The instinct, and why it is wrong

The common reasoning goes: if hazards are in the same place every time, players will memorise the level, and then the level is solved and no longer challenging. Therefore randomise.

The flaw is in the last step. A player who has learned a level has acquired a real skill, and that acquisition is the most satisfying thing an arcade game offers. Preventing it does not preserve challenge; it replaces a skill you can develop with an attrition you can only endure.

Candy Factory contains no randomness at all. Every piston, roller and stamping arm runs on a cycle locked to the floor's master clock. Any section that killed you can be crossed reliably once you have watched it for one full cycle, and players consistently describe it as one of the most satisfying games in the library.

What determinism gives the player

  • Failure is diagnosable. If you were crushed, you mistimed it. There is no other possibility to consider.
  • Practice compounds. Each attempt adds knowledge that remains valid on the next one.
  • Observation works. Watching a section teaches you how to cross it, which is a legitimate and rewarding activity.
  • Mastery is achievable. A floor can be cleared perfectly, and doing so feels earned rather than lucky.
  • Pausing is useful. Studying a cycle costs nothing and produces real progress, which is why our pause is instant and unpenalised.

That third point is worth emphasising. In a deterministic game, the correct response to a section that killed you twice is to stop and watch it. In a random one, the correct response is to try again and hope. Only one of those is a skill.

Advertising keeps Candy Arcade free. Ads never cover gameplay or controls.

What randomness actually produces

Random hazard placement does increase failure rate, which is why it looks like difficulty. What it produces is different.

  • Failures become ambiguous. Was that bad play or a bad roll? The player cannot tell, so they cannot correct.
  • Practice stops compounding. Thirty attempts teach roughly what one teaches, because the specifics never repeat.
  • Unfair situations occur. Unless carefully constrained, random placement eventually produces an impossible configuration.
  • The skill ceiling stays low, because there is nothing specific to get better at.

The last point is the important one. Randomness raises the floor of difficulty and does nothing to the ceiling, which is the opposite of what a well-designed challenge should do.

The middle ground: verified generation

Determinism does not mean hand-building everything. Generated content can be learnable if what is generated is the arrangement rather than the behaviour.

Candy Tunnel assembles formations from a set of verified templates. The sequence is new each run, so the game stays fresh, and each individual template is learnable, so encountering one you recognise is rewarded. That combination - novel sequences of known elements - is what most good procedural games actually do.

The rule: randomise which patterns appear and in what order. Never randomise how a pattern behaves once it has appeared.

Hazards with stated rules

The other middle ground is behaviour that varies with the situation but follows a rule the player knows.

Candy Monster Escape's creatures are not deterministic in position - they respond to where you are - but each one follows a single stated rule. Gumdrops move straight at you. Jellies aim at where you will be. Liquorice creatures patrol until you cross their sight line and then charge in a straight line without turning.

Those rules are published on the game page, which means the creatures' behaviour is predictable even though their positions are not. The player can exploit a gumdrop's directness by leading it into an obstacle, which is a skill that exists only because the rule is known.

When randomness is fine

Randomness is not universally wrong. It works well where it varies content rather than outcomes.

  • Which collectible type appears in a slot - no fairness implication.
  • Cosmetic variation: background elements, particle colours, decoration.
  • Which of several verified patterns comes next.
  • Spawn positions within a verified-safe region.
  • Power-up selection from a list, where every option is useful.

The test: if a random roll can determine whether the player survives a situation they played correctly, the randomness is in the wrong place.

Common mistakes

  • Randomising to prevent memorisationMemorisation is a skill being acquired. Preventing it replaces skill with attrition.
  • Randomising hazard behaviourIt makes failures ambiguous, so the player cannot correct anything.
  • Assuming randomness raises the skill ceilingIt raises the failure floor and leaves the ceiling alone, which is the wrong way round.
  • Generating without verificationUnconstrained randomness eventually produces an impossible configuration, which the player will blame themselves for.
  • Hiding hazard rulesBehaviour can vary if the rule is known. Unstated rules make varied behaviour indistinguishable from randomness.

Practical tips

  • In deterministic games, stop and watch anything that has killed you twice.
  • Use pause to study a cycle; it is instant, free and intended.
  • In template-based games, learn the individual formations - you will see them again.
  • Where hazards follow stated rules, exploit the rule rather than outrunning the hazard.
  • If a game's failures feel ambiguous, check whether its hazards are actually learnable.

Key takeaways

  • Memorisation is skill acquisition, not a problem to design against.
  • Determinism makes failure diagnosable and practice cumulative.
  • Randomness raises the failure floor without raising the skill ceiling.
  • Randomise which pattern appears, never how a pattern behaves.
  • Stated behavioural rules give predictability without fixed positions.

Frequently Asked Questions

Is it bad if players memorise a level?

No. A player who has learned a level has acquired a real skill, and that acquisition is one of the most satisfying things an arcade game offers.

What is wrong with random hazards?

They make failures ambiguous, so the player cannot tell bad play from bad luck and therefore cannot correct anything. Practice stops compounding.

Is Candy Factory really fully deterministic?

Yes. Every machine runs on a cycle locked to the floor's master clock, so any section can be learned and crossed reliably.

Can generated content still be learnable?

Yes, if you randomise which patterns appear and in what order but never how a pattern behaves. Candy Tunnel works this way.

Where is randomness actually fine?

Anywhere it varies content rather than outcomes: collectible types, cosmetic variation, pattern order, spawn positions within verified-safe regions.