coffee/fermentation · 15 min

coffee/ · 15 min

What fermentation actually does

Every route from cherry to green bean runs a fermentation. The only differences are what the microbes are given, and how long they get.

coffee/fermentation -> maind7b626d · 24 Sep 202629 sourcesask your agent about this post
The meal on the seed

00 read this with your agent

Read this with your agent first.

Copy one of these, paste it into whatever AI you use, and it will teach you from this page and only this page, at your level. An agent in a browser that supports WebMCP can use the page’s tools directly.

Teach me from scratch
What does 'anaerobic' on a bag mean?

01 the meal on the seed

There is a meal on the outside of the seed, and something will always eat it.

Every coffee seed comes wrapped in a sweet gel, and microbes start eating it the moment the fruit is picked.

Coffee is a fruit. Take the fruit off and you are left with a seed in a sugary slime, sitting in warm air full of hungry things: microbes, the living things too small to see.

A ripe cherry has a red skin, a sweet pulp, and, wrapped tight around each of its two seeds, a thin layer of clear jelly. Under the jelly is a papery hull called the parchment, and under that is the seed we roast. The full anatomy belongs to the natural process post; those five words are enough for this one.

The jelly is called mucilage, and it is the whole subject here. It is mostly water, about 842 grams in every kilogram, with a little protein and sugar.1 What gives it its body is pectin, the stuff that makes jam set: long chain molecules that glue plant cells to each other. An enzyme, a protein that does one chemical job, that cuts those chains is called a pectinase. Who supplies it is one of the oldest arguments in the field.

What the word fermentation is doing here.

Bread, yoghurt, beer, kimchi: the biology under all of them is the same. A living cell would rather burn sugar all the way down to carbon dioxide and water, but burning needs oxygen. When oxygen runs short, the cell takes a shortcut and stops halfway, and leaves the rest behind: alcohol, lactic acid, acetic acid, and a few hundred other things in trace amounts. Those leftovers are what we taste.

So here is the claim this whole post rests on. After picking, the sugar is there, the water is there, and the microbes are already on the skin. Fermentation is not an optional step. It is what happens by default. The choices a producer makes are choices about who eats it, for how long, with how much air and how much water.

fig 1The substrate, and the one enzyme that matters. The bar is one kilogram of mucilage in proportion, from source 1; the lens is a diagram, not a micrograph. Hover or tap the parts.diagram
For the enthusiast: the full recipe of the gel, and who really cuts the pectin

A kilogram of mucilage is about 842 grams of water, 89 of protein, 41 of sugar, 9 of pectin and 7 of minerals. It is not much material, but there is a lot of it: the mucilage and the soluble sugars that come with it make up around a ninth of the whole fruit by weight.1

Coffee pectin is heavily decorated with small methyl groups along its length, which matters, because a decorated chain is harder for an enzyme to cut than a bare one.1

The pectinase argument. A 2002 study isolated the pectin-cutting bacteria from fermenting coffee, measured their enzymes, and concluded that microbial pectin breakdown does not happen at all during fermentation, or is negligible: the enzymes it found worked best at pH 8.5, and a fermentation runs from 5.3 downwards.18 Two decades later a study that removed things rather than adding them found significant pectin-cutting in the first twelve hours coming from the coffee fruit's own enzymes, and then complete breakdown only when yeasts were present, with bacteria playing no critical part.9 Both can be true. The first looked at bacteria and the second found the activity in yeasts, and the fruit itself was doing some of the work all along.

02 who turns up

Nobody adds the microbes. They are already on the fruit.

Coffee is fermented by whatever already lives on the fruit and the farm, sorted into five working groups.

This is the part that surprises people who have brewed beer. There is no packet.19 Unless a producer steps in, a coffee fermentation is a spontaneous mixed culture: whatever was living on the cherry skin, the picker's hands, the soil, the pulping machine and the tank walls, all growing at once and competing for the same sugar. There are a lot of them: one survey of fifteen Brazilian farms counted a median of about sixteen million microbial cells on a single cherry.2

Five groups do the work. Yeasts are single-celled fungi, cousins of the one that raises bread. Lactic acid bacteria are the family that makes yoghurt. Acetic acid bacteria make vinegar, and need air to do it. Enterobacteria are common environmental bacteria that arrive on the fruit. Filamentous fungi are moulds, growing as threads.

fig 2The cast, each drawn in the ink it wears on every figure below. Drawn to be told apart, not to scale. Hover or tap the parts.diagram
For the enthusiast: the species, the counts, and what DNA sequencing changed

The Brazilian survey ran over two years and counted between thirty thousand and 2.2 billion microbial cells on a single cherry. Of 754 isolates, 626 were identified, making 44 genera and 64 species, and the authors concluded that the flora of a dry process is considerably more varied and complex than what is found in wet fermentations.2 The unit used is CFU, colony forming units, which just means "cells that were alive enough to grow into a visible colony on a plate". Counts are usually written as logarithms, so 5 log CFU per gram means a hundred thousand per gram.

Our picture of who is present changed sharply when sequencing arrived. In one Australian study, the old method of growing things on plates found 6 yeast species and 17 bacterial species; reading the DNA directly out of the same samples found 212 fungal and 40 bacterial species.3 A Colombian fermentation read the same way turned up 160 bacterial genera across 10 phyla.4 Most of that long tail is doing nothing much. But it means that any sentence beginning "coffee fermentation is carried out by" is a statement about who dominates, not about who is there.

