To understand what was lost, picture the vessel. Amazon riverboats of that era were wooden, two or three decks tall, open along the sides, with a forest of hammocks slung gunwale to gunwale so tightly that strangers swayed against each other all night. Families travelled with everything they owned; traders travelled with everything they hoped to sell. Cargo went wherever there was floor - sacks and crates piled in the gangways, on the stairs, along the rails - and the load line was a matter of opinion. No manifest worth the name was issued. Children rode unrecorded, and passengers stepped aboard at unlisted stops along the bank wherever a canoe came out to meet the boat. This is the reason the death toll is uncertain. Not because the loss was small, but because the system that carried these people had never troubled to learn their names.

The manoeuvre that killed her was routine: a turn to come alongside, in a current that at Óbidos can outrun a walking man. Survivors quoted in wire reports described the boat listing and then simply going, and they described something worse - cargo stacked in the gangways that jammed cabin doors shut. People in the hammocks on the open decks had a chance. People behind those doors, in the dark, in water, did not. Divers went down over the following days into a channel moving too fast to work in; by the second morning only ten bodies had been recovered. The captain was arrested. United Press International noted at the time that this was the second mass drowning on the river that year, another overloaded riverboat, the Novo Amapá, having capsized months earlier with heavy loss of life.

The New York Times headline the following day read “300 Die as Amazon Boat Sinks,” and three hundred has stuck ever since, usually with “more than” in front of it. Later accounts push higher. All of them rest on the same shaky subtraction: an estimate of how many boarded, minus the number seen alive. If five hundred is right and 178 survived, the missing come to more than three hundred. If more boarded than anyone guessed - which every description of the overcrowding suggests - the figure is worse. Many bodies were never recovered from water a hundred metres deep and running hard toward the sea. Families waited at landings downstream for people who never arrived. The honest statement is the uncomfortable one: several hundred people died, most of them poor river travellers on an ordinary journey, and the record does not permit precision about how many.

Because it happened in the Amazon, the story acquired monsters. In 2014 the angler and television presenter Jeremy Wade devoted an episode of River Monsters to the sinking, ruling out river dolphins, caimans and piranhas before settling on a hypothesis: that piraíba and redtail catfish, drawn to the harbour by waste from nearby fish-processing operations, dragged struggling swimmers under. It is a hypothesis, advanced on entertainment television more than thirty years after the fact, and it is not what killed the passengers of the Sobral Santos II. An overloaded wooden boat turned over in fast water before dawn with its exits blocked. Whatever came afterwards came afterwards. The disaster was a regulatory one, and treating it primarily as a fish story does a second disservice to people who were already badly served the first time.

There is no monument on the Óbidos waterfront and no plaque listing the dead, because there is no list. The wharf is still a working wharf. Boats still nose into the bank below the hill where the Portuguese built their fort in 1697, and passengers still cross the gangplank with hammocks rolled under their arms, though the hulls are steel now and the rules are stricter than they were. The same narrows that made this town worth fortifying, and later made it the place where hydrologists gauge the entire flow of the Amazon, also made it the worst possible place to lose a vessel: deep water, hard current, and a long way to swim in the dark.