROR. I had the privilege in seeing you present at Roast Summit where you corrected Scott Rao’s stance on roasting curves. I like that you are able to take an objective viewpoint on this industry and provide deep insight on roasting. Would love to pick your brain more if you have the time.
Thanks for taking the time to read and comment Bryan. If you’ll be in Portland for Expo I’d love to chat, otherwise feel free to reach out directly and we can find some time to talk shop. Cheers!
Hi Chris, the link to Morten’s YouTube has a small error; it prompts for a “mailto”.
There’s also a shorter version available: https://www.youtube.com/watch?v=zrscIUhii-U
or a blogpost: https://coffee-mind.com/why-rate-of-rise-is-a-bad-reference-point/
or as podcast: https://coffeemind.buzzsprout.com/1897439/11992254-why-rate-of-rise-is-a-bad-reference-point-for-optimizing-flavour-in-coffee-roasting
Thanks for your blog. (Coffee-)Mind-boggling stuff, but I appreciate it (including Morten’s work) very much!
Thanks Norman! Link updated in the text.
I appreciate you sharing this blog post. Thanks Again. Cool.