Blog

The thing that makes this case unusual is not the gold. It is the aircraft. At least twenty three of them, according to the federal prosecutors who built the case, flying people, diesel, generators, mining motors and extracted ore in and out of a protected Indigenous territory in the far north of Brazil over roughly four years. Six helicopters were seized in one enforcement operation alone, with the carcasses of five more found in two clandestine hangars. Several had their airworthiness certificates cancelled. Several had been stripped of identifying markings and modified in ways that affected their weight and balance, which is a polite aviation term for making an aircraft less safe in order to make it carry more. Investigators found wooden floors fitted inside some of them, built for one purpose, which was holding fuel drums steady in flight.

Early last year a federal judge in the state of Roraima convicted the man who ran it. Rodrigo Martins de Mello, a businessman widely known by a nickname taken from one of his companies, was sentenced to twenty two years and seven months in closed regime for leading a criminal organisation that extracted gold and cassiterite illegally inside Yanomami Indigenous Territory, for obstructing the investigation into it, and for laundering the proceeds. Three others were convicted with him, including his son, Celso Rodrigo de Mello, and his sister, Brunna Martins de Mello, each receiving somewhere between nine and ten years, with the exact terms reported differently by different outlets. The fourth, Leonardo Kassio Arno, received a comparable sentence for direct support to the operation.

Alongside the prison terms came the figure that made the ruling genuinely unusual. The court ordered compensation of 31,724,287 reais and 25 centavos, which at the exchange rate running at the time of the ruling came to roughly 6.1 million US dollars, and it directed that money in favour of the Yanomami people.

It matters what kind of money that is, because the coverage has blurred it. Some reporting described the sum as a fine. It is not, or at least not primarily. It was awarded as compensation for collective moral damages, a category in Brazilian law that exists precisely for harm done to a group rather than to an identifiable individual, and that is a different legal animal from the criminal fines a sentence also carries. The distinction is not pedantry. A fine is revenue for the state. Compensation for collective moral damage is, in principle, owed to the people harmed.

In practice, how such an award actually reaches those people is the part nobody has yet explained publicly in this case. Collective damages in Brazil frequently route through a dedicated fund rather than arriving as a transfer to a community, and the rulings that establish the award and the mechanisms that disburse it are often years apart. Every account of this judgment says the money is reverted in favour of the Yanomami. None of the available reporting sets out the actual payment route, and I am not going to invent one.

To understand why an air fleet was necessary at all, you have to understand the geography. Yanomami Indigenous Territory covers an area larger than many countries, straddles the border region with Venezuela, and has essentially no roads. The gold is in the riverbeds. Getting a mining operation in there means getting diesel in there, and diesel is heavy. A single pump and sluice setup eats fuel continuously. Multiply that by hundreds of camps and the limiting factor on illegal mining in that landscape stops being gold geology and starts being logistics. Whoever controls the fuel flights controls the industry.

It also explains the second mineral in the charge sheet, which most coverage of this case skips past. Cassiterite is tin ore, and tin ends up in solder, which is to say in almost every electronic device anybody owns. Gold is what gets written about because gold is dense and valuable and a single flight can carry a fortune of it. Cassiterite is the opposite. It is bulk. Moving commercially useful quantities of ore out of a roadless forest requires a great deal more lift than moving gold does, and it requires that lift repeatedly rather than occasionally. A fleet that is flying ore as well as gold is not a smuggling sideline. It is freight.

That is the business Mello was convicted of running, and the court’s description of it is the opposite of a rough operation. Prosecutors described a solid and structured criminal organisation functioning between 2018 and 2022, using real companies, real contracts and real invoices to wrap an illegal activity in the appearance of a legal one. One of the firms in the group was an artesian well drilling company. Another was an air taxi and tourism operator which, according to reporting on the case, at points held contracts with the health ministry. An air taxi licence is a reasonable thing for a business in a roadless region to hold. It is also, if you are so inclined, a licence to fly anything anywhere and file paperwork saying it was passengers.

The federal police operation that pulled the hangars apart happened in 2021, and the inventory from it reads less like a mining raid than a supply depot audit. More than ten thousand litres of fuel. Generators. Motors, including truck engines of the sort used to drive high pressure hoses against riverbanks. Ammunition. Provisions. The aviation regulator moved in afterwards and dismantled two hangars outright.

What is striking, looking at the sentence, is which law did the heavy lifting. Illegal mining itself is punished lightly in Brazil. The penalties for a first offence run from roughly one to five years, and for a repeat offence they can drop to between six months and a year, which is one reason enforcement for decades resembled a revolving door. Twenty two years and seven months is not what you get for digging gold out of the wrong river. It is what you get when prosecutors successfully reframe the activity as organised crime and money laundering and charge the person at the top of the logistics chain rather than the men holding the hoses. That reframing is the actual innovation in this case, and it is the part other prosecutors will be watching.

The wider context has to be stated, and then stated carefully. The years this operation was running were the years the situation inside the territory deteriorated to the point where the Brazilian government declared a public health emergency there. That emergency involved serious and well documented harm to Yanomami communities, and it is not this story. This story is an economic and environmental crime with a court ruling at the end of it, and it is better reported as one than as a backdrop for suffering. What is relevant and provable is narrower. Mercury and sediment from mining fouled the rivers that communities drink from and fish in. By the time enforcement arrived, around 5,564 hectares inside the territory had been damaged, and roughly ninety two per cent of that damage had already been done by the end of 2022.

What happened after the ruling is where the case stops being a clean story about accountability.

In the months that followed, reporting in Brazil established that Mello was not in custody. He was appealing, which under Brazilian procedure can leave a convicted defendant free while the appeal runs, and he was reportedly no longer in the country at all, with his operations said to have shifted across the border into Guyana. A federal prosecutor working on Yanomami cases summed up the broader pattern in one flat sentence, which was that nobody is locked up. He was not talking only about this defendant. Another convicted mining boss in the region received twenty six years and was similarly at liberty pending appeal. A third was sentenced to ten years and ten months and likewise not detained. Later reporting also alleged that money connected to Mello’s family had paid living costs for a fugitive former Brazilian intelligence official in the United States, an allegation that belongs to a different case and is unproven here, but which does raise the question that hangs over the six million dollar award. If funds can cross a border for that, the court’s compensation order is a number on paper until somebody finds assets.

Collecting on an award like this is its own long project, and it is not the court’s job. Somebody has to find assets, establish that they belong to the defendant or to people holding them for him, and freeze them, across whatever jurisdictions they have been moved through. Prosecutors in the region have shown they can do this when they commit to it. The same federal unit obtained an order freezing up to 146 million reais against companies accused of cooperating with mining in the territory, and has filed a claim against a mining firm for 1.7 billion. Those are vastly larger numbers than this ruling’s six million, which suggests the capacity exists. What has not been shown publicly is that it has been pointed at this particular judgment.

So on the arithmetic, the strongest environmental crime ruling of its kind in the region produced a prison term nobody is serving and a compensation award nobody has collected. That is an honest place to leave a story about a criminal, and it is a miserable one.

It is not where the story actually ends, though, because the Yanomami did not spend those years waiting for a verdict.

Their main association, Hutukara, built a reporting system of their own. Communities file alerts when they see or hear incursions, and in the course of one recent year that system issued more than fifty of them. The association has been pressing for a surveillance plan drawn up on Yanomami terms, with sixteen monitoring centres positioned along the river channels and drones to watch stretches of water that no patrol can reach on foot. They have asked for more of the permanent protection bases inside the territory, of which seven currently operate. They have asked for access to some of the gold seized in enforcement operations, worth roughly thirty nine million dollars, to fund their own monitoring. And they have asked, repeatedly, for a commitment that outlasts the current operation, because the plan that has been working is scheduled to expire.

The enforcement itself has worked, by the way, and the numbers are remarkable. Active mining inside the territory fell from about 4,570 hectares in early 2024 to about 56 hectares by early 2026. Something like 884 camps were destroyed. Eighty seven landing strips were put out of use. Around 250 kilograms of gold were seized. On the supply side, the fuel stations in one staging town were cut from six to two, after a period when they were reportedly moving around seven million litres of diesel a month. Officially registered gold production in the state fell from roughly 30,700 kilograms in 2022 to 5,808 in 2025, which tells you something about how much of the earlier figure was ever legitimate.

And yet the people doing the work describe something closer to a chase than a victory. The remaining operations have gone small and quiet. Nine out of ten of the mining patches recently identified were under a single hectare. Crews work at night, sink their equipment underwater between shifts and mark the spot with string, build berms to hide discoloured water, and spread themselves out under the canopy instead of clustering in camps big enough to show on a satellite image. One airstrip was blown up four times and restored four times. Supplies come in from across the Venezuelan border. The economic pressure behind all of it has only intensified, with a gram of gold reportedly fetching something over 130 dollars against around 60 a few years earlier. Most enforcement effort, by one assessment, goes on the logistics ring outside the territory rather than inside it, and investigators rarely pursue the buyers, the equipment suppliers or the financiers at all.

