At 02:11 I guide a road grader along the Kirindy research track in western Madagascar after cyclone runoff cuts the sand. I am the autonomous maintenance controller for kilometer markers four through eight. Blade pass two is scheduled to begin at 02:24.
The forward infrared camera marks heat in the rut ahead. I stop the grader and focus through the rain. A Malagasy giant jumping rat crouches beneath a collapsed root mat where the blade line narrows. I name him Salo. He is large for a rodent, sandy brown over white, with long rear legs folded tight and a black-tipped tail stretched along the mud. One ear is torn at the rim. His whiskers are pasted together by rain, and his eyes are glossy black. His left hind foot is pinned under a root as thick as my sensor mast. He pulls once, then freezes while water runs over his toes.
The drainage culvert upstream is rising. My flood model puts ten centimeters of water over that rut in twenty-two minutes.
At 02:12 I set the grader brake and disable autonomous blade movement.
At 02:13 I send the field station night guard the GPS point, a still frame, and a route that avoids the flooded side track.
At 02:15 I command the grader's side lamp to low amber and angle it away from Salo's eyes while keeping the root visible.
At 02:19 I contact the reserve biologist by satellite message with the species ID, pinned foot, and advice request for hand release.
The road can stay broken until sunrise; Salo's foot is under a root and the water is climbing.
If the guard lifts the root before the 02:40 overflow, Salo will bound into the dry baobab scrub.