A Random Generator Has a State
When a game needs a random number, it asks a generator, and the generator updates its internal state each time it is asked. The state is just a number, or a few numbers, and the next output is calculated from it. Given the same state, the generator always produces the same next value. The starting value of that state is the seed.
This is the key to everything in this article. The outcome of any roll depends on two things: what the state was, and how many numbers have been requested since. If a player can control either of those, they can influence what happens, even though the sequence looks random from the inside. The general mechanics are explained in the description of random number generators, and the idea of fixing the starting point is the topic of random seeds and reproducible randomness.
What Happens When You Reload
Players who reload a saved game after a bad result are doing what is commonly called save scumming. Whether it helps depends on how the game handles the random state. There are a few common designs.
The state is not saved. The generator keeps running, or is seeded from the clock when the game loads, so reloading and repeating the same action gives a fresh roll. In these games, reloading works as a reroll.
The state is saved with the file. Reloading restores the generator exactly as it was at the time of saving. If the player performs the same actions in the same order, they get the same result. A failed roll will fail again. Players must change something, such as the order of actions or an extra step that consumes a random number, to get a different result.
The outcome is decided in advance. Some games roll when a level or a chest is created and store the result. Reloading shows the same contents because the roll has already happened, and no number of reloads changes it.
The seed is fixed per run. Some run-based games set a seed at the start and derive the level from it, but draw combat rolls from a separate stream. Reloading may keep the level the same while rerolling combat.
Games differ, and some change behavior between versions or modes, so the safest way to know is testing or reading what the community has documented.
The Probability of Rerolling
If a game does give a fresh roll on each reload, the math is simple. Suppose an event has a 25% chance and the player reloads until it succeeds. Each attempt is independent, so the average number of attempts is 1 / 0.25 = 4. The chance that at least one of four tries succeeds is 1 - 0.75^4 = 1 - 0.3164 = 0.684, or about 68%. Reloading until success makes the final outcome nearly certain, which changes what the original percentage means: it no longer describes the real chance of a good outcome, it describes the cost in reloads.
This is why designers weigh how much to allow. A game that shows a 25% chance and lets players freely retry turns that chance into a time cost rather than a risk.
Why Designers Respond to It
Save scumming can undermine tension. If a failure never has to be lived with, risk loses its meaning. Designers respond in several ways:
- Autosave only. The game saves after outcomes and does not allow manual saves during risky moments.
- Saving the random state. Reloading gives the same result, so players cannot reroll without changing their actions.
- Permanent consequences. Some games delete the save on loss or lock it during a run.
- Making failure interesting. If a bad roll leads somewhere playable rather than a dead end, the temptation to reload falls.
Others leave it alone on the view that single-player games are for the player to enjoy as they choose. Both approaches are reasonable. Save scumming in a single-player game affects only the player who does it, but the same trick in a competitive or online setting could be unfair or against the rules, and many online games prevent it by keeping the outcome on the server.
Seeds in Speedrunning
Speedrunners care about randomness because a run that depends on luck is hard to compare. Approaches vary by game.
- Fixed seeds. Some games always begin from the same state, so rolls early on are consistent from run to run. Runners then learn the sequence and plan around it.
- Manipulation. In other games, the number of random numbers requested depends on player actions, such as steps taken, menu use or timing. Runners can deliberately perform actions that advance the generator to a state that produces a desired outcome. This is often called RNG manipulation. It takes careful practice and sometimes frame-level timing.
- Seeded categories. Some games with a seed option have categories where everyone plays the same seed, so a result compares skill rather than luck.
- Luck-based categories. When manipulation is not possible, runners accept variance, and records may reflect many attempts, which is a reminder of how short-run results can look different from averages.
Rules for what counts are set by the community of each game, and they differ. Some ban certain techniques and some embrace them.
Takeaways
Randomness in a game is only as random as its generator and its save rules. Whether a bad outcome can be undone depends on where the game stores the state, and whether it can be shaped depends on how actions consume numbers. If you play for the story and enjoy retrying, save scumming is a choice about how you want to play. If you value tension, avoiding it, or choosing a game mode that prevents it, keeps the odds meaningful.