Which is exactly what makes the Mello ruling matter, and exactly why one ruling is not enough. For once, the person prosecuted was not a miner. He was the man with the aircraft, the companies and the invoices, and a court found that the structure existed, named what it was, and attached a price to the damage. That is a template. It is also, at present, a template with nothing collected at the bottom of it.

The newly elected head of Hutukara, Dário Kopenawa, has been clear about what the Yanomami are not asking for, which is to have mining in the territory regularised and brought inside the law. He has described that proposal as a continuation of something much older wearing a new word. What his association wants instead is a protection system that does not have an expiry date attached to it, and recognition that the people living there already know how to look after the place.

At one Hutukara meeting, somebody put the request in the plainest possible terms. They wanted a sign on the riverbank. It would say that there is no mining here, and that this is Yanomami land.

In northern Ethiopia there is a ruined wall a little under two metres thick, built from broken phonolite stone set in solid clay mortar and threaded through with layers of wooden beams. It has been standing, in one state or another, for roughly 2,800 years. Earlier this year two engineers took one corner of that wall and one doorway from it, rebuilt both inside a computer as finite element models, loaded them with the worst material properties the stone and clay could plausibly have had, applied the safety factors a modern structural engineer is obliged to apply, and asked a simple question. How much more building would the wall still carry?

The answer came back at roughly sixteen floors. The reconstruction archaeologists had already drawn for the same building has eight.

That is a doubling, and it is worth being exact about what kind of doubling it is. It is a modelling output, not an observation. Nothing excavated at the site says a sixteen storey building ever stood there. What the study establishes is a ceiling on capacity, which is a different thing from a measurement of height, and the gap between those two ideas is where almost every write up of this work has gone astray. Within days of the paper reaching the general press, headlines were calling the place one of the world’s earliest skyscrapers. The two outside specialists who were asked about it both pushed back on precisely that word.

The building is called Grat Be’al Gibri, and it sits at Yeha, a small place in the north of the country that very few readers outside Ethiopia have heard of and that a specialist in the early first millennium BC would name without hesitation. The complex is enormous for its date, roughly sixty metres on a side. It stands on a stone podium about six metres high whose walls are around two point two metres thick. The ground floor walls above that are about one point nine metres thick. At the entrance there were square stone pillars something like ten metres tall. Radiocarbon dating puts the construction at around 800 BC. At some point in antiquity the building burned, and what survives is the lower part of a structure whose upper part is gone entirely.

A few hundred metres away stands the other reason anyone knows the name Yeha. The Great Temple there is generally described as the oldest standing structure in Ethiopia, dated to roughly 700 BC by comparison with buildings in southern Arabia and supported by inscriptions found locally. The archaeologist David Phillipson credited its survival to two things, the care its builders took to set it level on uneven bedrock, and its rededication as a Christian church, possibly as early as the sixth century AD. A building that keeps being used is a building that keeps being maintained. The palace next door had no such second life, which is why one of them is a roof over your head and the other is a footprint.

Yeha has been dug at, abandoned and dug at again for most of a century. A British traveller poked around the site in the 1890s. The Ethiopian Institute of Archaeology began excavating in 1952. That work was interrupted under the Derg, and a French team resumed in 1993. Rock cut shaft tombs were found in the early 1960s, with some speculation that one held a royal burial. More recently the German Archaeological Institute has worked there, and Mike Schnelle of the institute’s Orient Department, the second author on the new study, is a long standing Yeha specialist. His co-author, Martin Drieschner, is not an archaeologist at all. He works at the chair of structural analysis and dynamics at the Brandenburg University of Technology in Cottbus, which is to say he is a civil engineer who normally models things that have not been built yet.

That pairing is the whole reason this study exists, and it is more interesting than the number it produced.

Here is the problem the two of them had to solve. If you want to know whether a wall can carry a load, you need to know what the wall is made of, in numbers. Compressive strength, tensile strength, stiffness. For a concrete frame going up next year, you look those up, because the concrete has been specified and tested. For a wall built around 800 BC out of locally quarried rubble, locally dug clay and hardwood beams, you do not know them. You can take samples, and archaeologists have, but the material varies from stone to stone and the clay has had nearly three millennia to change. Any single number you plug in is a guess dressed up as data.

So rather than guess, they modelled the ignorance itself. The study runs what is called a nested fuzzy interval stochastic analysis, which in plain terms means the material properties are not entered as single values but as ranges, with a structure that distinguishes between genuine random scatter and the plain fact that nobody knows the right figure. The model is then run across those ranges rather than at one point. You come out not with one answer but with a spread, and crucially you come out knowing which of your uncertainties actually mattered.

That last part produced the finding most likely to survive the next decade, and it has nothing to do with floor counts. The variation in the wood barely moved the result. Change the stiffness or strength of the beams across their plausible range and the wall’s capacity hardly notices. What dominated was the masonry, and the specific failure mode that decided everything was tensile failure in the clay mortared rubble. The weak point of a timber reinforced wall, in other words, is not the timber.

Which is a slightly unexpected thing to learn about a construction system whose entire distinguishing feature is the timber. Horizontal wooden beams were laid into the rubble in alternating directions, longitudinal then transverse, course after course, with clay layers between. The wall section the engineers modelled was three point three six metres high and contained fourteen separate layers of beams. Two wood species have been identified at the site, African olive and Cordia africana, both hardwoods, and both understood to have been chosen in part for their resistance to insects. Timber lacing of this general kind shows up independently in a number of building traditions, often in places that shake, because wood gives a rigid masonry wall somewhere to put movement. The Yeha walls use horizontal beams only, which distinguishes them from the South Arabian practice of using both horizontal and vertical members.

That distinction matters more than it looks, because it bears on an argument about where the whole technique came from.

The conventional telling has long leaned on southern Arabia as the source. The Great Temple at Yeha is Sabaean in style. The inscriptions point across the Red Sea. The site belongs to the world historians call Ethio Sabaean, where South Arabian and African building traditions were in contact, and for a long time the default assumption in that contact was a one way transfer. Kwasi Konadu, formerly an endowed chair in Africana and Latin American studies at Colgate University, made the point bluntly when asked about the new paper. The result pushes against the older habit of treating Arabia as the origin of these systems in the northern Horn of Africa. What the engineering shows, he said in effect, is that African builders were calculating height, material and load with real competence roughly three thousand years ago. Davide Tanasi, a classical archaeologist at the University of South Florida, made the complementary point. Archaeologists already regarded this construction system as sophisticated. What was missing was a quantitative assessment of it, and equally sophisticated traditions existed in the Ethio Sabaean world where the two building cultures met.

So why was anyone counting floors in the first place? Because the question has a long paper trail, and it is not an engineer’s question originally. It is a textual one.

A tenth century Yemeni historian, al Hamdani, described a palace at Sanaa called Ghumdan and gave it a height of twenty storeys. Rock paintings and wall paintings in southern Arabia show palace like buildings of at least nine storeys. None of that is evidence in the archaeological sense, but it is persistent, and it has shaped how anyone reconstructing a monumental building in this region thinks about the vertical. At Yeha itself there is one piece of actual structural evidence for height. The footprint of the pillared entrance porch on the southeast side corresponds exactly to a central projecting bay of the building, which implies additional floors above the porch, and on that basis the complex has been argued to have stood at least three storeys. From there the published reconstruction takes a careful step further, five full floors with three smaller recessed levels above them, eight in total, and it is honest about leaning on the painted depictions to get there.

Eight, in other words, was never a measurement either. It was a reasoned drawing, built out of a bit of structural logic and a lot of comparative imagery. What the new study does is take that drawing to a structural engineer and ask whether the surviving wall could even have borne it.

It could, comfortably. Under favourable assumptions about the materials the modelled corner section carried the equivalent of something like forty eight additional floors and the doorway section around fifty six, numbers so far beyond any plausible building that they function mainly as a demonstration that the wall is nowhere near its limit. Push the materials to the worst end of their plausible range, combine that with heavy loading, then apply standard safety factors, and both sections still came out at around sixteen floors. The figures that reached the general press varied, with some coverage reporting at least nineteen rather than sixteen, which is a reasonable sign that readers should treat any single number here as indicative rather than final. The robust version of the finding is the comparison, not the digit. The wall system could carry roughly double what the reconstructed building would have put on it.