YeastsSaccharomyces, Pichia, Candida, Hanseniaspora, Debaryomyces

Single-celled fungi that reproduce by budding. They are the strongest pectin cutters in the tank and the main source of the fruity smells. Hanseniaspora uvarum and Pichia kudriavzevii dominated an Australian wet fermentation;3 Pichia nakasei dominated a Colombian one from start to finish;4 in a Brazilian natural process the commonest were Debaryomyces at 27 percent of yeast isolates, Pichia at 18.9 and Candida at 8.5

Lactic acid bacteriaLeuconostoc, Lactobacillus, Lactococcus

They eat sugar and make lactic acid, which drops the pH and shuts most competitors out. In a Colombian tank they held over 60 percent of the bacterial community at every single sampling point, with Leuconostoc peaking at 84 percent after 24 hours.4

Acetic acid bacteriaAcetobacter, Gluconobacter

They turn alcohol into vinegar, and they need air to do it. That single requirement is why they fade away in a tank of water and take over on a drying bed. They were the characteristic group of dry processing in the study that ran both routes side by side.6

EnterobacteriaEnterobacter, Citrobacter, Erwinia, Klebsiella

They arrive with the fruit and are the most numerous thing in the tank at hour zero. They are out-competed within a day as the acid builds, and in the wet route their counts fall continuously from the beginning.7

Filamentous fungiAspergillus, Penicillium, Fusarium, Cladosporium

They need time and air, so they belong to the long dry routes. Two Brazilian studies by the same group disagree about which one dominates: the farm survey found Cladosporium, Fusarium and Penicillium each at about a third of the fungal isolates with Aspergillus at only 3 percent,2 while the fermentation study put Aspergillus at 42.6 percent.5 I cannot reconcile those, so both are here. The risk side of moulds is the natural post's subject, not mine.

And sometimes, one you chosea pitched starter culture

A producer can add a known organism at the start. That does not add new chemistry; it changes the starting line, which is section 06. What the practice looks like on a farm is the honey and experimental post's subject.

03 the order they arrive in

The order matters more than the roster, and acid is the referee.

The microbes take turns, and the acid the winners make is what keeps the spoilers out.

Nobody wins by being best. They win by going first, or by surviving what the first ones leave behind. Ecologists call this succession: one community changing its surroundings enough that another can replace it, the way a burnt forest grows back through grass, then scrub, then trees. In a tank it takes a day instead of a century.

At hour zero the enterobacteria lead, because they were the most numerous thing on the fruit. Within a few hours the oxygen in the water is used up, and everything that needed air is in trouble. The yeasts pull ahead and cut the pectin, turning sugar into alcohol and carbon dioxide. Then the lactic acid bacteria overtake everybody and keep the lead to the end.7

And the tank turns sour. pH is the scale for that: the lower it goes, the more acid. A Colombian tank went from pH 5.2 to 4.2 over two days, and as the acid built, the lactic acid bacteria held back the spoilage organisms.4 They are not just making a flavour; they are making the tank uninhabitable for everything that would ruin it. A long fermentation is not dangerous because it is long. It is dangerous when the acidification fails.

The mucilage is gone within about a day and a half.9 After that the microbes are still alive and their food is not, which is where section 09 begins.

fig 3This figure is the post in one interaction. Scrub the hours and read who is winning, what the acid is doing and what the cup is becoming. Curves are composited: the endpoints come from sources 3, 4, 6, 7, 8 and 9, and the shapes between them are modelled. The pH marks are measured values from two separate ferments, sources 4 and 6, not a curve. The population axis shows order only: the counts behind it are in different units, yeasts about 5.5 log CFU per gram and lactic acid bacteria about 5.2 log CFU per millilitre, and cannot share one scale.composite · interactive
For the enthusiast: the counts, a second pH series, and where the curves come from

The sequence is drawn from the studies that sampled a tank hour by hour. Yeasts grow to something around 5.5 log CFU per gram.8 In the Belgian group's Ecuadorean trial the lactic acid bacteria achieved what the authors call quantitative prevalence over every other group and kept it to the end.7

An Ecuadorean trial measured pH 4.5 after 16 hours of fermentation and 4.0 after 36.6 In the Australian work the mucilage was completely degraded by the end of a 36 hour fermentation.9

The figure composites those endpoints across studies (sources 3, 4, 6, 7, 8 and 9) and models the shapes between them. A count of cells is not the same thing as how much work those cells are doing.

04 what they make

Six things come out, and only two of them are acids.

The microbes turn sugar into acids, alcohols and, above all, esters: the molecules that make a coffee smell of fruit.

The tank does not smell of one thing. By the end of a long fermentation there are over a hundred different volatile compounds in the water, light enough to reach your nose.7 Nearly all of them are in trace amounts. Six families do the heavy lifting.

  • Lactic acid. Soft, milky, mildly sour: yoghurt rather than vinegar. The signature of the whole washed route.
  • Acetic acid. Vinegar, unambiguously. A little reads as brightness; a lot is the first defect most people can name.
  • Ethanol. Plain drinking alcohol, made by yeasts from sugar, and, it turns out, partly by the seed itself. Hold that thought for section 05.
  • Sugar alcohols. Mannitol and glycerol, both faintly sweet.
  • Higher alcohols and aldehydes. Heavier cousins of ethanol, with smells that run from banana skin and roses to green apple.
  • Esters. Where the fruit comes from.

And esters, which are where the fruit comes from.

An ester is the single most useful word in this post. Take an acid. Take an alcohol. Push them together and they join, and as they join a molecule of water falls out. The thing left over is an ester, and esters are what almost every fruit smells of. The acid on its own is sour and the alcohol is sharp; joined, they smell of fruit, and neither parent does. Acetic acid plus isoamyl alcohol gives isoamyl acetate: banana and pear drops.

fig 4The post's own example: acetic acid and isoamyl alcohol join, a molecule of water drops out, and what is left smells of banana and pear drops. Hover or tap the parts.diagram

That is why a coffee can smell of strawberry when there is no strawberry within a thousand kilometres. Nothing strawberry-flavoured was added. A bacterium made an acid, a yeast made an alcohol, and the two met.

