When the Physics Engine Becomes the Puzzle
Donkey Kong Bananza shipped with a destruction system so detailed that Nintendo had to build custom physics architecture for the Switch 2 just to run it. Every chunk of terrain that DK tears through follows real simulation rules – mass, momentum, fracture patterns. That level of fidelity was designed to make the game feel alive. It also handed speedrunners a cheat sheet they are still decoding.
Within weeks of launch, the Bananza speedrun community had moved past simple sequence breaks and into something more technically ambitious: using the physics engine itself as a propulsion system. The current world record holder in the any% category is not just skipping cutscenes. They are weaponizing rubble.
How Destruction Becomes Movement
The core exploit at the heart of most major Bananza skips involves what the community has labeled “chunk surfing.” When DK destroys a large section of terrain, the resulting debris does not just fall – it follows calculated trajectories based on the angle of the strike, the material density of the layer, and DK’s current velocity. Runners discovered that by destroying terrain at a specific angle while jumping, the ejected chunks carry enough momentum transfer to boost DK horizontally or vertically far beyond what standard movement allows. Think of it less like a glitch and more like finding a gear the developers left in the engine.
Material types matter enormously here. The game’s geology system – the same one that makes Bananza’s destructible world one of the Switch 2’s technical benchmarks – differentiates between soft rock, dense ore, and composite layers. Each behaves differently when fractured. Softer materials produce faster but lighter debris. Dense ore generates slower chunks with far more knockback force. Runners have effectively memorized the game’s geological map and plan routes around which material sits at which depth, treating the terrain itself as a resource rather than an obstacle.
The Community Building the Rulebook in Real Time
Speedrun categories for Bananza are still being actively negotiated on the game’s dedicated Discord and across Speedrun.com. The any% board currently allows chunk surfing and all physics exploits, but a growing faction is pushing for a “Natural Movement” category that restricts terrain manipulation to what the game’s intended movement toolkit allows. The debate is genuine and unresolved – some runners argue the physics exploits are not glitches at all, just advanced techniques the engine permits, which is a fair reading of how the system was built.
The “No Major Skips” category, already formalized, bans the most aggressive sequence breaks but still permits chunk surfing within individual levels. That compromise has attracted a different kind of player – people more interested in optimizing combat and route choice than in learning the geometry of a single boost frame-by-frame. The category split is producing two very different-looking runs of the same game.
Tool-assisted runs are revealing the ceiling. TAS submissions have demonstrated that the physics engine has consistent internal behavior down to the subframe level, meaning the exploits are not random or luck-dependent. They are reproducible, learnable skills. That finding accelerated real-time runner interest considerably, because it confirmed that what looks chaotic on screen is actually deterministic underneath.
A particularly active thread in the community right now focuses on DK’s grab mechanic and how it interacts with falling debris. When DK grabs a large chunk of destroyed terrain mid-air and then releases it, the angular momentum from the throw carries back into his own hitbox under certain timing windows. Runners are calling it “throw rebound” and it is not fully mapped yet. Some attempts suggest it could save anywhere from ten to thirty seconds in specific sections, but the input timing is tight enough that no one has made it consistent in a live run.
Nintendo’s Shadow Over the Board
There is an open question about patches. Nintendo has a documented history of updating physics interactions in its flagship titles when runners find exploits that completely bypass intended level design. Bananza’s destruction system is complex enough that targeted patching would be difficult without destabilizing other behaviors, but that has not stopped the community from archiving every current technique on pre-patch versions of the game.
The archiving instinct is itself telling. Runners are treating the current state of the engine as a finite resource that could be reduced at any point. Categories for legacy versions are already being discussed, which means the Bananza board could end up split not just by ruleset but by game version – a layer of bureaucracy most platformer communities do not deal with until years after launch.
Where the Records Stand and Where They Are Going
The any% world record has been broken multiple times in the past month alone, which is typical for a game this new but unusual for how dramatic the drops have been. Each major cut has come from a newly discovered chunk surf application rather than from tighter execution of existing routes. The route is still being written, not optimized.
Sub-categories targeting individual worlds are drawing serious attention from runners who specialize in level-scale optimization rather than full-game routes. These shorter runs let people experiment with specific physics interactions without committing to a two-plus hour attempt, and they are producing the most creative tech discoveries at the moment. The “World 3” board in particular has become a testing ground because that section has the highest density of dense ore layers, making it the ideal environment for studying high-force chunk behavior.
The community has mapped roughly half the game’s geological layers in detail. The other half still holds unknowns, and at least two major sections reportedly contain material type combinations that no runner has fully stress-tested yet. Whatever is buried in those layers may rewrite the any% route again before anyone has had time to call the current one finished.