The authors are also clear about something that pushes the reserve higher still. Only two sections were modelled, both from the ground floor, where the walls are at their thickest and carry the most. A real building of this type would have thinned its walls as it rose, shedding weight with every level, so the structure as a whole had more margin than two ground floor samples can show.

None of which makes it a skyscraper, and the word is doing real damage. A wall carrying nineteen floors is not a structure that stood nineteen floors high. Konadu’s formulation is the one to keep. What the analysis actually establishes is that the building system was more capable than the surviving ground floor would lead you to assume, which is a finding about engineering knowledge rather than about silhouette. Tanasi put the arithmetic plainly. The floor counts indicate how much additional load the wall could theoretically take before failing. That is a stress calculation, and reading a building height out of it is a category error.

Drieschner, for his part, seems most pleased about the method rather than the result. What he found exciting, he said, was that tools from modern civil engineering, finite element modelling and the handling of fuzzy material parameters, transfer directly onto a 2,800 year old structure. That is the quietly useful part. There are ruins all over the world whose material properties nobody can pin down, which has historically meant that structural questions about them got answered with judgement and analogy. A method that takes the uncertainty as an input rather than a problem to be papered over can be pointed at any of them.

There is a limit to all of this that is worth stating plainly, because a computer model of a ruin is not an experiment on a building. What gets modelled is an idealisation, a clean geometry with defined material zones, and the real wall was never clean. The clay mortar being sampled today has had almost three thousand years of wetting, drying and settling, so even the ranges fed into the model are ranges for an aged material rather than a fresh one. The fire that destroyed the upper structure will have done things to the stone that no material range captures. A model like this is best read as a boundary, telling you what is impossible and what is comfortably possible, and leaving a wide middle where it cannot help.

And there is one piece of evidence the model does not need, which is that the thing is still there. A wall built of rubble and clay, in a building that burned, has held its line through roughly 2,800 years of weather in the open air. That is not a controlled test, and nobody would write it up as one, but it is the kind of fact that should make an engineer suspect the margins were generous before any software is opened. The eight storey reconstruction was conservative in the ordinary professional sense of the word, which is to say its authors chose the smallest building the evidence would support rather than the largest, because that is what you do when the evidence is a painted wall in another country. What the new work adds is the other end of the range, and the finding is that the two ends are further apart than anyone had cause to assume.

The building at Yeha burned a very long time ago and nobody knows how tall it was. That has not changed. What changed is that the eight storeys on the reconstruction drawing are no longer the ceiling of what the evidence permits, and the reason they were ever the ceiling turns out to have been the limits of the drawing rather than the limits of the wall. Someone built that thing out of rubble, clay and olive wood, in a place where the whole technique was supposedly borrowed, and built it with so much capacity to spare that two engineers with a modern model had to push the materials to their worst imaginable state before the numbers came anywhere near the building above.

The thing that made the discovery possible was a disaster. Drought and bark beetles killed off a stretch of forest in the Taunus hills north of Frankfurt, the trees came down, and archaeologists were suddenly able to get at ground that had been locked under woodland for centuries. What they found, sitting a few centimetres beneath the forest floor, was an entire Roman village.

That was the headline a year ago. The village, a civilian settlement attached to the small Roman fort called the Kapersburg, turned out to have paved streets, stone buildings, a room with heating channels running hot air through its walls, and the remains of kitchens with hearths built up out of stone and broken pottery. Archaeologists estimated it had been lived in for about a century and a half, and that alongside a garrison of roughly 160 soldiers there were something like 600 to 700 civilians.

What has emerged since is smaller, and considerably more interesting. In the latest work at the site, the team found a temple, two inscribed votive stones, and a statuette of Victoria with her wings and palm branch still recognisable. And lying near the temple, among fragments of handmade pottery, they found a brooch.

The brooch is the reason this story matters. According to the Hessian state heritage office, it is the first archaeological evidence for use of this area after the Roman frontier was given up. Which means that for the first time, there is something physical to say in answer to a question that has hung over this landscape for a very long time: when Rome walked away from its German frontier, did anybody stay?

To see why a single small object can carry that much weight, you need to know what the Kapersburg was part of.

The Upper German and Raetian Limes was the land frontier of the Roman Empire in this part of Europe, a line of palisade, ditch, wall, watchtowers and forts running for hundreds of kilometres across what is now southern Germany. It was not a wall in the sense people usually picture. It was a continuous controlled edge, studded with small installations where soldiers watched, taxed, escorted and reported. The Kapersburg was one of those installations: a modest fort, not a legionary fortress, holding a garrison you could fit into a couple of modern buses.

And then, somewhere in the middle of the third century, the Romans let the whole thing go. The frontier was pulled back to the Rhine and the Danube, the forts were vacated, and a line that had taken generations to build and maintain simply stopped being a frontier. German scholarship calls the episode the Limesfall, the fall of the Limes. The exact dating is argued over, and the honest position is that it was a process rather than an event, with pressure from outside, civil war and imperial crisis inside, and no single day anyone can point to.

What makes it one of the quiet great events of European history is how it is usually told. It tends to be written as a date, or as a line on a map that moves. But every one of those forts had a village attached to it, and every one of those villages was full of people who were not soldiers: traders, smiths, innkeepers, families, the whole apparatus that grows up wherever a few hundred men with regular pay are stationed. When the army left, the army’s pay left with it. What happened to everyone else is one of the least documented aspects of the whole affair, because people leaving quietly is almost archaeologically invisible.

This is exactly why the civilian settlements, the vici, are worth so much more to archaeologists than the forts. Forts tell you about the army, and the army is already comparatively well documented through inscriptions and military records. A vicus tells you how people actually lived on the edge of an empire: what they ate, what they bought, what they made, who they married, what they worshipped.

Before this project, almost nothing was known about the Kapersburg’s vicus. Only a single structure outside the fort had ever been documented, a bathhouse near one of the gates. The rest was under trees.

That is more normal than it sounds. The Limes is a World Heritage site and has been studied intensively since the nineteenth century, but study has always concentrated on the forts and the frontier works themselves, because those are what is visible, what is monumental and what the early investigators were looking for. Louis Jacobi, the nineteenth century investigator whose name is attached to the early archaeology of this region’s Roman military sites, is still cited in accounts of the Kapersburg today. The villages beside the forts were comparatively ignored, and in a case like the Kapersburg, where the whole settlement lay under mature woodland, they were effectively unreachable even for archaeologists who wanted them.

So when the trees died, the project that followed was not a lucky accident so much as a long-deferred opportunity that finally became possible. The excavators have been working a landscape that scholarship had written about for a century without ever being able to open.

The work has been running since 2024, carried out by hessenARCHAEOLOGIE, the archaeological arm of the Hessian State Office for Monument Preservation, in cooperation with the University of Frankfurt. The excavation director is Manuel Pieper. The district archaeologist and Limes coordinator for the region is Dr Kai Mueckenberger. Professor Markus Scholz of the University of Frankfurt, whose field is Roman epigraphy, is working on the inscriptions. Michael Gottwald is also on the team, and Katharina Griesshaber of the Saalburg museum, the reconstructed Roman fort nearby that has been the centre of Limes scholarship in this region for well over a century, has been involved in the finds work.

The area opened up is substantial, roughly 200 metres by 300 metres, and what the team has been able to do is trace the shape of the settlement: lines of road, terraces cut into the slope, and the positions of buildings. One of them held part of a smithy, later flattened when the settlement was reorganised, which is the kind of detail that tells you a village was not static but rebuilt itself over the decades the way any living place does.

The temple is small. The building is around five metres long, rectangular, with the walls of its inner sanctuary about 50 centimetres thick, and the archaeologists have described its state of preservation as unusually good. The finds inside and around it are what identify it: two votive stones carrying inscriptions, fragments of statuary including a hand thought to belong to a figure of Mars, animal sculpture fragments, and the Victoria statuette, of which a nearly complete torso survives with the wings, the palm branch and the victor’s laurel all still legible.

Mars and Victoria, on a frontier. It is not hard to work out what a garrison community was asking for.

Victoria is the more revealing of the two, in a way. Mars in a soldiers’ village is almost expected furniture. Victoria is specific: she is not protection, she is outcome. A community that commissions and keeps a figure of victory is a community with a stake in results, which fits a place whose entire economy depended on the garrison next door continuing to exist and continuing to be paid.

There is something worth pausing on there, though, which is that a temple to the war god in a soldiers’ village is not simply a piece of military decoration. Votive stones are transactional objects. Somebody paid for them. Somebody made a vow, in a specific set of circumstances, and then paid a stonecutter to record that the vow had been kept. Those inscriptions are the thread that could still produce actual names out of this site, which is Scholz’s job, and the team is waiting on the reading of a Latin dedication that may even give the settlement its Roman name. For now, nobody knows what the people who lived there called the place.