The cleanest proof that the yeasts matter comes from an experiment that removed them. Natamycin is an antifungal: it kills yeasts and leaves bacteria alone. Beans from a tank dosed with it carried 25 times less of one fruity ester, ethyl acetate, than beans from a tank without it; the gap was still there after roasting, and that coffee scored lower.8

For the enthusiast: the concentrations, strawberry and furaneol, and the yeast experiments in full

Over 170 volatiles were found on the beans, about seventy percent of which were not there before the fermentation started.7

In the Ecuadorean trial lactic acid reached 8.2 milligrams per millilitre in the fermentation water after 64 hours and 2.2 grams per kilogram on the beans; acetic acid 2.4 milligrams per millilitre in the water and 4.5 grams per kilogram on the beans; ethanol 4.6 grams per kilogram on the beans. The authors argue a large share of that ethanol was made by the coffee seed itself, because a submerged seed with no oxygen does exactly the same chemistry a yeast does.7 Glycerol was present at 0.08 percent when yeasts were allowed to grow and was not detectable at all when they were suppressed, which makes it a clean yeast fingerprint.8

Higher alcohols are made when a microbe strips down an amino acid rather than a sugar. Isoamyl alcohol smells of banana skin and solvent; 2-phenylethanol smells of roses and honey. Aldehydes are a half-step further, sharper and greener, and acetaldehyde is the green-apple note you get in a fermentation that is still young. Acetic acid plus ethanol gives ethyl acetate, which is nail varnish in quantity and something like pineapple in trace.

Strawberry deserves its own sentence, because it is the note people find hardest to believe. The textbook strawberry molecule is furaneol, formally 4-hydroxy-2,5-dimethyl-3(2H)-furanone, which smells of caramel and is one of the key aroma compounds of many fruits: it has been isolated from strawberry, raspberry and tomato, and is a key odorant of pineapple. It is also a product of the Maillard browning reaction, which means a roaster can make it out of nothing fruity at all.26 Furanones are in the small set of compounds that Czerny, Mayer and Grosch identified as having the greatest impact on roasted arabica flavour, in the classic experiment where they rebuilt coffee aroma out of 27 pure odorants and then removed them one at a time to see what a panel noticed.25 So when a cupper writes strawberry on a scoresheet they are not reaching for a metaphor. They are naming a smell produced by a family of small molecules, most of them esters and some of them furanones.

In the Ecuadorean measurements, esters were about half of all the volatile compounds found in the fermentation water, alcohols about a quarter and aldehydes about a tenth, and the total aroma intensity rose fivefold between a 16 hour and a 64 hour fermentation.7

Two more sets of numbers, from studies that added yeasts rather than removing them. When Hanseniaspora uvarum and Pichia kudriavzevii were pitched into a wet fermentation, sucrose in the mucilage fell from 13.03 to 2.01 grams per hundred grams in the uninoculated control and to nothing at all where both yeasts were pitched together, fructose fell from 27.02 to 4.05 in the control, and esters came to about 30 percent of the total volatiles measured in the green beans.23 And a 2026 study profiling fermented against washed beans measured total esters at 74.5 milligrams per kilogram in the fermented set, with 2-phenylethanol, the rose and honey one, at 27.5.27

The natamycin experiment in full: green beans with 25 times less ethyl acetate, 21 times less isoamyl alcohol, 8 times less acetaldehyde and 3.7 times less ethanol, and those gaps were still there after roasting. The yeast-free coffee had only a mild fruity aroma and scored lower on every axis the panel used, 6.5 overall.8

Ethyl acetate, yeast present vs suppressed
25×more
Isoamyl alcohol, same comparison
21×more
Aroma intensity, 64 h vs 16 h ferment
5×higher
Volatiles on the bean that appeared during fermentation
70%

First two from source 8, last two from source 7.

05 does any of it get in

Now the hard question: how much of that actually reaches the seed?

Some of what the microbes make does get into the seed, but nobody knows how much of your cup it accounts for.

Everything in the last section happened outside the bean. The bean is in a box, and the box is not obviously open. This is the centre of the post, and the place where I am going to refuse to give you an answer, because the field does not have one.

Set the problem up properly. A molecule made in the fermentation water has to cross what is left of the mucilage, then the parchment, a dead fibrous hull, then the silverskin, a thin skin on the seed itself, and then get into the seed's cells. Then it has to survive washing, drying and a roaster at over 200 degrees.23 That is a long way to travel for a trace compound.

And the seed is alive. A green coffee bean will sprout if you plant it, and everything done to it during processing is, from its point of view, an emergency. A stressed seed does its own chemistry, and some of it makes exactly the same molecules the microbes make.7 So finding a compound on a bean does not tell you who made it.

Two columns, and both of them are right.

the case forThings get in, and we have watched them do it.

Researchers made three aroma compounds with a heavy form of hydrogen in them, so they could not be confused with anything the bean or the microbes made, and put them in the water around whole beans. All three turned up inside.10 The hull is not a wall.

the case for cautionLess than you think, and much of it was never microbial.

Once a bean is ground for testing, a compound stuck to its outside looks the same as one inside it. The seed makes some of the same things itself, and washing takes a lot of it back off.712

There are at least three explanations, and all three can be partly true at once: things diffuse in from outside; fermentation changes the ingredients the roaster later browns; or the living seed changes itself. The honest summary is that crossing happens and has been directly observed, and nobody has measured how much of the final cup it is responsible for. Those are two different questions and the literature routinely runs them together.

