From high score to completion time
Competitive play measured by score — the original metric of the arcade era — assumed that games had no ending reachable by most players, or that reaching the end was less interesting than accumulating the highest score on the way there. Twin Galaxies, the organisation that maintained arcade high score records from 1981, operated in a world where most arcade games' dominant metric was points rather than completion time: Pac-Man, Donkey Kong, and Centipede were designed as games of indefinite duration where the score was the achievement, not a marker en route to a conclusion.
The shift toward completion time as a competitive metric followed the shift toward games with defined endings. Super Mario Bros. (1985) could be completed; completing it quickly was a different challenge from completing it at all. Nintendo Power magazine published completion times alongside strategy guides; players who had finished the game began timing their subsequent playthroughs. The comparison of completion times between players was initially informal — players comparing notes in person or by mail — and became more systematic with the internet's ability to connect players who would never meet physically but could share video evidence of their runs.
Glitches, skips, and out-of-bounds
The distinction between completing a game and completing it quickly created incentives to find routes through the game that hadn't been designed as routes. A developer who designs a game expects players to follow the designed path; a speedrunner who wants to minimise completion time treats the designed path as the slowest available path and looks for shortcuts. Some shortcuts are intentional — designed level layouts with multiple paths, faster routes that reward players who explore. Most significant speedrun shortcuts are unintentional: bugs in the game's collision detection, memory management errors, or AI behaviour that can be triggered in sequences the designer didn't anticipate.
The discovery and documentation of unintended shortcuts requires a different kind of game knowledge than completing the game normally. A player who wants to understand how Mario walks through a wall needs to understand how the game's collision detection works — what conditions cause the game to check for collision, what happens when the check fails, and whether the failure state can be reproduced reliably. This is a form of reverse engineering: understanding the software beneath the game rather than the game as the designer intended it to be experienced. Speedrunners who documented major glitches — wrong warps in Zelda that skipped entire game sections, out-of-bounds areas in 3D games reached by clipping through geometry — were often doing work equivalent to software security research: finding conditions where the program behaved in ways its authors didn't intend.
The Legend of Zelda: Ocarina of Time has been the subject of speedrunning work for over twenty years. The current any% (complete the game in any way possible) world record is under seven minutes, compared to a typical first playthrough of 25 to 40 hours. The route involves a series of specific glitches: wrong warps that skip major game sections, out-of-bounds areas reached by specific movement sequences, and a file corruption technique that allows arbitrary code execution — writing values to memory addresses that control game flow in ways that skip to the game's ending sequence without completing the designed game. The techniques required to achieve this were discovered and documented through thousands of hours of collective research by the speedrunning community, each discovery building on previous understanding of the game's memory structure.
Speedrunning as community practice
The modern speedrunning community coalesced around video recording technology and streaming platforms. Early speedruns were documented by video cassette recordings and distributed on sites like Speed Demos Archive (founded 2003) — lists of run times with downloadable video proof for verification. The ability to record and distribute evidence transformed speedrunning from an individual practice into a social one: runners could watch each other's routes, identify differences that produced time differences, and discuss technique through forum posts that accumulated into community knowledge about how to approach a specific game.
Speedrunning Twitch streams created an audience for speedrunning that didn't exist before live streaming. Watching a skilled player complete a game in a fraction of the expected time, while explaining the techniques being used in real time, was compelling entertainment for players who had completed the same game normally. The commentary that speedrunners provided — explaining the memory addresses being manipulated, the frame windows for specific inputs, the probability of specific random elements — turned speedrunning into a form of technical education about games as software. Viewers who watched a Mario 64 speedrun learned more about the game's memory structure than they would have learned from any normal playthrough.
Games Done Quick, a charitable speedrunning event that began in 2010 as a small streamed marathon, grew to attract millions of viewers and raise tens of millions of dollars for charity through bi-annual events. The charitable framing was partly genuine — speedrunners who participated were donating their time to causes they supported — and partly a sustainable commercial context for a competitive activity that had previously lacked institutional support. The growth of Games Done Quick from a hobbyist event to a multi-million-dollar charitable organisation was the most visible evidence of speedrunning's cultural legitimacy: a community practice that had begun as players competing to break games had become an entertainment form with a mass audience and philanthropic impact.
What speedrunning revealed
Speedrunning's most interesting contribution to game culture is epistemological: it demonstrated that games are software, and that software behaves according to rules that may not align with the experience the designer intended to create. The game as it appears to a normal player is a surface; beneath that surface is the code that produces the surface, and the code has properties that the surface doesn't display. A wall that cannot be walked through in normal play is a collision detection check in code; if the check can be bypassed, the wall can be walked through. The game that takes thirty hours to complete in designed sequence takes seven minutes when the code's vulnerabilities are understood.
This perspective on games — as software systems rather than as authored experiences — changes how the medium is understood. A game's intended experience is one reading of the system; a speedrunner's run is another reading of the same system, equally valid from a software perspective if less valid from a design intent perspective. The debate about whether a speedrun that uses arbitrary code execution to skip the entire game "counts" as completing it is a debate about whether games are defined by their designed experiences or by their software systems. The question has no agreed answer, which is why speedrunning communities maintain separate categories (any%, with major glitches, without major glitches, 100%) that accommodate different positions on what completing a game means.