Which brings us back to the brooch.

Here is why it carries the weight it does. Dating a Roman site is normally done with the enormous, well-studied mass of Roman material culture: coins, samian pottery, brick stamps, specific forms of vessel whose production dates are known tightly. Those things stop at the Kapersburg where you would expect them to stop, in the third century, because they were made and traded inside a system that withdrew. The problem with trying to find out whether anyone was still around afterwards is that the people who might have been there were not using that system. If they left no coins, no stamped tile and no imported tableware, then as far as the usual methods go, they left nothing at all.

A late antique brooch breaks that silence, because it is datable on its own terms. Brooch typology is one of the more finely resolved areas of the period, and a form that belongs later than the withdrawal, found in the temple area, is a datable object sitting in a layer of time that was supposed to be empty here.

The handmade pottery matters in exactly the same way, and it may matter more. Roman pottery in a settlement like this was wheel-thrown, standardised, produced in workshops and distributed through trade. Pottery made by hand is a different technological signal entirely. It points to people producing their own vessels outside that network, which is characteristic of the non-Roman groups of the region rather than of a Roman provincial village. In a Roman provincial landscape, handmade sherds are not a whisper. They are a shout.

What the archaeologists say about who those people were is appropriately careful, and it is worth reproducing their caution rather than smoothing it over. The state heritage office’s position is that the brooch and the pottery represent the first archaeological evidence for use of the area in the period after the Limes, and that this can potentially be connected to Alamannic settlement in the Wetterau, the fertile basin to the east. Potentially. The language in the release is conditional, and the piece of the claim that is firm is narrower than the piece that is interesting.

It is also important to be precise about the scope of what is being claimed, because it is the sort of thing that grows in the retelling. This is not the first evidence that anybody at all remained in Germany after the Roman period, which would be an absurd claim; the broad outlines of post-Roman settlement in the region are well established. It is the first archaeological evidence for post-frontier use of this area, at this site, on this stretch of the Limes. That is a local claim, and it is a genuinely significant one locally, because it converts an assumption into an object.

And note what the brooch does not do. It does not prove continuous occupation. It does not prove anyone lived in the vicus after the garrison walked out. A brooch and some sherds are entirely consistent with people coming back to a ruin, using it, camping in it, taking stone from it, or simply passing through a place that still had standing walls and a usable road. The archaeologists use the word use rather than the word settlement, and that distinction is doing real work.

What has changed is the shape of the question. For this stretch of frontier, the question was whether there was any physical trace of human presence after Rome gave the line up, and the answer to that was previously no. Now it is yes, and the interesting arguments can begin: who, how long, how often, and whether they knew or cared what the place had been.

The other detail worth verifying rather than simply repeating, because it is the part of this story that travels furthest, is the depth. Every account of the Kapersburg work says the archaeology begins a few centimetres below the present forest floor, and the state heritage office used almost exactly that phrasing as the title of its own announcement: a Roman village a few centimetres under the woodland floor. What no source publishes is an actual measured figure. It is a description, consistently given by the excavators themselves, and not a survey number, so it should be read as what it is.

It is nonetheless remarkable, and it explains both why the site survived and why it is now vulnerable. Woodland is an excellent preservative. Nobody ploughs it, nobody builds on it, and the soil accumulates slowly, so a Roman street surface can lie just under the leaf litter for seventeen centuries and still read as a street. The flip side is that there is almost nothing protecting it. Forestry machinery, root damage from falling trees, erosion on a slope and the simple passage of boots all reach that deep without trying.

So the dead trees that made the discovery possible are also a warning. The reason archaeologists could suddenly see this village is that a drought and an insect killed the forest that had been hiding it, and the same clearance that let them in leaves the remains exposed in a way they have not been for a very long time.

The work is not finished, and the team has been explicit that it does not expect to finish soon. Further seasons are planned with students from the University of Frankfurt, which is how a site like this gets excavated and how field archaeologists get trained at the same time. The inscriptions are still being read. The settlement’s Roman name is still unknown.

But the question that stood over this patch of woodland has already changed, quietly and for good. For as long as anyone has studied the Limes here, the period after the frontier was abandoned has been a blank at the Kapersburg, something inferred from the general history of the region rather than seen in the ground. A brooch the size of a coin, lying in the dirt beside a small temple to a war god whose empire had already pulled back a hundred kilometres, has turned that blank into a fact.

Somebody was there. Nobody yet knows their name, and for now that is a better place to be standing than where the question stood before.

There is a figure in a botany journal, published earlier this year, that is captioned with a phrase you almost never see in a scientific paper. It reads: first known field photographs of Ptilotus senarius.

Not the first in a decade. Not the first since the 1960s. The first that exist. Before the summer of last year, there was no photograph anywhere on earth of this plant alive, in the ground, with its flowers open. There were two dried specimens, pressed flat and gummed to card, and that was the entire visual record of a species.

This is usually told as a rediscovery story, and it is one, but the rediscovery is the less interesting half. A rediscovery means something was lost and then found. What happened here is stranger and says something uncomfortable about how much of the natural world we have described without ever properly looking at it.

Ptilotus is a mostly Australian group in the amaranth family, and if you have seen one you have probably seen it called a mulla mulla, those soft, bottlebrush looking flower spikes that come up in dry country and look slightly artificial, like something from a florist’s shelf rather than a paddock. There are around a hundred species. Ptilotus senarius is a small, slender thing, easy to walk past, and it is known from one patch of northern Queensland, the Gilbert River country inland between Croydon and Georgetown.

It was collected twice. Once in 1925, by a collector named L.J. Brass, from what his label records as grassy hills. Once more in 1967, by S.J. Dansie, out of melaleuca forest, tea tree country. Then nothing. Nobody brought back another one for fifty eight years.

Here is the part that reframes everything. The species was not described in 1925 or in 1967. It was described in 2014, by Anthony R. Bean of the Queensland Herbarium in Brisbane, in the journal Austrobaileya. He worked it out from those two sheets, recognised that what was on them did not match anything already named, wrote the description, chose the epithet and published it.

Which means a botanist examined two dried fragments collected decades before he started, concluded they were a species new to science, gave it a formal name that will outlast him, and had never once seen the living plant. Nobody alive had. By the time it had a name, it had been missing for forty seven years. Targeted searches went out between 2008 and 2014 and came back with nothing. The species existed in the literature and nowhere else anyone could point to.

A herbarium sheet is an extraordinary object and it is also a profoundly limited one. It will give you the architecture of a plant with great precision, the branching, the leaf shape, the arrangement of the flower parts, and if you are careful you can read a great deal of a plant’s life out of a sheet a century old. What it will not give you is colour, because colour goes in the press and never comes back true. It will not give you posture, how the thing actually stands in the wind, whether the stems hold up or lean. It will not tell you reliably what it was growing with, or what the ground was made of, beyond whatever the collector thought to scribble on the label. And it cannot tell you what a living open flower looks like, which for this family is often where recognition lives.

So the people going out to look for Ptilotus senarius in that dry Queensland scrub were, in a real sense, looking for something they would not necessarily recognise if it brushed their boot.

It helps to understand how much room there is to hide. The Gilbert River country is flat to gently rolling savanna, hot, seasonal, and enormous, the kind of landscape where you can drive for a long time between fences and see the same three kinds of tree. Small herbs and slender shrubs in that environment are often there for a few weeks after rain and gone again, and a plant that is only recognisable when it is in flower is effectively invisible for most of the year. The number of botanists who go looking in such places is small, the number of days they can spend there is smaller, and the odds of one of them standing in the right fifteen metre circle during the right fortnight are not good. That is not negligence. It is arithmetic.

In the middle of last year, a man called Aaron Bean was working on private property in the Gilbert River region. He is a horticulturalist by trade and an endorsed Australian bird bander, which is a licensed role and not a weekend habit. Banding, the practice of fitting numbered rings to wild birds so individuals can be tracked across years, requires training, assessment and a permit precisely because handling wild animals badly kills them. People who do it are, by necessity, people who have been taught to look carefully at living things and to write down what they see.

He found a Ptilotus he could not place, and what he did next is the whole hinge of this story.

He did not take one snapshot. He photographed the inflorescences with their slender stems, then went in close on a single open flower, then stepped back for the plant’s habit and the habitat around it. That is three deliberate frames covering exactly the three things an identification needs, and it is the difference between a photograph that is evidence and a photograph that is a memory. He recorded the ground, granite based soils, and the plants growing alongside. He counted: roughly fifteen individuals inside a radius of about fifteen metres. Then he uploaded the lot to iNaturalist, the public biodiversity platform, without a species name on it, because he knew he could not supply one and he knew where the people who could were standing.

