A speedrunner learns a game the way a musician learns a piece: section by section, with a timer running and a record of every attempt. The work starts long before the timer does. Runners study routes, break the game into segments, and practice the hard parts in isolation until they stop being hard.
The payoff is patience, not lightning hands. A serious run of a single game can take months of speedrun practice, most of it spent on a handful of seconds. Here is how that process actually works, from the first route sketch to the splits file that tracks every improvement. We covered a connected angle in From Itch.io Demo to a Million Copies: How Small Games Actually Get Found.
None of this happens in a vacuum, either. The categories runners compete in shape what a route even tries to do, as our guide to speedrun categories explains, and the community that judges and celebrates these runs has its own long history, including the charity marathons that turned speedrunning into a spectator event.
Where does a route come from?
A route is the plan: which rooms to enter, which items to skip, which cutscenes to trigger and in what order. Nobody invents one from scratch in a vacuum. The first step is usually reading. Leaderboards show what the current record looks like. Forums and Discord servers hold years of accumulated notes on tricks, skips and dead ends. A newcomer inherits a route, runs it, and only then starts asking what could change.
Route discovery is largely a process of subtraction. Every room, every pickup and every line of dialogue costs time, so the runner asks one question over and over: what happens if I don't do this? Sometimes the answer is nothing, and a whole section of the game falls out of the plan. Sometimes the game refuses to progress, and that failure itself becomes knowledge worth writing down.
Route theory also depends on the category. An any% run cares about reaching the credits as fast as possible, so it will happily skip content a 100% run must visit. Glitchless routes trade those skips for consistency. The same game can host several entirely different route documents, each with its own notes, its own community arguments and its own history of abandoned ideas. For related coverage, see Any%, 100% and Glitchless: How Speedrun Categories Actually Work.
Why break a run into segments?
A full run is too large to improve all at once. So runners slice it. A segment might be a single room, a boss fight, or a stretch of platforming between two checkpoints. Each segment gets a name, a target time and its own practice loop. The runner plays that slice repeatedly, ignoring the rest of the game entirely, until the target feels routine.
This is where the real speedrun practice happens. A tricky jump that costs five seconds when flubbed might take hundreds of attempts to land reliably. Practicing it inside the full run would mean replaying twenty minutes of game to reach it each time. Segmenting removes that cost. Load a save, try the jump, note the result, reload. The feedback loop shrinks from minutes to seconds.
Segmenting has a second benefit: it separates luck from skill. Some segments depend on random elements, like enemy behaviour or item drops. Isolating them lets a runner measure how much a segment varies and decide whether chasing a lucky outcome is worth the resets. That decision, made deliberately rather than on tilt, separates a planned run from a hopeful one.
What do splits actually track?
Splits are the run's ledger. A timer program records the time at each segment boundary, then compares it against the runner's best or average performance at that same point. After a run, the runner can see exactly where the attempt was won or lost, segment by segment.
Two comparison numbers matter most. The personal best shows the cumulative time at each split from the runner's best full run. The best-ever split, sometimes called a gold split, shows the fastest the runner has ever completed that segment in isolation. The gap between those two lines is the runner's headroom, and it is the most honest answer to the question every runner eventually asks: where is my time actually going?
Splits also discipline the practice plan. If one segment consistently runs seconds behind its gold while the rest are tight, the next practice session has a target. Instead of grinding the whole game, the runner grinds that one room. Improvement becomes a scheduling problem rather than a vague hope.
How does practice actually get scheduled?
Ask a room of runners how they train and you will get competing answers, and the disagreement is genuine rather than one side being wrong. Some swear by full runs from start to finish, arguing that segment practice builds false confidence because it removes the fatigue and nerves of a real attempt. Others barely do full runs at all, treating them as exams rather than study sessions.
What most approaches share is a structure. A common pattern looks like this:
- Warm up on the segments most likely to be flubbed, so the hands are ready before anything is measured.
- Drill the one or two weakest segments with repeated isolated attempts.
- Run a full attempt, treating it as data collection rather than a performance.
- Review the splits and pick tomorrow's practice target from the evidence.
Reset discipline is part of the routine too. A runner who quits every attempt at the first mistake never practices the late game under pressure. A runner who never resets wastes hours on doomed attempts. Where that line sits is a personal judgement, and runners argue about it the way musicians argue about scales versus repertoire.
There is also the unglamorous layer: input practice away from the game, route notes reviewed between sessions, and video review of past attempts. Watching your own run back is uncomfortable and useful in equal measure. The timer does not lie, but memory does.
What does this mean for someone starting out?
The practical entry point is smaller than it looks. Pick a game you already know well, because familiarity with the normal game is the raw material route planning works with. Pick a category with an active leaderboard and published rules, so your runs mean something. Then run the game once with a timer, badly, and let the splits tell you what to practice first.
Our analysis of how runners improve points to one consistent theme: the timer is a feedback tool, not a scoreboard. Runners who treat each attempt as information improve steadily. Runners who treat each attempt as a referendum on their talent tend to burn out. The split file, unglamorous as it is, is the difference between the two.
Speedrunning's broader culture, from its mod-scene roots to its charity marathons, gets plenty of attention. But the day-to-day craft looks like this: a route document, a segment list, a splits file with a few hundred entries, and a runner who knows exactly which five seconds they are hunting today.
Where the learning ends up
The evidence of this process is public. Leaderboards, route wikis and Discord archives preserve years of collective problem-solving, and every record run sits on top of them. What remains unknown in any given game is simply the next skip, the next route revision, the next half-second someone has not found yet. That open frontier is why a game released a decade ago can still produce a first.
Sources: themetrorailguy.com · en.wikipedia.org · in.linkedin.com · linkedin.com