fig 5The only direct measurement of the crossing I could find. The circled 12 hour values are read straight from source 10. The paper also sampled at 0, 6, 24 and 48 hours, but only the 12 hour values are published in its abstract, so the arrows show only the direction it reports. The vertical scale is logarithmic so that all three fit, and the layers are not drawn to scale.measured points only · interactive
For the enthusiast: the evidence on both sides, point by point

the case forFive findings

  • A labelled tracer crossed and was counted. The three compounds were labelled with deuterium. At 12 hours, in the medium with yeast, the beans held about 11.2 micrograms of 2-phenylethanol per gram, 1.3 of isoamyl acetate and 0.2 of butanal.10 This is direct evidence and it is not ambiguous.
  • Glycerol was inside the bean, and only when yeasts were alive. In the natamycin experiment, lactic acid accumulated inside the beans at about three times the concentration in the yeast-active run compared with the suppressed one, and glycerol was measured at 0.08 percent with yeasts and not detected at all without them.8 Glycerol is a yeast product. It was inside the seed.
  • Mannitol ends up on the bean, and a bean cannot make mannitol. After the full process, including washing, soaking and drying, extended-fermentation green beans carried 1.4 times more mannitol and 5 times more lactic acid than the standard lot from the same farm and the same day.7 Mannitol is a microbial product.
  • Suppress the microbes and the roasted coffee changes. The natamycin experiment killed the yeasts and nothing else, and the resulting green beans held a fraction of the esters and alcohols. Crucially, the difference survived the roast, and the cuppers scored the two coffees differently.8 If nothing crossed, this could not happen.
  • Non-volatile compounds cross too. The same French group followed nonvolatile compounds into the bean under both laboratory and real farm conditions with two yeast strains, and measured the transfer rather than inferring it.11 A later paper models transfer into the bean and destruction inside it as two separate processes, which is what lets the first be real even where the second cancels it.22

the case for cautionFive findings

  • Adsorption is not penetration. The authors of the most detailed metabolite study say plainly that compounds at high concentration in the water could be adsorbed onto the beans or trapped between the endosperm and the parchment.7 Grind a whole bean for analysis and you cannot tell the difference between inside the cells and stuck to the outside.
  • The seed makes a lot of it itself. GABA rose tenfold during fermentation, and although lactic acid bacteria can make GABA, the authors conclude the majority of it must have been the bean's own, produced as a hypoxia and germination signal. They make the same argument for a large share of the ethanol.7
  • The parchment measurably resists. In the tracer study itself, parchment resistance significantly affected the transfer of 2-phenylethanol, and two of the three tracers were consumed inside the bean faster than they arrived, so their concentration went down after 12 hours.10 Arrival is not accumulation.
  • Washing takes a lot of it back off. Metabolite concentrations in the soaking water fell to a few percent of the fermentation water levels,7 and in a large parameter study soaking tempered the fermentation effects and standardised the green bean, whatever had been done before it.12
  • In the dry route, the inside barely moved. Following both routes at once, the outer layers changed clearly and the endosperms showed fewer changes.6
For the enthusiast: diffusion, the three explanations, and four reasons this is hard to settle

The mechanism usually offered for the crossing is plain diffusion: a molecule moves down its concentration gradient through the parchment and into the endosperm until the chemical potential on the two sides is equal. The most recent critical review of the field treats diffusion as the working picture, and says of the newer fermentation methods that they lack sufficient scientific evidence.24 What nobody has done is measure the seed's own chemistry before and after processing closely enough to say which compounds actually cross, and in which direction.

ADiffusion. Compounds made outside the seed move into it and are still there when it reaches the roaster. This is the one with direct evidence behind it,10 and also the one the marketing overstates.
BChanged precursors. Fermentation alters the sugars, amino acids and acids the seed carries into the roast, so different browning chemistry happens later. On this account the compound you smell never existed before roasting and nothing crossed anything.
CThe seed’s own metabolism. The seed is alive, and germination-associated activity begins inside it during processing.29 A long, warm, low-oxygen fermentation changes what a living seed does to itself, which is not the same thing as absorbing anything.7
01There is no control group. You cannot grow a coffee that has never met a microbe, so there is nothing to compare against. Even a lot whose mucilage is scrubbed off by machine has had hours of microbial life on it first.
02Three sources, one molecule. Ethanol can come from a yeast, from the fruit, or from the seed's own oxygen-starved metabolism. Without a label on the atoms you cannot tell which.7
03The instrument grinds the bean. Almost every analysis mills the whole bean and extracts it, which mixes what was inside the cells with what was stuck to the surface and what was sitting between the seed and the parchment.
04The roast is in the way. A difference in the green bean may not survive to the cup, and a difference in the cup may have been created in the roaster rather than in the tank. Section 07 is about exactly that.

My own reading, and it is only a reading: the thing that changed in the last ten years is not the answer to how much gets in. It is the discovery that the seed was never a passive object sitting in somebody else's soup.where I actually land on this

06 taking the air away

Two ways to edit the guest list: shut the lid, or bring your own.

Sealing the tank or adding a chosen microbe does not invent new flavours; it changes who gets to eat.

Sealed tanks, carbonic maceration, pitched cultures. What producers do with these, and whether it is craft or marketing, is the honey and experimental post's argument. What I owe you is the mechanism, and it is the same every time: you are not inventing chemistry, you are changing who is in the room.

Oxygen is the biggest single switch.

Anaerobic means without oxygen. Seal a tank and the microbes use up the oxygen within hours; the carbon dioxide they make then keeps the air out. The vinegar makers stop, because making vinegar needs oxygen. The moulds stop for the same reason. The yeasts switch from growing to fermenting, which means more alcohol and more aroma. The lactic acid bacteria barely notice. So a sealed tank belongs to the yeasts and the lactic acid bacteria, and that is the whole mechanism behind "anaerobic" on a bag of coffee. The seed notices too: sealed and short of air, it runs its own oxygen-free chemistry harder.7

Pitching a culture moves the starting line.

To inoculate is to add a known organism at the start, in numbers big enough to out-grow whatever was there; what you add is a starter culture. Beer and bread do this. Coffee mostly does not. The mechanism is competition, not novelty: a pitched lactic acid bacterium soured a Brazilian tank fast enough to cut the fermentation from 24 hours to 12.14 Yeast starters lean towards alcohols and bacterial ones towards acids, so choosing a starter is choosing between two flavour registers.16 And finding an organism in a good coffee is not evidence that it made the coffee good.17

Below, the same cherry down four routes on one clock. The routes themselves live in washed, honey and natural; this is only what the microbes get on each.

fig 6The same cherry down four routes, on one clock, so the only thing on show is how long the microbes get and what they are given. The equipment belongs to the sibling posts; this is the biology. Phase boundaries are typical rather than fixed and vary widely by farm and by weather, so the axis shows order and relative length only, with no hours.schematic · interactive
For the enthusiast: the one controlled study of sealed tanks, and the starter trials

Given air, a yeast prefers to burn sugar completely and grow; starved of air, it ferments, which means less growth and much more ethanol and more of the aroma compounds. Lactic acid bacteria barely notice, because most of them do not use oxygen anyway.