It would be easy, and wrong, to write him as a lucky amateur who happened to point a phone at the right weed. He has given no published account of the moment, so what he was thinking is not on the record and this piece will not invent it. But what he did is on the record, in the figure captions and the methods of the paper he is now a co author on, and it is the work of somebody who understood that an unidentified plant plus a location plus a flower in focus is a scientific object, and that the same plant with a blurry phone picture and no notes is nothing at all.

The other thing he had, which no professional botanist had, was permission to be there. That sounds trivial and it is not. Mesaglio, the paper’s lead author, has made the point directly: iNaturalist has become an invaluable tool for recording biodiversity on private property, which researchers mostly cannot walk onto. A great deal of the world’s remaining undocumented plant life is behind somebody’s gate.

Within a few weeks, the record was spotted. Anthony R. Bean, who keeps an eye on Ptilotus submissions coming through iNaturalist, opened a photograph of a plant he had named more than a decade earlier from two dried specimens and recognised it alive for the first time.

The surname, yes. Both men are named Bean, both are in Queensland, both ended up on the same paper. It is a coincidence that reads like an invention. Whether they are related has not been established in anything published, and the paper’s own account of what happened, that the describer saw the record and then contacted the photographer, reads like two people who did not already know each other. So it stays a coincidence here. It is a good one and it does not need to be more than that.

What followed was the unglamorous part that actually settled the question. A specimen was collected from the site in the following month and lodged at the Queensland Herbarium in Brisbane, and the identification was confirmed by comparing physical material against physical material. The paper is explicit about this, and the sentence deserves attention from anyone inclined to think citizen science has replaced collections: species rediscovery requires an integrative approach and cannot be achieved solely through photographic vouchers. The photographs were the catalyst. The sheet was the proof.

Which produces a neat symmetry. A herbarium sheet is what a photograph cannot replace. A photograph of a living plant in its place is what a herbarium sheet could never supply. The whole affair is a demonstration that the two are not rivals, and that the century old cabinet in Brisbane and the app on a phone are doing different halves of one job.

Timing mattered more than it appears. The record was recognised quickly, which meant the follow up collection could be made while the plants still carried buds and flowers. A fertile specimen is worth a great deal more than a sterile one. Had the upload sat unexamined for a few months, the window would have shut and the confirmation would have waited another year, or longer, since nobody can promise the plants come back.

The rediscovery was published earlier this year in the Australian Journal of Botany, with Thomas Mesaglio of UNSW Sydney and the National Herbarium of New South Wales as lead author, alongside Anthony R. Bean of the Queensland Herbarium and Aaron Bean himself. Mesaglio is a known figure in the odd territory where public photo platforms and formal taxonomy meet, and the paper is as much an argument about infrastructure as it is a note about one plant.

The numbers it marshals are worth sitting with. Something like nine hundred plant species are documented as having gone extinct in the wild since the mid 1700s. Against that, over four hundred rediscoveries have been recorded, running at roughly sixteen a year across the past three decades. Sixteen a year. Rediscovery is not a miracle, it is a rate, and an increasing share of it now begins with somebody photographing something they do not recognise and putting it where experts can see it. The platform’s real contribution is not enthusiasm. It is that thousands of botanically competent people are looking at the same feed, so an unidentified plant from an obscure corner of Queensland gets in front of the one person in the world who named it.

It is worth being precise about what iNaturalist actually is, because calling it an app undersells it badly. A user uploads photographs with a date and a location. Other users propose identifications, and when enough independent agreement accumulates a record is promoted to a status that makes it usable in research, at which point it flows into national and global biodiversity databases that scientists query as a matter of routine. In other words it is not a gallery, it is a pipeline, and the thing coming out the far end is data with a name and a coordinate attached. The weaknesses are obvious, since a photograph can be wrong or unidentifiable and a confident crowd can be confidently mistaken. The strength is scale and speed. A record from a cattle property in inland Queensland was in front of the relevant specialist in the world within weeks, which is faster than almost any formal channel would have managed.

As for the plant, its situation has improved only in the sense that it is now known to exist. Queensland’s most recent flora census moved it from extinct in the wild to critically endangered, which is a bureaucratic sentence containing a real reversal: a species officially gone is officially present again. Mesaglio has noted it appears susceptible to cattle grazing pressure, which in that country is not a small problem. Less than half of its potential habitat has ever been surveyed. The honest total of living Ptilotus senarius known to science is the fifteen or so plants in that fifteen metre circle. There may be thousands more in unvisited scrub. There may not be.

One small thing about the record is telling. When the story was picked up and re run some months after publication, several outlets rounded the gap up and announced a plant unseen for sixty years. It was fifty eight, 1967 to 2025, and the difference is only two years, but the entire reason anyone can say fifty eight with confidence is that a collector in 1967 wrote down a date and a place and somebody filed the sheet. Records are the only thing standing between a species and a vague anecdote. It is a slightly comic footnote on a story about the value of precise documentation that the retelling could not keep the number straight.

The ending here is not the discovery. It is the figure.

For eighty nine years, counting from Brass in 1925, this plant existed to science as two flattened brown fragments in a drawer. For eleven more it existed as a name, formally valid, attached to a thing nobody had seen standing up. Now there are three photographs, taken on a hot day on somebody’s property in northern Queensland by a man who stopped, got close, got the open flower in focus, counted what was around him and wrote it all down. They are printed in the paper, they are free for anyone to use, and they are the first and only images of a living Ptilotus senarius that human beings possess.

That is the result. Not that the plant turned out to be alive. That we finally know what it looks like.

About a hundred metres downstream of where Iron Gate Dam used to stand, a sonar camera sat in the Klamath River through the autumn of 2024 and watched. It did not take pictures in any ordinary sense. A multi beam sonar unit throws high frequency sound into the water and reads what comes back, which means it works in sediment so thick that a camera would see nothing at all, and it can tell you which way a fish is swimming, how long it is, how fast it is going and how far from the bank it passed.

Between the first of October and Christmas Eve that year it counted 7,742 Chinook salmon moving upstream past a place where, for more than a century, no salmon could go. The following autumn the same stretch of river gave up 13,310. Add the two and you get 21,052 fish in two migration seasons, which is where the round figure of twenty thousand that circulated in news coverage actually comes from.

Some of them went a very long way. Fish were documented at Beatty Gap on the Sprague River, at an elevation of about 1,250 metres and more than 580 kilometres from the Pacific, which is the farthest from the ocean that a salmon has ever been recorded in that basin. That is more than 360 river miles inland, uphill, through a fishway, into country where the last salmon anyone saw was seen before the First World War.

And then there is the figure that got the headlines, which is 88 per cent, and it is worth being very careful about what it means, because almost everything written about it has been looser than the paper that produced it.

Damon Goodman, the Shasta Cascade regional director at California Trout, is lead author on the study, published late last year in Scientific Reports. The 88 per cent is a measure of river length, not of fish. The authors took the historically documented extent of Chinook distribution in the upper basin, measured from the furthest upstream historical observation on record downstream to the Iron Gate site, worked out how much of that length is now occupied, and divided one by the other. The answer came to 88 per cent, across roughly 345 kilometres of the Klamath mainstem, the Link River and eleven tributaries combined.

So it is not 88 per cent of the salmon that used to be there. It is not 88 per cent of the whole Klamath basin. It is not a population restored to 88 per cent of anything. It is a statement about geography: of the river length where we have hard evidence salmon once went above the dam sites, almost nine tenths of it now has salmon in it again.

The authors are blunter still about the word they use for it. Recolonisation, in their definition, means movement into and use of habitat following reconnection. It does not mean the establishment of permanent, self sustaining populations. They say so in the paper. They also say what would be needed before anyone could claim that, which is evidence of sustained performance over time, successful reproduction in particular, and a broader ecosystem response. Two years of data is two years of data.

Two seasons is also exactly two seasons, and that matters because the story has a tendency to inflate in the retelling. The dams came out and the fish came back in the two autumn runs that followed. That is remarkable on its own and needs no help. Those two runs represented about 18 per cent and about 19 per cent of the basin wide return in their respective years, which tells you that the upper basin is not yet carrying the bulk of the run, and nobody involved claims it is.

Here is the part of this story that gets told wrong most often, and it is the part that actually explains why there is a river to count fish in.

