A joint Brazilian and French team began digging in 1973 in what is now the southeastern portion of Serra da Capivara National Park. Guidon reported the results in 1986, and the numbers were startling: radiocarbon dates on charcoal running from about 48,000 to 32,000 years before present, with later analyses extending the range as far back as 60,000 BP. Alongside the charcoal lay flakes and cobbles of quartz and quartzite that Guidon read as tools.\n\nShe worked out fifteen distinct occupation levels grouped into three cultural phases. The deepest and oldest she named Pedra Furada. Above it sat Serra Talhada, running from roughly 12,000 to 7,000 BP, with an unambiguous kit of knives, scrapers, retouched flakes and cores. Above that, the late Agreste phase. The shelter also carries hundreds of rock paintings, generally dated between about 11,000 and 5,000 years ago. Nobody disputes the upper levels. Everything turns on the bottom of the trench.

In 1994 Tom Dillehay — no defender of Clovis-first orthodoxy, having excavated Monte Verde in Chile himself — argued that the charcoal at Pedra Furada could have come from natural fires rather than human hearths. Others went further. James Adovasio of Florida Atlantic University holds that the supposed tools are simply rocks that fell from the cliff above and shattered on impact, an interpretation with an obvious mechanism: there is a great deal of cliff directly overhead. His assessment was blunt. "The Pedra Furada stuff," he said, "is not even up to capuchin standards."\n\nThat was not a throwaway line. Wild bearded capuchins living in the same park smash stones against embedded rocks, producing flakes and cores that look remarkably like early hominid work. The observation, published in 2016, raised a genuinely awkward possibility: that some of what has been read as human tool-making at Pedra Furada was made by monkeys, or by natural rockfall, or both.

Guidon's answer was that the geology does not do what the critics need it to do. "The carbon is not from a natural fire," she said. "It is only found inside the sites. You don't get natural fires inside the shelters." The French palaeolithic specialist Jacques Pelegrin has taken a more measured version of the same position: he accepts that natural processes could in principle mimic flakes this simple, but considers it unlikely here given the evidence for continuous human presence at the shelter.\n\nSupport has also come from the neighbourhood. In 2013, stone tools at Toca da Tira Peia, elsewhere in the same park, were dated by optically stimulated luminescence to around 22,000 years ago — far younger than Guidon's oldest claims, but still well before any Clovis-first model allows. At the nearby Sitio do Meio, excavated over many seasons and reported on in 2014, the lithics are better preserved and the stratigraphy clearer, which helps anchor the regional chronology even though the site is not as old.

The Clovis-first model, dominant since the 1930s, held that the Americas were empty until roughly 11,500 BP. That model has already been dismantled by sites nobody seriously contests — Monte Verde in Chile at about 14,800 BP, Meadowcroft Rockshelter in Pennsylvania at around 16,000, the Buttermilk Creek Complex in Texas at about 15,500. The mainstream position now allows a first movement south of Alaska somewhere in the range of 15,000 to 17,000 years ago.\n\nPedra Furada asks for something much larger. Its oldest dates belong to what is called the long chronology — arrival perhaps 21,000 to 40,000 years ago, or in Guidon's most expansive version, humans crossing the Atlantic from Africa. That is not widely accepted. David Meltzer of Southern Methodist University put the structural objection plainly: if there were pre-Clovis people in South America, why is there no matching signal in North America? Guidon, for her part, told The Guardian in 2000 that the resistance was territorial — that North American archaeologists could not accept losing the oldest site.\n\nCaution has a track record here. A caribou-bone scraper from Old Crow Flats in Canada, thought in the 1970s to be 27,000 years old and used to argue the long chronology, was redated in the 1990s to about 2,000. The lesson stuck. Pedra Furada remains what it has been for four decades: a site where the stakes are enormous, the evidence is ambiguous, and the honest answer is that we do not yet know.