Controlled studies of this are thinner on the ground than the marketing would suggest. The one I found that varied it properly reported the best overall cup score at 96 hours and 38 degrees under carbonic maceration, and found bacterial diversity correlating positively with the sensory characteristics.13 One study, one origin, one protocol. Treat that number as a data point, not a recipe.

The lactic starter was Lactiplantibacillus plantarum, and the cut from 24 hours to 12 is a real operational result before it is a flavour one.14 Different organisms leave different fingerprints: sprayed onto Brazilian cherries for a dry process, Saccharomyces cerevisiae gave caramel notes and Candida parapsilosis gave apple and cherry.15 In a semi-dry process, four yeasts all produced caramel and fruity notes and the Saccharomyces came out best; DNA fingerprinting confirmed the added yeasts stayed dominant throughout, neither butyric nor propionic acid was detected in any treatment, and the panel found a caramel note in the inoculated coffee that was simply absent from the control.20 Run through natural and pulped natural processing of one Brazilian variety, fermenting 27 hours at 16.5 to 24 degrees, S. cerevisiae CCMA 0543 scored 84.75 and 84.92 on the hundred-point scale against lower uninoculated controls, with compounds present in the inoculated lots that were absent from the controls entirely.28 Yeast starters were better producers of volatile alcohols and bacterial starters better producers of acids, which is what you would predict from the biology.16

Two honest caveats, both from people who work on starter cultures for a living. A pitched organism has to beat a community that is already established and adapted to that farm, which is much harder than pitching into sterile wort. And a review of the whole field notes that most of the organisms isolated from spontaneous coffee fermentations turn out to lack the attributes needed to improve sensory quality at all.17

07 through the fire

A fermentation metabolite is not a cup note.

The roast burns off most of what the microbes made; fermentation mostly changes the ingredients the roast cooks.

A metabolite is anything a living cell makes as it eats. This is the step most writing about fermentation skips, and skipping it is how you get tasting notes that claim a specific yeast put a specific fruit in your cup. Between the green bean and the cup there is a roaster at over 200 degrees for several minutes; one trial here roasted at 225 degrees for seven.23 It is not gentle.

Some things go straight through. Lactic acid and some of the acetic acid survive, and are tasted as sourness and brightness. Some things are eaten. Free sugars and amino acids are the fuel for the browning reactions that make coffee taste like coffee, and fermentation changes how much of that fuel there is. And a lot simply goes. Ethanol boils off far below roasting heat, and many of the light esters are gone or rearranged long before the beans start to crack.

So the largest effect of fermentation on the cup is probably indirect: it does not add a flavour, it changes the raw material the roaster is browning. The fruit smell over a fermentation tank is not the smell that comes out of the roaster, and expecting it to be is the single commonest mistake in this subject.

fig 7Three fates, and the middle one is the largest. What the roast mostly gets from a fermentation is a different set of ingredients, not a set of finished flavours.schematic
For the enthusiast: what does make it through, and the experiment nobody has run

The effect is not zero, and the experiment that shows it is the natamycin one again. Isoamyl alcohol and ethyl acetate remained higher in the yeast-present coffee after roasting, not just before it, and the cup scores followed.8 Something makes it all the way. What we do not have is a study that labels a microbial compound, follows it into the bean, follows it through a roast and quantifies what fraction of the cup it accounts for. Until somebody does that, the chain is established at every link except the last one.

08 steering it

You cannot plant a microbe, but you can set the table.

A producer steers a fermentation through its conditions, and time matters most.

A producer is in the odd position of managing a workforce they did not hire and cannot see. Everything they do is indirect: change the conditions, and a different part of the community does well. Six dials, roughly in order of how much they move the result.21

fig 8Six dials in the order the control literature keeps arriving at, time first by a distance. Size shows the order only; nobody has measured the gaps. Hover or tap the parts.schematic

Time is the biggest lever by a distance. In the study that varied everything at once, how long the fermentation ran had the greatest effect on both the green bean and the cup, and long gave a fruitier, more acidic cup.12

One control deserves more attention than it gets: washing and soaking at the end. Soaking after fermentation tempered its effects and made the green beans alike, whatever had been done before.12 A producer chasing a fermented character and then soaking hard is undoing their own work.

And a reality check. Most fermentation on most farms is judged by hand and nose: you rub a bean and feel whether the slipperiness has gone, and you smell the tank. That is a perfectly good instrument in the hands of somebody who has done it ten thousand times, and it is not a thermometer.

For the enthusiast: the six dials one by one

Time

The biggest lever by a distance. In the study that varied everything at once, fermentation duration had the greatest impact on both green bean composition and cup quality.12

Temperature

Sets the rate and the roster at once. A tank high in the Andes cools overnight and slows right down; the same tank in a Brazilian valley does not. Controlled carbonic maceration found 38 degrees the best of the temperatures it tried.13

Water

Under water, heat spreads out and oxygen runs out. On a bed, water activity, how available the water is to a microbe, falls and shuts groups down from the outside in. In one Brazilian natural, the beans started the fermentation at about 68 percent moisture.5

Oxygen

Present or absent decides whether the vinegar-makers and the moulds are in the competition at all. Section 06.

The vessel

Open tank, sealed drum, plastic bag, the floor of a mill. The vessel decides the oxygen, the temperature stability and, importantly, which microbes were living on it before you started.

pH, and what you add

Acidity falls on its own, but it can be pushed. Adding pectinase and lactic acid broke the mucilage down in under eight hours instead of thirty-six.9 A pitched culture does the same thing biologically.