The four dams, J.C. Boyle, Copco 1, Copco 2 and Iron Gate, did not come out because a conservation nonprofit asked nicely. They came out because tribes in the Klamath basin spent decades fighting for it, in hearings, in courtrooms, in shareholder meetings, in the press and on the river itself. The Yurok, the Karuk, the Hoopa Valley and the Klamath Tribes pushed through administrative processes that outlasted careers. Amy Cordalis, an attorney who represented the Yurok Tribe and now runs the Indigenous conservation group Ridges to Riffles, is one of the names attached to that long effort. The turning point in public opinion came in 2002, when roughly 33,000 adult Chinook died in the lower river in warm, low, crowded water, and tribal members hauled the bodies out by hand in front of cameras. Klamath salmon runs had already fallen by something close to 90 per cent.

The mechanics of actually getting four dams out were their own long grind. The structures belonged to a power utility, and the decisive lever was not a conservation campaign but a relicensing process, because a dam that needs a new federal licence has to prove it can meet modern fish passage requirements, and these four could not do so at anything like a defensible cost. Agreements were signed, collapsed, renegotiated and signed again across more than a decade. A purpose built nonprofit, the Klamath River Renewal Corporation, was created to take ownership of the dams and demolish them, which is a strange and quite modern solution to the problem of who carries the liability for knocking down somebody else’s infrastructure. Its chief executive, Mark Bransom, spent years as the public face of a project that had no precedent at this scale anywhere in the world. Federal regulators gave the final approval, the reservoirs were drawn down, and the last of the concrete came out in the course of 2024. The bill came to around 500 million dollars, carried by utility customers in two states and by the states themselves, with a further 162 million dollars pledged by the federal interior department for restoration work and tens of millions more awarded since.

For the Klamath Tribes in the upper basin, the absence was older and more total. Salmon stopped reaching them around 1909. Their chairman, William Ray Jr, marked the recent return by reaching back to testimony recorded in 1911, in which an elder named Berthe Lotches said that the salmon taken each year from the Sprague River had supplied more than half the food of the people living along it. The Klamath word for those fish is c’yaals, and the tribes do not describe them as a resource. They describe them as relatives.

Which brings us to the author list, because this is where the story stops being a nonprofit press release and becomes something more accurate. The study carries twelve authors, and four of them are tribal scientists, named, as authors and not as acknowledgements. Toz Soto, a fisheries biologist and senior research and policy adviser for the Karuk Tribe, and Alex Corum, also of the Karuk Tribe. Jordan Ortega, who is the anadromous fish specialist with the Klamath Tribes’ Ambodat department. And Oshun O’Rourke of the Yurok Tribe.

They are not passengers on the paper. Ortega is the person who confirmed a radio tagged Chinook in the upper basin in the autumn of 2025, the fish that let the Klamath Tribes say out loud that salmon were back above the dam sites. Ambodat’s monitoring programme runs radio telemetry and passive integrated transponder tagging, visual surveys and the grim, essential work of collecting biological samples from salmon carcasses after they spawn, because a carcass tells you whether a fish actually bred or merely arrived. Barry McCovey Jr, who directs the Yurok Tribe’s fisheries department, has been running crews on this river for years. The Karuk Tribe’s own water quality sampling is the source of one of the most striking numbers in the whole affair: before the dams came out, 58 per cent of samples exceeded public health limits for toxic algae, and after they came out, none did.

That is not background. That is the data.

It is also worth killing one misunderstanding before it spreads, because the paper’s own title uses the word rewilding, and rewilding in most people’s ears means somebody released animals somewhere. Nothing of the kind happened here. No fish were put into the upper Klamath to produce these counts. The dams came out, the river reconnected, and salmon that were already in the system swam up it of their own accord, through a 4.6 metre high fishway at Keno Dam and a stretch of hydraulically awkward water that nobody was confident they would manage. There are hatchery programmes in the basin, including fry releases from Fall Creek, and Soto has publicly criticised the timing of one of those releases as unfavourable, but the fish counted on the sonar were volunteers.

Goodman’s own assessment is that the biological response has been much faster than he, and most of the people working on it, expected. That is a scientist admitting his own forecast was wrong in the pleasant direction, which is rarer than it should be.

Now the harder half, because this is a river, not a fable.

George Pess, a co author who spent years as a federal fisheries scientist and is now science director at the Conservation Angler, has put it about as plainly as it can be put: this is not a magic bullet, and the Klamath still has plenty of problems. In the summer after the second count, an outbreak of the parasite Ceratonova shasta killed hundreds of juvenile Chinook. Snowpack was low that year and water deliveries were cut to below normal spring levels. Spring run Chinook, the ecotype that vanished from the upper river by the 1970s and the one that matters most to the tribes, has not come back, although reconnection at least makes its return conceivable.

Passage above the old dam sites is still imperfect. At Keno Dam, no fall run Chinook were confirmed passing in 2024 at all. The following year a motion sensor camera, which only works in daylight, caught 2,497. At Link River, the figure for that year was 210. Keno Dam does not meet federal or state fish passage standards. Upstream, the endangered suckers of Upper Klamath Lake are still in decline, and the long argument over water for a project that irrigates well over 240,000 acres of farmland has not gone anywhere. Coho, steelhead and Pacific lamprey all appeared in the reconnected river within the first year, and what happens to them next is genuinely unknown.

There are also people in this basin for whom the river is an irrigation supply rather than a fishery, and pretending otherwise would be dishonest. The Klamath Project delivers water to well over 240,000 acres of farmland, and in a low snowpack year every acre foot sent downstream for fish is an acre foot not sent to a field. The Klamath Water Users Association and districts such as the Klamath Drainage District have their own executives making their own arguments, and those arguments did not end when the dams did. Low water in the refuges has its own body count: botulism in Tule Lake National Wildlife Refuge killed well over thirteen thousand birds in a single recent year. Taking the dams out removed one fight from the basin and left the older, harder one exactly where it was.

None of that cancels the 88 per cent. It explains it. Because the most useful thing this study says to every other river with a dam on it is not that salmon are resilient, which everyone already believed and nobody could prove on this scale. It is that for the upper Klamath, the binding constraint turned out to be access. Water quality was bad, disease pressure was real, the climate was against them and the runs were a fraction of what they had been, and still, the moment the concrete was gone, the fish went most of the way back up in two years. The information that told them where to go had survived a hundred years of being unable to act on it.

That is an argument you can take to a licensing hearing. It is also the answer to the question that hangs over every dam removal proposal, which is whether it is worth doing when the river is already in such poor shape. The Klamath was in poor shape. The fish came anyway.

What remains is the counting, and the counting is the point. Somebody has to sit with the sonar footage, frame by frame, and decide what is a Chinook and what is a coho and what is a stick. The team built Bayesian models that classify fish by length and by the timing of their passage, and brought in machine learning help that reached 98.4 per cent accuracy near the bank and about 75 per cent across the full width of the river, roughly doubling the speed of review. The machine learning side came from an unlikely direction. The team leaned on a project called Fisheye, built by researchers out of Caltech, MIT and the University of Massachusetts Amherst as part of the same Visipedia effort that produced iNaturalist and the Merlin bird identification app, which is to say the computer vision lineage behind this salmon count is the same one behind the app that birdwatchers use to identify a song in a hedge. It still needs people. It still needs crews walking tributaries in the cold looking for carcasses, and technicians reading tag detections, and someone from Ambodat standing on the Sprague in October confirming that the signal in the receiver is a fish that started in the ocean.

Those crews were there before the dams came out, counting fish that were not arriving, in a river that had been argued over for a generation. They are the reason there is a baseline to compare any of this to. The headline number belongs to them as much as to anyone, and the honest version of it is narrower and better than the one that travelled: nearly nine tenths of the river they fought to reopen now has salmon moving through it, for the first time in over a century, and the question of whether those salmon stay is one only the next ten autumns can answer.

The paper that came out of this is titled, in full, VLT/SPHERE images of the candidate companion of Betelgeuse. Read that word in the middle again. Candidate. It is still there in the title of the paper that supposedly ended a hundred years of argument, and the thirteen authors put it there on purpose.

What they have is genuinely remarkable. Miguel Montargès of LIRA at the Observatoire de Paris led a team that pointed the SPHERE instrument on the European Southern Observatory’s Very Large Telescope at Betelgeuse on two nights, and pulled out of the glare a faint point of light sitting about 52 milliarcseconds from the star, at a brightness ratio of roughly one to twelve hundred. In their own measurement the separation is 52.32 milliarcseconds with an uncertainty of 0.18, and the position angle is 117.12 degrees give or take 0.6. Those are not vague numbers. That is a thing, in a place, measured.

What they do not have, and say they do not have, is proof that the thing is orbiting Betelgeuse rather than sitting behind it. For that you need to see it move. The paper states the condition plainly: formal confirmation that this is a gravitationally bound companion awaits a second observing epoch, and the object should turn up on the opposite side of the primary after half an orbit. Given a minimum orbital period of around six years, that means waiting about three. Until then, candidate.

