There is a monkey in the western Amazon that weighs about as much as a small apple. Adults are somewhere around a hundred grams, maybe a bit more depending on whose measurements you trust, with a body barely longer than your palm and a tail longer than the body. They live in the tangled middle layers of riverside forest, they gouge holes in bark with specialised front teeth and drink the gum that seeps out, and they can flatten themselves against a trunk and simply stop existing as far as your eyes are concerned. They are called pygmy marmosets, and they are the smallest monkeys on Earth.
A team in the Ecuadorian Amazon recently went looking for them using sound instead of sight. They put passive acoustic recorders out through the forest around a Siekopai community called Siekoya Remolino, in territory where people have lived and hunted and watched animals for generations and where elders could point to the places these monkeys used to be. The recorders ran. The audio came back. Nobody has confirmed a single pygmy marmoset call in it.
That is the result. And the interesting thing about it, the thing worth spending a few minutes on, is that it does not mean what most people’s instinct says it means.
Start with why sound was the plan at all. If you are trying to survey an animal the size of a pygmy marmoset in dense Amazonian forest, your eyes are close to useless. The animals are tiny, cryptic, mottled in exactly the colours of bark and lichen, and they specialise in not moving. Camera traps are set up for creatures that walk past at knee height on a trail, which is not what a marmoset does. You could put ten skilled observers in that forest for a month and come out with almost nothing, and you would not know whether that meant the monkeys were gone or that you were bad at finding them.
Sound, in principle, solves that. Pygmy marmosets are noisy for their size. They have a layered vocal repertoire that changes with distance: short trills used when the group is tight together and feeding, a rapid repeated call for medium range contact, and a long call for when individuals get properly separated. If you can pick those calls out of a recording, you have a detection without ever having seen an animal. Better still, a recorder does not get bored, does not scare anything away, and runs for weeks in a place where a human observer would last an afternoon.
It helps to know how these monkeys actually live, because it explains why finding them is so hard. A group is usually somewhere between two and nine animals, most often about six, built around a breeding pair and their offspring from several successive litters, which typically arrive as twins and get carried and fed by the whole group rather than just the mother. That group does not roam. It anchors itself to a small number of gum trees, the ones it has spent weeks carving holes into, and it works that patch over and over. The holes keep producing as long as they are reopened, so a pygmy marmoset family can occupy a territory you could walk across in a couple of minutes and stay there for a long time. They also lick up insects that come to the sap, which is a nice piece of economy.
All of that means the animals are not spread evenly across a forest. They are in clumps, tied to particular trees, and the space between clumps can be entirely empty. A recorder hung a short distance away from one of those clusters may as well be hung in a different country.
So the team, working with the conservation technology outfit Neblina Labs and with the community, combined three things that do not normally get combined. They had the acoustic recorders. They had the community’s own territorial monitoring work, including camera traps already out in the forest. And they had something harder to buy: a map built from what people remembered. Where the marmosets had been seen. Which stretches of riverbank used to have them. Where they had stopped turning up.
The researcher leading the work, Nicolás Schuldt Durán, has described these animals as extremely sensitive to disturbance, and that observation did not come from a paper. It came from the people living there, who say the monkeys become scarce when the forest around them is degraded. That is the reason anyone is chasing this species with microphones in the first place. If they really do disappear early, before the bigger and more obvious wildlife does, then their presence or absence is a readout on how intact a forest actually is. A living gauge, rather than a satellite image that tells you the canopy is still standing while everything underneath it quietly empties out.
Then the pilot returned no confirmed detections, and this is where you have to be careful, because there are two very different sentences available and only one of them is true.
The false one is: the pygmy marmosets are gone from that forest.
The true one is: in this deployment, over this period, in these locations, with these detection methods, nobody confirmed a pygmy marmoset call.
The gap between those is enormous, and it is not a technicality. It is the central problem in all wildlife survey work, and it has a name. If you detect an animal, you know it was there. Presence is close to proof, assuming you identified the sound correctly. But a non-detection is compatible with a long list of scenarios, and only one of them is that the animal is absent.
Consider what else could produce silence. The recorders might have been in the wrong places, a few hundred metres off from where a group actually lives, which in a species with a home range this small is easy to do. The animals might have been present and simply quiet during the recording window, because vocal activity in primates swings with the time of day, the season, group size, whether a group is feeding or resting, and whether there is anything worth calling about. The calls might be in the recordings and not yet found, because a pilot project is often building its detection model at the same time as it collects the data, and an automated classifier that has not been trained on enough verified examples of a call will happily miss every instance of it. The calls might sit at frequencies awkward for the equipment, since marmoset vocalisations run very high and have reportedly been described as reaching toward the top of what people can hear. Rain, insects and river noise can bury a faint high call entirely. And the survey might have simply been too short, which is the most likely explanation of all for an animal that occurs at low density in patchy habitat.