In the soaking water itself, metabolite levels fell to a few percent of those in the fermentation water.7 A producer who wants consistency across lots has a very effective tool in soaking, and may not know it. And any discussion of dialled-in fermentation control is describing a small and well-capitalised slice of the industry.

09 when it goes wrong

Where a style becomes a fault, and who gets to say.

Left too long, a ferment runs from fruit to vinegar to cheese and onion, and people disagree about where style ends and fault begins.

The chemistry is a gradient. The line across it is drawn by people, and they do not all draw it in the same place.

Once the sugar has gone, the community starts on whatever is left, and what comes out gets worse the further past that point you go. First the fruity esters peak and fade. Then acetic acid keeps climbing, and the cup goes from bright to sharp to plainly vinegary. Then propionic and butyric acid appear: sweaty cheese and rancid butter, which together are what people mean when they say a coffee tastes of onion.17 At the far end are sulphur compounds and phenol, which make a stinker: a single bean you can taste in a whole cup.

This is not inevitable. In a carefully run Brazilian natural process, butyric acid was not detected in any sample at any stage.5 Over-fermentation is a failure of control, not an inherent property of long processing.

My own view, for what it is worth. The useful question is not whether a coffee tastes fermented. It is whether the producer could do it again on purpose. A vinegary lot that was aimed at is a style; the identical lot arrived at by leaving the tank an extra night because it rained is a fault wearing a style's clothes, and the fact that they taste the same is exactly why the distinction is worth insisting on.

fig 9The gradient, and the fact that nobody agrees where to cut it. Compound order and the onion threshold from sources 5 and 17; there is no hour scale, because this is one of the few things in the post nobody has published a clean timeline for, and the curves show the order only.order only · interactive
For the enthusiast: the onion threshold, and why the line moves

The review literature puts the propionic and butyric threshold at around one milligram per millilitre, below which they are background and above which they are the coffee.17 The review attributes it to an earlier study by López and colleagues, which I have not read, so I cite the review. Stinkers are graded out by hand and by machine.

In a semi-dry process with pitched yeasts, neither butyric nor propionic acid was detected in any treatment.20

And the line genuinely moves. The Ecuadorean trial ran a fermentation four times longer than standard and got a fruitier, more acidic cup that the panel liked.7 The same extra two days on a warmer tank with a different starting community would produce something that gets rejected. Both results are real, and the difference between them is not duration, it is which organisms were doing the eating and whether the pH fell fast enough to keep the wrong ones out.

The three routes are the subject of the other three posts on this branch: the wet mill and the washed route, drying whole cherries, and honey and the experimental methods. This post is the biology that all three of them share.

10 for your agent

Use your agent with this page.

Any part of this post can be explained at your level by the AI you already use. Every section heading and figure caption has an ask agent button: pick explain it simply, go deeper or how do we know, and it copies a prompt about that exact spot. Select a passage for one about just that.

I'm reading "What fermentation actually does". https://hetanshmehta.com/blog/coffee/fermentation Explain the whole post simply, for someone with no science. - With this page's WebMCP tools: call get_outline(), then explain_section({section, level: "child"}) for each section. - If you can open links, read https://hetanshmehta.com/txt/blog/coffee/fermentation.txt. - If neither, say so and I'll paste the post in. Stick to what the page says, and tell me where it stops.
  • get_outline the map: sections, claims, figures
  • explain_section any section, for a child, student or expert
  • explain_figure any figure, the same three ways
  • get_evidence the study behind a claim
  • get_glossary any term, in plain words
  • start_guided_reading a teaching plan, a quiz, or just the answer
For the enthusiast: how the page answers an agent

This is the most technical post on the site, which makes it the one most likely to lose somebody. The readers I have in mind are a fifteen year old who got curious, a microbiologist who will notice every simplification, and a person who just bought an expensive bag and wants to know what anaerobic means. One piece of prose cannot be pitched at all three, so the page carries tools instead.

They are registered with WebMCP when the browser has it, which today means behind a flag, and the page does nothing at all when it does not. All six are read-only and answer from this page's own data: the explanations, the glossary and the evidence notes are written by hand, not generated on the spot, and they live in one small file the page fetches the first time an agent asks. Nothing else goes over the network, and nothing needs a key. A plain-text version of the whole post, prose and sources, lives at a stable URL for agents without a browser, and it opens with a section written for an agent that is helping someone read.

For the penetration question in section 05, get_evidence returns both sides and what is unresolved, because that is the honest answer:

> get_evidence({ claim: "penetration" }) { "claim": "penetration", "status": "contested", "evidence_for": [ { "source": 10, "point": "Deuterium-labelled tracers put in the liquid were recovered inside the beans, 11.2 micrograms per gram of 2-phenylethanol at 12 hours." }, … ], "evidence_against": [ { "source": 7, "point": "The same authors note compounds may be adsorbed onto the bean or trapped between endosperm and parchment …" }, … ], "unresolved": "No study has followed an isotopically labelled microbial metabolite through drying and roasting into the beverage …" }

hover, tap or focus · every number above is in here

Sources · 29 references

Peer-reviewed unless marked otherwise. Where two studies disagree, both are here and the disagreement is stated in the post rather than resolved. On the penetration question in section 05, sources 6, 7, 8, 10, 11, 24 and 29 are the ones to read against each other.

