Why we treat roasting like a system, not a recipe
If coffee roasting were a recipe, we would all be done by now.Follow the steps. Hit the numbers. Repeat.
But anyone who has roasted more than a handful of coffees knows that this breaks down very quickly. Not because people aren’t skilled, but because coffee doesn’t behave like an ingredient in a cookbook.
Roasting isn’t a recipe problem. It’s a systems problem.
The problem with recipes
Recipes are comforting. They suggest control. In coffee, a recipe usually means a roast profile: a curve, a set of targets, a sequence of events that worked once and should work again. And sometimes, they do.
Until they don’t.
- Change the harvest.
- Change the density.
- Change the moisture.
- Change the roaster.
- Change the weather.
Suddenly, the same recipe produces a different result. Not slightly different. Meaningfully different.
At that point, the recipe hasn’t failed. The assumption behind it has.
Variability is not a bug
Coffee is a biological product. It changes.Even when two coffees share an origin, a variety, and a process, they are not the same system. They carry differences in structure, chemistry, and behaviour that show up under heat.
Treating roasting as a recipe assumes stability. Treating roasting as a system assumes variability.
The second assumption holds up better over time.
In engineering, we don’t design systems by hoping conditions stay constant. We design them to behave predictably when conditions change. Roasting deserves the same respect.
What a system changes
When you think in systems, a few things shift.
You stop asking, “What’s the right profile?” You start asking, “What inputs matter, and how do they interact?”
You stop chasing exact curves. You start tracking ranges, tolerances, and responses.
You stop relying on memory. You build references that accumulate over time.
This is where data becomes useful, not as a replacement for judgment, but to make learning stick.
Data doesn’t kill the craft. It preserves it.
One common fear is that data-driven roasting removes intuition. In practice, it does the opposite.
By logging what happens, across coffees, machines, and time, we free intuition from guesswork. Patterns become visible. Assumptions get tested. Surprises get documented instead of forgotten. Craft isn’t about doing something once. It’s about doing it again, on purpose.
Systems make that possible.
Humans stay in the loop
No system roasts coffee on its own. Data can tell us what changed. Models can suggest what might matter. They cannot decide what tastes good, or what feels right for a coffee.
Roasters make those calls.
The goal of a system is not automation for its own sake. It’s support. Clear signals. Fewer blind spots. Better conversations between what we taste and what we measure.
Judgment stays human. Learning becomes collective.
Notes from our Lab Bench
We see this every time we rely too heavily on a single “working” profile.
A roast that performs beautifully on one coffee often drifts when applied elsewhere. Not because it’s wrong, but because the system underneath has changed.
What works better is defining intent, tracking behaviour, and adjusting within known bounds. When we do that, results stabilise, even when the inputs don’t.
The most useful data points aren’t the perfect curves. They’re the moments where things didn’t behave as expected and we had enough context to understand why.
That’s where learning compounds.
Where this leaves us
Roasting isn’t about finding the perfect recipe.
It’s about building a system that behaves well across difference, change, and time.
At Heart of Coffee, that means fewer magic numbers and more shared understanding. Fewer one-off successes and more repeatable outcomes. Less reliance on memory and more on accumulated knowledge.
Recipes are easy to share. Systems are harder to build.
We prefer the harder work.
English 
French