It is worth understanding how hard even this much was, because the picture is not a picture in the way people usually mean. Betelgeuse is one of the brightest stars in the night sky, visibly orange, sitting on Orion’s shoulder, and it is enormous. The companion is twelve hundred times fainter and sits so close in the sky that from here the gap is about the width of a human hair seen from a kilometre away. Imaging it is the optical equivalent of photographing a candle held next to a floodlight, where the candle is close enough to the floodlight that both land on the same few pixels.

The way around that is to remove the floodlight mathematically. The team used SPHERE’s ZIMPOL camera through two narrow filters, one on a slice of continuum near the hydrogen alpha line and one on the line itself, then ran the data through processing originally built for hunting planets around other stars, subtracting a model of the bright primary and looking at what refused to subtract away. The leftover signal came out at 6.1 standard deviations with one method and 5.1 with another. In this field that is a detection, not a hint, and it is the reason the team is confident about the object even while being careful about the word companion.

The history behind this is the actual story, and it is longer and messier than any announcement captures. The suspicion that Betelgeuse is not alone goes back generations. It appeared in binary star catalogues in the early 1980s and again in the 2000s. In 1986 a group led by Margarita Karovska reported the star as a binary. None of those claims, as this paper notes drily, ever got independent confirmation, and the field moved on without settling anything.

What kept the question alive was a rhythm in the star’s brightness. Betelgeuse varies on a cycle of roughly four hundred days, which is the star pulsating, and that part is uncontroversial. Sitting underneath it is a second, slower variation with a period of around 2,200 days, about six years, and that one has no good explanation in terms of the star itself. Red giants and supergiants commonly show these long secondary periods and nobody has ever properly accounted for them. In 2024 two separate groups, one led by Jared Goldberg and one led by Morgan MacLeod, argued that for Betelgeuse specifically the long period is an orbit, and that if it is, a companion ought to be at its widest apparent separation from the star at a particular moment in late 2024.

That is a prediction with a date on it, which is the kind of thing astronomers find impossible to ignore. Several teams went looking. Several came back empty. Goldberg’s group reported a non-detection in the ultraviolet with the Hubble Space Telescope, and a group led by Anna O’Grady reported a non-detection in X-rays with Chandra. Then a team led by Steve Howell at NASA, using the ‘Alopeke speckle imager on the Gemini North telescope, got something, but only barely: a tentative detection at 1.5 standard deviations, which is weak enough that a careful person would not bet a career on it. Howell’s group published it, called the object Alpha Ori B, nicknamed it Betelbuddy, and estimated it at around one and a half times the mass of the Sun, likely a hot young blue-white star orbiting inside Betelgeuse’s own extended outer atmosphere.

Here is the coincidence that makes this sequence genuinely odd. Howell’s team observed on the ninth of December 2024. Montargès and colleagues observed on the third and the sixth of the same month, three to six days earlier, and did not know what the other group had. Two instruments, two methods, two teams, one week.

That ordering matters more than it sounds, and it is the thing most coverage of this got backwards. The SPHERE images are not a later, second epoch confirming an earlier detection. They are an earlier epoch, taken first, processed and published afterwards. The red circle you see drawn on the published figures marks where Howell’s team found their object, and the SPHERE signal lands inside it. What that gives you is two independent instruments seeing the same thing in the same place in the same week, which is strong. What it does not give you is the one thing a second epoch would give you, which is motion.

Then there is the mass, which is where the numbers need handling carefully. The paper derives a range of 2.6 to 3.1 times the mass of the Sun. The European Southern Observatory’s own announcement rounded that to roughly two to three solar masses, and much of the reporting repeated the line that the companion is heavier than predicted. Both of those statements are defensible and neither is quite the full picture.

More massive than what, is the question. It is heavier than Howell’s estimate of around one and a half solar masses, which is a real and interesting discrepancy between two detections of the same object. It is heavier than the older assumption that any companion would be a modest, roughly solar-mass star, which is the comparison the observatory’s announcement was making. But Goldberg’s 2024 analysis had already predicted something in the two to three solar mass band, and this result sits at the top of that prediction rather than outside it. So the honest version is that one earlier estimate was undershot and another earlier prediction was confirmed at its upper end, which is messier than a surprise but more useful.

And the mass is not a measurement. It is an inference, and the paper is explicit about the chain. The team took the companion’s brightness in their continuum filter, compared it against PHOENIX model atmospheres, assumed the two stars formed together, assumed Betelgeuse itself is somewhere between eight and ten and a half million years old, and read a mass range off pre-main-sequence evolutionary tracks at that age. Change any assumption in that chain and the number moves. Call it a young star of two to three solar masses and you are on solid ground. Call it a measured mass and you are not.

The weakest link in the chain is one astronomers have been complaining about for decades, and the paper says so in almost those words: nobody knows exactly how far away Betelgeuse is. This sounds absurd for one of the most studied stars in the sky, and the reason is the star itself. Distance is normally pinned by parallax, the tiny annual wobble of a star’s apparent position as the Earth moves around the Sun, and that technique assumes your target is a point with a fixed centre. Betelgeuse is not. Its surface is a boiling, blotchy thing with convective cells the size of planetary orbits brightening and fading, which shifts where the centre of light appears to be and corrupts the measurement. The paper works with a distance of about 168 parsecs derived from the star’s own oscillations, with an uncertainty of plus 27 and minus 15, and notes that this distance produces orbital periods consistent with the 2,200 day signal. Which is a satisfying consistency and still an uncertainty of nearly 20 per cent, and distance feeds straight into brightness, and brightness feeds straight into mass.

A separate loose end: the companion’s position angle does not line up with where Betelgeuse’s known rotation axis says it should be, by more than twenty degrees. The authors are relaxed about it, offering two straightforward fixes. The rotation axis may be tilted out of the plane of the sky, or the companion may simply not have reached its widest separation yet on an orbit with a modest tilt. Either resolves it, but nothing in the data picks between them.

One thing this result does help with is the mystery it grew out of. If the six year brightness rhythm were just the companion stirring up dust and periodically blocking light, the numbers do not work; the authors conclude that dust sublimation alone cannot produce the brightness swings actually seen. What is left is dynamics, the companion physically disturbing the supergiant’s vast distended envelope as it goes round. That is a real contribution to a problem that stretches well beyond this one star, since long secondary periods turn up in red giants all over the sky.

Now the part search engines will insist on, so it is worth saying flatly. None of this means Betelgeuse is about to explode. It is a massive star late in its life and it will end as a supernova, and on the timescales astronomers use for stellar evolution that is soon, which in human terms means it could be tens of thousands of years from now. The episode a few years ago that convinced a lot of people otherwise, when the star faded dramatically and the speculation ran well ahead of the evidence, had an entirely undramatic explanation.

There is also a reason to care beyond Betelgeuse itself. Howell’s team noted that if this object is where they think it is, it is the first close-in stellar companion ever detected around a supergiant, orbiting inside the primary’s own bloated outer atmosphere rather than at a comfortable distance from it. That is an unusual place for a star to be. A supergiant’s envelope is enormous and tenuous, more like a very hot atmosphere than a surface, and a second star ploughing through it for millions of years is a situation nobody has had a real example of before. It would make Betelgeuse a laboratory for what happens when a small star is slowly swallowed by a large one, which is a process astronomers believe is common and have very few places to watch.

Montargès imaged that dimming as well, which is the quiet shape of this whole story. The person who photographed the star going dark is the person who then found what may be orbiting it.

That is also, incidentally, the reason the Great Dimming mattered so much at the time. When Betelgeuse faded by more than half its usual brightness a few years ago, enough that the shape of Orion looked wrong to anyone who knew it, the serious question underneath the public excitement was whether the star’s surface was doing something nobody had modelled. The answer turned out to be dust, a huge cloud of it flung off the star and drifting across the line of sight, and the imaging that showed the star’s blotchy, lopsided face while it happened came from the same instrument and the same lead author as the companion images.

His own framing of the result, in the observatory’s announcement, is that it concludes a century-long quest, and that he jumped out of his chair when he saw the processed images. He also said something more revealing, which is that he had honestly not thought they had the sensitivity to see the thing at all, and that they can see it precisely because it is heavier than anyone expected. The detection happened because a prediction was wrong in a helpful direction.

So what is actually settled, and what is not. There is an object, measured in position and brightness, by two instruments within one week, exactly where a model said something should be. It is consistent with a young star of two to three solar masses. That is a lot. Whether it is bound to Betelgeuse or merely in front of it is not settled, by the authors’ own statement, and will not be until somebody looks again and finds it on the other side.

Which means the century of argument has not ended. It has been handed forward, with a far better set of coordinates than anyone had before, to whoever points a telescope at Orion’s shoulder in about three years and checks whether the faint thing has moved.