  1. Esquivel, P. and Jiménez, V.M. (2012). Functional properties of coffee and coffee by-products. Food Research International, 46(2), 488-495. Source of the mucilage composition (water 84.2 percent, protein 8.9, sugar 4.1, pectic substances 0.91, ash 0.7, after Belitz et al.), of the fraction weights of the cherry (after Bressani), and of the high degree of methyl esterification of coffee pectin. doi.org/10.1016/j.foodres.2011.05.028
  2. Silva, C.F., Schwan, R.F., Sousa Dias, E. and Wheals, A.E. (2000). Microbial diversity during maturation and natural processing of coffee cherries of Coffea arabica in Brazil. International Journal of Food Microbiology, 60(2-3), 251-260. doi.org/10.1016/S0168-1605(00)00315-9
  3. Elhalis, H., Cox, J. and Zhao, J. (2020). Ecological diversity, evolution and metabolism of microbial communities in the wet fermentation of Australian coffee beans. International Journal of Food Microbiology, 321, 108544. doi.org/10.1016/j.ijfoodmicro.2020.108544
  4. de Oliveira Junqueira, A.C., de Melo Pereira, G.V., Coral Medina, J.D., Alvear, M.C.R., Rosero, R., de Carvalho Neto, D.P., Enríquez, H.G. and Soccol, C.R. (2019). First description of bacterial and fungal communities in Colombian coffee beans fermentation analysed using Illumina-based amplicon sequencing. Scientific Reports, 9, 8794. doi.org/10.1038/s41598-019-45002-8
  5. Silva, C.F., Batista, L.R., Abreu, L.M., Dias, E.S. and Schwan, R.F. (2008). Succession of bacterial and fungal communities during natural coffee (Coffea arabica) fermentation. Food Microbiology, 25(8), 951-957. Source of the isolate percentages for the dry route and of the statement that butyric acid was not detected in any sample. doi.org/10.1016/j.fm.2008.07.003
  6. De Bruyn, F., Zhang, S.J., Pothakos, V., Torres, J., Lambot, C., Moroni, A.V., Callanan, M., Sybesma, W., Weckx, S. and De Vuyst, L. (2017). Exploring the impacts of postharvest processing on the microbiota and metabolite profiles during green coffee bean production. Applied and Environmental Microbiology, 83(1), e02398-16. The study that ran wet and dry routes side by side and reported far fewer changes in the endosperms than in the outer layers. doi.org/10.1128/AEM.02398-16
  7. Zhang, S.J., De Bruyn, F., Pothakos, V., Torres, J., Falconi, C., Moccand, C., Weckx, S. and De Vuyst, L. (2019). Following coffee production from cherries to cup: microbiological and metabolomic analysis of wet processing of Coffea arabica. Applied and Environmental Microbiology, 85(6), e02635-18. The single densest source on this page: metabolite concentrations in water and on beans at 16 and 64 hours, the volatile class breakdown, the tenfold GABA rise, and the argument that much of what appears is endogenous bean metabolism rather than microbial. doi.org/10.1128/AEM.02635-18
  8. Elhalis, H., Cox, J., Frank, D. and Zhao, J. (2020). The crucial role of yeasts in the wet fermentation of coffee beans and quality. International Journal of Food Microbiology, 333, 108796. The natamycin experiment: yeasts suppressed selectively, with the ester, alcohol and cup-score differences that followed. doi.org/10.1016/j.ijfoodmicro.2020.108796
  9. Elhalis, H., Cox, J. and Zhao, J. (2023). Yeasts are essential for mucilage degradation of coffee beans during wet fermentation. Yeast, 40(9), 425-436. Seven treatments with bacteria and yeasts selectively suppressed; the source for endogenous pectinolysis in the first 12 hours and for the eight-hour breakdown when pectinase and lactic acid are added. doi.org/10.1002/yea.3888
  10. Hadj Salem, F., Lebrun, M., Mestres, C., Sieczkowski, N., Boulanger, R. and Collignan, A. (2020). Transfer kinetics of labeled aroma compounds from liquid media into coffee beans during simulated wet processing conditions. Food Chemistry, 322, 126779. The labelled-tracer experiment, and the source for every number in the crossing figure. doi.org/10.1016/j.foodchem.2020.126779
  11. Hadj Salem, F., Vasai, F., Duez, C., Sieczkowski, N., Boulanger, R. and Collignan, A. (2022). Mass transfer kinetics of nonvolatile compounds into coffee beans during wet processing: study at the laboratory scale and in real conditions using two yeast strains. ACS Food Science & Technology, 2(5), 852-861. doi.org/10.1021/acsfoodscitech.2c00022
  12. Zhang, S.J., De Bruyn, F., Pothakos, V., Contreras, G.F., Cai, Z., Moccand, C., Weckx, S. and De Vuyst, L. (2019). Influence of various processing parameters on the microbial community dynamics, metabolomic profiles, and cup quality during wet coffee processing. Frontiers in Microbiology, 10, 2621. Source for fermentation duration being the largest lever, and for soaking tempering and standardising the result. doi.org/10.3389/fmicb.2019.02621
  13. Brioschi Junior, D., Carvalho Guarçoni, R., de Cássia Soares da Silva, M., Gomes Reis Veloso, T., Kasuya, M.C.M., da Silva Oliveira, E.C., da Luz, J.M.R., Moreira, T.R., Debona, D.G. and Pereira, L.L. (2021). Microbial fermentation affects sensorial, chemical, and microbial profile of coffee under carbonic maceration. Food Chemistry, 342, 128296. doi.org/10.1016/j.foodchem.2020.128296
  14. de Melo Pereira, G.V., de Carvalho Neto, D.P., Medeiros, A.B.P., Soccol, V.T., Neto, E., Woiciechowski, A.L. and Soccol, C.R. (2016). Potential of lactic acid bacteria to improve the fermentation and quality of coffee during on-farm processing. International Journal of Food Science & Technology, 51(7), 1689-1695. doi.org/10.1111/ijfs.13142
  15. Bressani, A.P.P., Martinez, S.J., Evangelista, S.R., Dias, D.R. and Schwan, R.F. (2018). Characteristics of fermented coffee inoculated with yeast starter cultures using different inoculation methods. LWT, 92, 212-219. doi.org/10.1016/j.lwt.2018.02.029
