Coffee batch size is the mass of green coffee charged into a roaster, judged against that machine's rated capacity — and it changes the roast so fundamentally that a profile which works at one batch size will not simply work at another. Halve the load and you have not made a smaller version of the same roast. You have changed the thermal mass in the drum, the ratio of beans to hot metal and hot air, and the way the probe reads temperature. The curve comes back different, and it comes back different for reasons you can predict.
This is the most common frustration among people who move between machines or between production sizes: the numbers were copied faithfully, the gas settings were matched, and the coffee tasted wrong anyway. Batch size is not a scaling factor you multiply through. It is a variable in its own right.
What coffee batch size means and how it is measured
Batch size is normally expressed as a fraction of a roaster's rated capacity: a two-kilogram charge in a machine rated for four kilograms is a 50% batch. That rating describes what the drum can physically hold and what the builder believes it can heat, which is not always the same thing, and ratings are not standardized across manufacturers.
The commonly repeated working range for roast batch size runs from roughly half of rated capacity up to the full rating, and sources differ on where inside it the comfortable middle sits. The most frequently quoted figure comes from roasting writer Scott Rao, who has written that many — if not most — machines produce their best coffee at somewhere between 50% and 70% of nominal capacity. Other roasters place the useful band higher. Experienced production roasters routinely run right up to the plate on machines with enough burner and enough airflow to support it, and plenty run well below half capacity on purpose for sample work and profile development.
Treat all of those numbers as guidelines rather than laws, because none of them is a property of roasting in general. The usable range belongs to your specific machine — its drum mass, its burner output, its airflow range — and the only reliable way to find its edges is to roast at them and taste the results.
The core idea: thermal mass and ratio
A drum roaster is a heat-exchange system with three actors — the hot metal of the drum and its structure, the hot air moving through it, and the green coffee. Green coffee is the heat sink. It arrives cool and damp, absorbs energy for several minutes before it gives any back, and its mass determines how much energy the system must supply to carry the whole load through drying, browning and first crack.
Change the mass and you change two ratios at once.
- Bean-to-metal. A bigger charge means more beans per unit of drum wall, so any single bean spends proportionally less of its time in direct contact with hot steel — but the bed is deeper, so beans buried in the pile are shielded from the airflow, and the ones at the wall may sit there longer before the tumble lifts them clear.
- Bean-to-air. Airflow through a machine is broadly fixed by its fan and its restriction. Push less coffee through the same airflow and each bean receives a larger share of the convective heat and a larger share of the chaff-clearing draft. Push more coffee through it and the same air has to do more work.
Everything else in this article follows from those two ratios shifting. The machine's own stored heat — the mass of the drum itself, which does not change when the charge does — becomes proportionally larger or smaller relative to the coffee it is heating. That is the whole mechanism in one sentence.
What under-filling does
A small batch in a large drum is dominated by the machine rather than by the coffee. There is less thermal mass to absorb the heat already stored in the steel, so the load recovers quickly after charge and climbs fast. That sounds like an advantage, and sometimes it is, but it arrives with characteristic problems, and they are the reason small batch coffee roasting is a skill of its own rather than an easier version of the same job.
- The roast gets twitchy. With less mass to buffer it, bean temperature responds sharply to small changes in gas or airflow. An adjustment that would nudge a full batch will shove a light one, and holding steady momentum through the middle of the roast becomes harder rather than easier.
- The probe reads less reliably. A bean probe measures whatever it is immersed in. If the bed is shallow enough that fewer beans wash over the sensor, it spends part of each rotation reading air rather than coffee. Readings run noisy, often low, and comparisons against your full-batch logs stop meaning much. This is the most under-appreciated consequence of a light charge.
- Scorching risk on hot metal. A thin bed leaves individual beans in contact with the drum wall with less coffee around them to share the conducted heat, so an aggressive charge temperature can mark bean faces before the load has built any thermal inertia at all.
- Bean movement can go wrong at the extremes. Very light loads sometimes tumble badly, skating across the drum rather than folding through a bed, which undoes the evenness a small batch should otherwise deliver.
Sample-roast-sized batches sit at the far end of this spectrum by design — a charge measured in tens or a few hundreds of grams, on a machine built around that mass, with a probe and an airflow path sized to match. That is exactly why a sample roast is a good screening tool for green coffee and a poor template for a production profile: it is not a small version of your production roast, it is a different thermal system.
What over-filling does
Push past what the burner and fan can support and the machine stops leading the roast. The coffee does.
- Sluggish recovery. A heavy charge pulls the system down further and holds it down longer, and if the heat source cannot replace that energy quickly enough, the whole curve flattens. Roasts stretch out, momentum decays through the middle, and the classic outcome is a flat, papery, "baked" cup where the aromatics never really arrive.
- Poor bean movement. A drum that cannot properly lift and fold an overloaded bed leaves beans sitting where they are. Beans also expand as they roast, so a drum that is full at charge is fuller still by first crack.
- Uneven roasting. The outside of the bed and the interior of the bed do not experience the same environment. You end up with a batch that is the average of two roasts rather than one consistent one, which shows up as a muddled, unresolved cup even when the average color reading looks correct.