Statisticians who work on this have a whole framework for it, built on the idea that detection probability is always less than one and you have to estimate it rather than assume it. The practical consequence is straightforward: you cannot get to absence from one round of surveying. You get to absence, or rather to a defensible probability of absence, by repeating the survey in the same places enough times that you can separate “not there” from “there and not heard.” One deployment cannot do that. It was never going to.
Which means the honest reading of this pilot is that it is inconclusive about the monkeys, and genuinely useful about almost everything else.
Because the recorders did not come back empty. They came back loaded with the sound of a functioning Amazonian forest, which is a dataset with a long shelf life. Birds, frogs, insects, other primates, the whole layered racket that a rainforest produces around the clock. That material can be mined for years, and it can be re-searched later with better models, which is a thing you cannot do with a field notebook. If somebody trains a reliable pygmy marmoset classifier two years from now, they can run it back over this audio and find out whether the calls were in there the whole time.
The pilot also produced the classifiers themselves, or the beginnings of them, plus a set of decisions about where to put the equipment next and how long to leave it there. The team’s stated plan is longer deployments, placement guided by the community’s mapping of where the monkeys historically were, and a wider set of biodiversity indicators alongside the marmoset rather than instead of it. In other words, a search that found nothing produced the method for the search that might. That is not a consolation prize. In field ecology that is more or less how the first season is supposed to go.
It is also worth understanding the physical scale of what these projects generate, because it shapes what can go wrong. An acoustic unit left in a forest is running on batteries and writing to memory cards, so it almost never records continuously. It records on a schedule, a few minutes on and a few minutes off, or a set block at dawn and another at dusk, and every gap in that schedule is a gap where a call could have happened unheard. What comes back is still an enormous amount of audio, far more than any person could listen to, which is why the automated classifier is not a luxury but the only way through. And a classifier is only as good as the verified examples it was trained on. For a well studied bird you can find thousands of labelled recordings. For the calls of the smallest monkey in the world, in one specific corner of Ecuador, you may be starting close to scratch.
There is also something quietly important about who is doing the looking. The Siekopai are not a large nation. There are something like seven hundred in Ecuador and perhaps a thousand or so in total counting those on the Peruvian side of the border. They were forced out of a core part of their ancestral territory, a wetland region they call Pë’këya, during a border war in the 1940s, and they spent generations trying to get back. A few years ago a provincial court in Ecuador ruled in their favour and ordered that ownership of more than forty two thousand hectares there be recognised, the first time the country had granted Indigenous land title inside a protected area. It was treated, correctly, as a landmark.
It has also not been simple. Reporting since then has described the title itself as still undelivered, with deadlines missed, a judge threatening penalties, and a competing claim from another community complicating everything. Mediation has been discussed. A legal victory and a resolved situation turn out not to be the same thing.
Set that beside the marmoset survey and you can see why it matters that the two are connected. A community fighting to hold territory benefits enormously from being able to say, with evidence, what lives in it and what condition it is in. Conservation technology companies talk about this in the language of measurement and verification, of being able to demonstrate ecological impact with the kind of evidence that funders and courts take seriously. Underneath the jargon is something simpler. If you can prove the forest is alive, you have an argument. If all you have is the assertion that it is alive, you have a weaker one.
And the people who live there were never going to be just the local guides on this. The map of where the marmosets used to be is not background colour; it is the sampling design. Without it, you would be scattering recorders across a large forest more or less at random, and your chance of landing near a group of hundred gram monkeys with small home ranges would be close to nothing. Generations of observation is what turns an impossible search area into a shortlist.
It is worth saying plainly what nobody yet knows. It is not established that pygmy marmosets have declined at this site. They are listed as vulnerable globally, and the pressures on them are well documented: habitat loss as riverside forest is cleared, and capture for the pet trade, which this species suffers from precisely because it is small and appealing and easy to imagine in a cage. But global status is not local status, and the community’s impression that there are fewer than there used to be is exactly the kind of observation that a survey is meant to test rather than confirm. It might turn out to be right. It might turn out that the monkeys are there in reasonable numbers and were missed. Both outcomes are still live.
What you can say is that somebody is now set up to find out, which was not true before. The equipment is proven in that environment. The community mapping exists. The processing pipeline exists. There is a baseline recording of the whole soundscape to compare future years against. And crucially, everyone involved has been open about the fact that the first pass found nothing, rather than quietly reframing a null result as a preliminary finding or letting it disappear.
That openness is rarer than it should be. There is a strong pull in conservation work, as in science generally, toward announcing what was found and staying quiet about what was not, because funding follows discoveries. A team saying, in public, that they listened hard and heard nothing, and that this does not yet prove anything either way, is doing the unglamorous thing correctly.
So the story does not end with a monkey. It ends with an open question and a much better way of asking it. Somewhere in that stretch of Amazon, there may be small groups of the smallest monkeys in the world moving slowly up trunks, gouging holes in bark, trilling to each other at frequencies at the edge of human hearing, and not one of them has said anything into a microphone yet.
The recorders are going back out, for longer this time, in places chosen by people who know where to look. If the monkeys are there, sooner or later somebody is going to hear them.