  16. Martinez, S.J., Bressani, A.P.P., Dias, D.R., Simão, J.B.P. and Schwan, R.F. (2019). Effect of bacterial and yeast starters on the formation of volatile and organic acid compounds in coffee beans and selection of flavors markers precursors during wet fermentation. Frontiers in Microbiology, 10, 1287. doi.org/10.3389/fmicb.2019.01287
  17. Haile, M. and Kang, W.H. (2019). The role of microbes in coffee fermentation and their impact on coffee quality. Journal of Food Quality, 2019, 4836709. Review. Source for the propionic and butyric acid threshold and for the observation that most organisms isolated from spontaneous fermentations lack quality-improving attributes. doi.org/10.1155/2019/4836709
  18. Avallone, S., Brillouet, J.M., Guyot, B., Olguin, E. and Guiraud, J.P. (2002). Involvement of pectolytic micro-organisms in coffee fermentation. International Journal of Food Science and Technology, 37(2), 191-198. The dissenting paper on pectin: concludes that depolymerisation of pectic substances by pectolytic micro-organisms does not occur, or is negligible, during fermentation. doi.org/10.1046/j.1365-2621.2002.00556.x
  19. de Melo Pereira, G.V., Soccol, V.T., Brar, S.K., Neto, E. and Soccol, C.R. (2017). Microbial ecology and starter culture technology in coffee processing. Critical Reviews in Food Science and Nutrition, 57(13), 2775-2788. doi.org/10.1080/10408398.2015.1067759
  20. Evangelista, S.R., da Cruz Pedrozo Miguel, M.G., de Souza Cordeiro, C., Silva, C.F., Marques Pinheiro, A.C. and Schwan, R.F. (2014). Inoculation of starter cultures in a semi-dry coffee (Coffea arabica) fermentation process. Food Microbiology, 44, 87-95. doi.org/10.1016/j.fm.2014.05.013
  21. Elhalis, H., Cox, J. and Zhao, J. (2023). Coffee fermentation: expedition from traditional to controlled process and perspectives for industrialization. Applied Food Research, 3(1), 100253. Review, used for the control parameters and for the framing of spontaneous versus controlled fermentation. doi.org/10.1016/j.afres.2022.100253
  22. Hadj Salem, F., Achir, N., Sieczkowski, N., Boulanger, R. and Collignan, A. (2023). Modelling the transfer and degradation kinetics of aroma compounds from liquid media into coffee beans during simulated wet processing conditions. Journal of Food Engineering, 343, 111303. The follow-up that separates transfer from degradation inside the bean, which is the distinction section 05 turns on. doi.org/10.1016/j.jfoodeng.2022.111303
  23. Elhalis, H., Cox, J., Frank, D. and Zhao, J. (2021). Microbiological and chemical characteristics of wet coffee fermentation inoculated with Hanseniaspora uvarum and Pichia kudriavzevii and their impact on coffee sensory quality. Frontiers in Microbiology, 12, 713969. Source of the sucrose and fructose figures and of esters at about 30 percent of green-bean volatiles. doi.org/10.3389/fmicb.2021.713969
  24. Hurtado Cortés, V., Bahamón Monje, A.F., Bustos Vanegas, J.D. and Gutiérrez Guzmán, N. (2024). Challenges in coffee fermentation technologies: bibliometric analysis and critical review. Journal of Food Science and Technology, 61(12), 2223-2234. The review that treats diffusion through parchment and endosperm as the working picture, and finds the newer fermentation methods lacking sufficient scientific evidence. doi.org/10.1007/s13197-024-06054-5
  25. Czerny, M., Mayer, F. and Grosch, W. (1999). Sensory study on the character impact odorants of roasted arabica coffee. Journal of Agricultural and Food Chemistry, 47(2), 695-699. The aroma recombination experiment: coffee smell rebuilt from 27 pure odorants, then compounds removed one at a time. doi.org/10.1021/jf980759i
  26. Schwab, W. (2013). Natural 4-hydroxy-2,5-dimethyl-3(2H)-furanone (Furaneol). Molecules, 18(6), 6936-6951. Source for furaneol’s caramel-like smell, its place among the key aroma compounds of strawberry, raspberry, pineapple and tomato, and its being a Maillard product as well as a fruit compound. doi.org/10.3390/molecules18066936
  27. Duan, S., Dong, J., Liu, S., Yu, L., Li, Y., Yin, H., Fang, C. and Du, Z. (2026). Chemical and sensory profiling of fermented, washed, and artificially flavored coffee beans: insights into flavour quality, authenticity, and food safety implications. Food Chemistry, 525, 150338. doi.org/10.1016/j.foodchem.2026.150338
  28. Bressani, A.P.P., Martinez, S.J., Sarmento, A.B.I., Borém, F.M. and Schwan, R.F. (2021). Influence of yeast inoculation on the quality of fermented coffee (Coffea arabica var. Mundo Novo) processed by natural and pulped natural processes. International Journal of Food Microbiology, 343, 109107. doi.org/10.1016/j.ijfoodmicro.2021.109107
  29. Waters, D.M., Arendt, E.K. and Moroni, A.V. (2017). Overview on the mechanisms of coffee germination and fermentation and their significance for coffee and coffee beverage quality. Critical Reviews in Food Science and Nutrition, 57(2), 259-274. The review that puts germination metabolism alongside microbial metabolism as a cause of what processing does to the seed. doi.org/10.1080/10408398.2014.902804

Three things I could not verify and so did not state as fact. First, a single canonical figure for the thickness of the parchment: published values vary and I could not reach a primary measurement, so the crossing figure draws the layers to no scale and says so. Second, a published timeline of defect compounds against hours of over-fermentation; the compound order is sourced, the hours on that figure are illustrative. Third, the propionic and butyric threshold of about one milligram per millilitre: the review I cite attributes it to an earlier study by López and colleagues, which I have not read, so I cite the review rather than an underlying study.

merged to main from coffee/fermentation · d7b626d · 24 Sep 2026 · the three process posts on this branch lean on this one← the blog