A few days ago, in a government office in Chennai more than a thousand miles from the village in question, a registrar signed off on a name. The name is Mandar of Jharkhand. What it protects is a drum with a body made of fired clay, headed with leather at both ends, built in a village called Jarjatta in Raidih block of Gumla district, where by the district administration’s own count about 25 families have been making them for four generations.

Getting that signature took five years. It took a documentation file assembled by Dr Satydeep Singh, a visiting faculty member at the National University of Study and Research in Law in Ranchi, who began the work around 2020 and lodged the application in 2021 through the Chief Minister’s Small and Cottage Industries Development Board. It took three successive Deputy Commissioners of Gumla, Sushant Gaurav, then Karn Satyarthi, then Dileshwar Mahto, each inheriting a file from a predecessor who had been posted elsewhere. It took anthropology written down nearly a century ago by scholars who were studying something else entirely.

And here is the thing worth sitting with. Read every report of the registration, in every outlet that carried it, and you will find the number 25, the word generations, and the names of the officials. You will not find the name of one person who makes a mandar. The makers are a figure in a document. That absence is not a small detail at the edge of this story. It is close to the centre of it.

Start with what the instrument actually is, because the sound is the reason any of this happened. A mandar is not a drum with a wooden shell. The shell is clay, thrown and shaped and then fired until it is hard enough to hold tension, and it is specifically local clay, the sticky reddish nagda earth that the region’s potters have always worked. Leather goes over both ends and is pulled tight. Both heads are played. And the two heads are not the same, which is the part that makes the instrument interesting rather than merely traditional.

For how that difference is produced, the best reporting available anywhere is not from Jarjatta and is not from the registration period. It is a piece the Hindi daily Ranchi Express published in July 2024, two years before the registry acted, and it is set in Tapkara village, in Torpa block of Khunti district, which is a different district altogether. It should be read as a description of how mandars are made in that part of the region, by the maker that reporter happened to find, and not as a description of Jarjatta or of anything happening this year. With that said plainly, it is the only place the craft is reported properly.

What that report describes is a paste called kharan, red stone ground to powder and mixed with clay, laid onto the leather heads. It goes on thicker on one head than the other. That is the whole trick. The heavier loading drops the pitch and shortens the ring on one end, the lighter loading leaves the other end brighter and more open, and a single player working both hands gets two voices out of one body. The two heads are also not the same size. The report gives roughly 30 inches across on the heavy side and roughly 12 on the light side, which tells you the instrument is less a drum than a pair of them joined at the waist.

The same report sets out three sizes with three names. The muchi mandar runs around two feet. The thongi runs around three. The jaspuria, at around three and a half feet, is the one that travels furthest, turning up not only across Jharkhand but in Chhattisgarh, Odisha and Madhya Pradesh. A single drum takes two to three days of work. None of that is in any of the registration coverage, which is exactly why it is worth going and finding it.

The loading is not a quirk of one village. Weighting a drumhead with a paste to tune it is one of the oldest ideas in the region’s percussion, and it works for a reason that has nothing to do with tradition. Added mass on a membrane pulls its pitch down and damps the higher overtones faster than the low ones, so a heavily loaded head produces a short, dark, definite thud, while a bare or lightly loaded head of smaller diameter rings higher and longer. Put one of each on the same resonating body and you have a drum that can hold down a pulse with one hand and answer itself with the other.

Why clay, rather than wood, is harder to answer with any confidence, and makers tend to answer it the way craftspeople everywhere do, which is that the local earth is the right earth. Fired clay is rigid and it is brittle, so a clay shell does not flex and absorb energy the way a carved wooden one does, and the makers’ claim that the nagda earth gives the instrument a sound no other material gives it is exactly the claim the registration has now written into law. It is also, usefully, the one part of the case that can be checked by ear.

Two people speak in that 2024 piece, and what they say is not triumphant. Samman Ram, who makes drums in Tapkara, said the market had thinned badly. He remembered selling well at the Rath Yatra fairs and said demand had fallen considerably since. Lakshmikant Narayan Badai, a school principal in Torpa who also plays nagpuri music, put the instrument at the foundation of the region’s entire musical life and said no cultural occasion feels finished without it. He also said that the village evenings where people gathered and played, forty or fifty years ago, have simply stopped happening.

So the drum was already in trouble before anybody started filing paperwork about it. That is the condition a geographical indication is supposed to address, and it is worth being precise about what one is, because the phrase gets thrown around as though it were a prize.

It is not a prize. It is a statutory register. A geographical indication is a legal recognition that a particular product’s qualities or reputation are tied to a particular place, and that only producers in that place may sell under the name. It is not owned by a company the way a trademark is. It is a collective right, held on behalf of everyone who makes the thing in the place where the thing is made. Which sounds generous until you consider what it costs to obtain, because the register does not take anybody’s word for it. You have to prove the link. In writing. With evidence a registry examiner will accept.

That is straightforward for a product with an industry behind it and records going back decades. It is very hard for an object made in a village by families who never kept a single document about it, because the tradition’s own memory lives in hands and in festivals, and a registry cannot read either. So the proof has to be assembled out of whatever other people happened to write down.

Which is where the anthropology comes in. The documentation drew on the work of Sarat Chandra Roy and Tarak Chandra Das, researchers who studied music and dance in the Chotanagpur region generations ago and whose descriptions, written for entirely different purposes, became the evidentiary spine of a 21st century legal application. Nobody writing those pages was trying to protect anyone’s livelihood. They were describing what they saw. A century later that description is what let a lawyer argue that the drum had been there long enough to count.

The administrative side of it is almost comically mundane and genuinely admirable. A file sits in a district office. The officer who understood it gets transferred. A new officer arrives, is handed a folder about a drum, and does not drop it. That happens three times. There is no moment of drama anywhere in the sequence, and the whole thing worked because of it.

Where the drum is actually used explains why the state cared at all. The mandar is the instrument of Sarhul, the spring festival that marks the flowering of the sal trees; of Karma, which falls in the monsoon months; and of Sohrai, the harvest and cattle festival. It is what the dances are danced to. Take the mandar out and you do not have a quieter festival, you have no festival, because the dance steps are built on the pattern the two heads make between them. That is the sense in which Badai’s remark about nothing feeling complete is a technical statement rather than a sentimental one.

What the registration will change is a more sober question than the coverage suggests. The real protections are narrow and specific. The name is now reserved, which means drums that look identical but are made elsewhere, from other clay or from other materials entirely, cannot be sold as the Jharkhand article. Buyers who want the real thing have a way of knowing they have it. The district has talked about organising the makers into a producer company, which would let them sell as a group rather than individually at fairs, and about market links beyond the local circuit. All of that is worth having.

But a geographical indication does not create demand. It is a fence, not a field. Plenty of registered products in India have the certificate and still have the same thin market they had before, because nobody built the thing that was supposed to come after. The tag is the beginning of a commercial project, and whether the project happens depends on district officials who will themselves be transferred, and on 25 families deciding it is worth their children’s time to learn a trade that currently pays badly. The register cannot settle that.

There is also the question of who would ever enforce it. A registered geographical indication is a right somebody has to assert, which means noticing that drums are being sold under the name from outside the area, and then doing something about it. In practice that work falls to whoever is organised enough to do it, which is why the producer company idea matters more than the certificate does. Twenty five families with a legal right and no body to exercise it are in much the same position as twenty five families with no legal right at all.

Which brings it back to the names. Across every account of this registration, the makers of the mandar appear as a count. Twenty five families. Four generations. No one is quoted. No one’s hands are described. The only people with names are the ones who handled the paper.

That is partly just how news works when the news is an administrative act, and partly something stranger and more structural. The legal case for a geographical indication is a case about a place. You prove that a village makes a thing, that the village has made it a long time, that the clay from that ground gives the thing a quality no other clay gives it. The argument is deliberately impersonal, because a right held by one named person would be a trademark, and the whole point of this instrument is that it is held by everybody there. The anonymity in the file is not an oversight. It is the shape of the protection.

It still leaves a gap in the public record that nobody has filled, and it is a gap worth filling for an entirely practical reason, not a sentimental one. If the producer company happens, somebody will have to run it. If the market links happen, buyers will want to know whose drum they bought. A craft sells on names. The register has given Jarjatta a protected name for a sound, which is real and took five years and three Deputy Commissioners and a century of borrowed anthropology to achieve.

The thing it has not given anyone yet is a name for the person who lays the paste on thicker on one side than the other, so that one drum can say two things at once. That is the next piece of documentation this story needs, and unlike the first one it would not take five years. It would take an afternoon in the village with a notebook.