- Conduction defects. Scorching — a burnt flat face where a bean has sat against over-hot metal — and tipping, the small burnt marks at the ends of the bean that are usually attributed to heat being driven in faster than the bean can absorb it, both become more likely when beans are pressed against the drum and agitation is poor. Some roasters also distinguish facing, a related mark from prolonged contact with the drum wall; the vocabulary is not used identically by everyone, but the underlying cause is the same one — too much conducted heat, too little movement.
- Exhaust struggling. The same fan now has to clear more chaff, more moisture and more smoke from a longer roast. Chaff that does not leave can smolder; moisture and smoke that linger can leave a heavy, ashy character on the coffee.
Why turning point and probe response move with the load
Two instrument behaviors shift the moment you change the charge, and both are easy to misread as machine faults.
The first is the turning point — the low mark where the falling probe reading stops and begins to climb. On most drum machines a larger charge pulls the reading down further and takes longer to bring it back, so the turning point tends to land lower and later, while a smaller charge lands higher and earlier. How pronounced that is depends on probe placement and on how much stored heat the drum is carrying, so the size of the shift is worth measuring on your own machine rather than assuming. Either way it is expected behavior, not something to correct, and it means turning-point targets are only comparable between roasts of the same size. The mechanics of that moment are covered in full at coffee turning point.
The second is rate of rise. The height of the RoR curve at its start and how gracefully it declines are both functions of how much mass the heat is working on, so a batch-size change rewrites the curve you have been reading for months. Rate of rise explains what that measurement actually tracks and how to interpret its shape.
Underneath both sits the probe itself. Immersion, position and response speed all interact with bed depth, and a sensor that is well buried at 80% of capacity may be intermittently exposed at 40%. When people say a small batch "reads strangely", the coffee is usually behaving fine and the instrument is the thing that changed.
At a glance: small, working-range and over-filled batches
| Behavior | Small batch (well under capacity) | Batch in the working range | Over-filled batch |
|---|---|---|---|
| Recovery after charge | Fast; turning point higher and earlier | Predictable and repeatable | Slow and deep; turning point lower and later |
| Momentum through the roast | Twitchy; responds sharply to small changes | Steady; adjustments behave proportionally | Decays; the curve flattens if the burner cannot keep up |
| Evenness | Generally good, if agitation is adequate | Best case for the machine | Poor; bed exterior and interior roast differently |
| Probe reliability | Reduced; the sensor may be partly exposed | Good immersion, trustworthy readings | Well immersed, but no longer representing a uniform bed |
| Typical faults | Scorching, over-fast development, noisy data | Faults come from profile choices, not from the load | Baking, tipping, facing, retained chaff and smoke |
| Chaff and exhaust | Cleared easily | Within the fan's design range | Fan works harder; retention and smoky character |
Scaling a profile means re-developing it, not copying it
The practical takeaway is blunt: when the batch size changes, times and temperatures do not carry over. What can carry over is the intent behind the profile — its relative shape, the proportion of the roast spent drying versus browning, and roughly how much development you want after first crack. Everything numeric underneath that has to be rebuilt.
A workable approach looks like this. Decide first what you are holding constant; most roasters hold the phase proportions and the post-crack development, because those are what the cup responds to most directly. Then adjust charge temperature and early gas to land the turning point and the early climb where you want them, expecting to raise charge temperature for a bigger batch and lower it for a smaller one. Then roast, log, taste, and adjust again. Needing two or three iterations to reach a stable profile at a new batch size is normal, not evidence that something is broken.
The end of the roast deserves the same skepticism. A given drop temperature does not represent the same degree of development across batch sizes, because bed uniformity and probe immersion have both changed — color measurement and tasting are the arbiters, not the number on the display.
Honest limits: all of this is machine-dependent
Almost every statement above carries a "depends on the machine" behind it. Drum thickness and material determine how much heat the roaster stores, and therefore how strongly a small charge is dominated by it. Burner sizing determines how far you can push a large batch before recovery collapses. Fan capacity determines how much coffee the airflow can serve. Two roasters with identical ratings can have genuinely different usable batch ranges, and the differences between drum, fluid-bed and hybrid designs change the picture again — coffee roaster machines covers how those architectures differ. Fluid-bed designs in particular tie batch size to airflow rather than to drum-wall contact, which changes which faults you are guarding against.
Nor is a bigger charge automatically less controllable. A well-specified machine running near capacity can be more stable than the same machine running a quarter load, precisely because the extra thermal mass smooths out the noise. The problem is never batch size in the abstract. It is a mismatch between the load and what that particular machine can deliver.
The bottom line
Coffee batch size sets the thermal mass of the roast and the ratios of beans to metal and beans to air, and those ratios drive nearly everything you subsequently measure. Roughly 50–100% of rated drum capacity is the range most machines work within, with the sweet spot usually somewhere in the middle and the true edges specific to your equipment. Under-fill and you get speed, twitchiness and less trustworthy data; over-fill and you get sluggishness, unevenness and conduction faults. Keep batch size consistent within a profile, log it alongside every roast, and when you do change it, treat the new size as a new profile to be developed rather than an old one to be copied.