The Lab

Tens of thousands of games, for a single rule.

Before a rule goes into a product, we attack it: mass simulation, exhaustive computation, games around a table. None of it shows on screen. It's the reason a game still holds up on the thirtieth play.

The measurements

We remove nothing without proving it first

Removing is easy; proving something has to go is expensive.

84%of high-level games ended in repetition before the no-return rule
0–16%repetitions left once taking a move back was removed
57,947,401transitions explored to prove a sixteen-square board is perfectly fair
2%win rate for the hoarding strategy after the rebuild, across two thousand games
Six decisions

None of these came from instinct

Each one came from a number. Several run against what we believed at the start.

  • SwashThe no-return rule

    Without it, two cautious players slide the same piece between two squares forever. Across twelve thousand simulated games, up to 84% of high-level games ended in repetition. Repetitions are now 0–16%.

  • SwashThe 4 × 4 is solved

    An exhaustive endgame base: 9,879,533 positions and 57,947,401 transitions. No opening placement forces a win, which proves the game is perfectly fair.

  • Swash DeepCross-placement

    Letting each player lay out their own pieces decided the game before the first move. Each now places the opponent's Pearl — balance restored without adding a rule.

  • MercatoThe weakest link

    Two strategies crushed everything: three stars and filler, or buy cheap and sit on the cash. A score reading your weakest player pushed the hoarding strategy down to a 2% win rate.

  • OrbitThe hidden trajectory

    With a full preview of the curve, the game solved by sweeping angles until the line hit the target. Removing it forces you to build an intuition for gravity — which is the point of the game.

  • BankPar runs backwards

    We gave expert players more shots. Wrong way round: expert courses aren't harder, the players are better. Par now tightens as you improve.

The method

Always in this order

An unbalanced rule is fixed at stage two. After that it costs ten times more.

  1. Rules

    Written out in full, edge cases included: deadlock, repetition, draws, running out of time.

  2. Validation

    Balance measured across thousands of games, exact solving where the size allows it.

  3. Visuals

    Final and approved before a single interface is built.

  4. Interface

    Built once, on those visuals. Never twice.

  5. Integration

    Engine, network, release — then nine automated checks, any one of which blocks delivery.

The test bench

And what we threw away

Free placement of pieces. A diagonal piece on the 4 × 4. A clock increment. A cap on bounces. A tenth arcade game.

Free placement decided the game before the first move. A diagonal piece broke the simplicity of the small board. A three-second increment per move handed the computer an advantage it hadn't earned. Capping bounces removed the whole point of the game.

And the tenth arcade game works fine — it just isn't as good as the other nine, so it stays out of the catalogue. What has been validated is never touched again: it's the only way to move forward without undoing what worked last week.

11prototypes buried
14,000+games simulated first

Easiest thing is to try one

The games run in your browser, free. The apps are in the catalogue.