In a crowded room, someone is moving furniture. Sometimes a cabinet ends up against the wall, and it becomes easier to breathe. Other times it gets shoved into a corner, and the room feels even tighter, though everything is still standing. That's roughly how avalanches behave in a granular fault assembled in the lab from four thousand transparent disks.

The disks were squeezed between two stationary cylinders, while a plate below rotated slowly at 48 millimeters per hour. A load of about 20 kilograms pressed down from above. Every 4 seconds, 24 cameras photographed how the disks pressed against each other: transparent plastic under polarized light reveals stresses as colored bands. Six piezoelectric microphones picked up every click, and force sensors recorded how pressure on the walls changed. Over one hundred hours of a single experiment, about four hundred thousand acoustic bursts and one hundred thousand mechanical avalanches were collected.

Previously, it was thought that an avalanche always releases stress: grains slip, the layer compacts, energy dissipates as heat and sound. But when events were sorted into four bins depending on whether the layer contracts or expands and whether the force on the wall rises or falls, the picture turned out to be more complex. The most frequent avalanches do release energy. However, the largest ones often do the opposite: the layer expands, and the force on the stationary boundary grows. Such events were called negative avalanches, because they don't discharge the material but charge it, like a cocked spring.

The difference also shows up in motion. Ordinary avalanches swirl within a narrow shear band, like water in a brook. Negative avalanches barely move the grains, but their weak displacements creep into the jammed, nearly motionless region and blur the band's boundary. They sound louder while moving less. So an avalanche can do more than just release accumulated stress: it can redistribute it into a more dangerous configuration, from which the next event will emerge stronger.

For natural faults, this changes how we view earthquake preparation. If after one slip the fault doesn't relax but instead locks stress deeper, then forecasting from precursors becomes even less reliable. For now this is a laboratory result, and before any practical application we need to test how such avalanches behave in real rocks. But the very possibility that the room becomes tighter after rearrangement already makes us listen differently to the creaking of furniture.