All field notes

Security · Retrospective

Anti-cheat when rewards are real: shipping PvP

A game controller inside a glass inspection frame, a blue light scanning across it.
Concept illustration · AI-generated

On 22 April the team shipped Playermon version 3.0.0. The headline was PvP Battleground, a mode where players assemble a team of Playermons and challenge each other directly. It had been a while coming. Registered testers had been playing the PvP alpha since early February, and weekly PvP game nights on Discord through March and April gave us a steady stream of matches and feedback.

The same release carried a line that drew less attention. In the patch notes, PvE battles gained "significantly upgraded security and anti-cheating systems". This post is about that line.

Why PvE needed it

PvE was where the rewards were. Each Anchorland season ended with a leaderboard paid in SGEM. The table we posted in February paid 50,000 SGEM to the champion and paid every place down to 50th. A leaderboard that pays is a target. A cheat in a game without rewards spoils someone's evening. A cheat on a paid leaderboard takes SGEM that should have gone to a player who earned it, and it teaches everyone else that the table does not mean what it says.

I am not going to describe what our checks look for. Explaining a detection rule in public mostly helps the people it is meant to catch. What I can describe are the principles I hold any game with real rewards to, because they are not secret and they decide where the engineering effort goes.

The client is the player's machine

Everything that runs on a player's device is under the player's control. They can read it, change it, and replay or forge the messages it sends. A game that pays rewards has to treat every report from the client as a claim, not a fact.

In practice that comes down to a few rules:

  • The result is decided where the player cannot edit it. If the client says "I won" or "I dealt this much damage", that is input to be checked, not a score to be recorded.
  • Randomness that affects rewards does not come from the client. A player who can choose their own random numbers will choose good ones.
  • Rules are enforced at the point of payment. Whatever the interface allows, the service that credits a reward checks again that the reward is owed.
  • Detection is paired with review. Some patterns only look suspicious in aggregate, and a person should look at the evidence before an account loses what it holds.

None of this is new, and none of it is specific to blockchain games. What changes when rewards are tokens is that a mistake is hard to reverse once a reward has been paid out to a player's wallet. Catching a forged result before payment is far cheaper than chasing it afterwards.

Three details in the same release

Three smaller items in version 3.0.0 fit the same thinking.

The first is fast-forward. Players could now speed through the execution phase of a battle. A fast-forward button is a quiet test of a design: if playing a battle faster could change who wins, the result was being decided in the wrong place. Playback speed should affect what the player sees, never the outcome.

The second is the off-season. Between competitive seasons, the whole Anchorland map was unlocked, with no leaderboards and no rewards. That gives players room to explore and try strategies with nothing at stake, which also means there is nothing to gain by cheating. Separating practice from payment is a security choice as much as a design one.

The third is a change to the Intelligence stat. A defender with higher Intelligence than its attacker now takes less damage from that attacker's later attacks. Changes like this are why anti-cheat cannot be a fixed list of thresholds. Every rule change alters what a legitimate result looks like, so the checks have to move with the game.

Testing where nothing is at stake

On 24 April, two days after the release, we announced support for Polygon's Amoy testnet. The Mumbai testnet, where we had tested most of our features before deploying them, was going offline. A testnet is where a team can be wrong without anyone losing anything, which makes it the right place to try to break a feature. A paid leaderboard is the wrong place to find out that a feature can be broken.

PvP shipped as a way to challenge friends directly. The rules above apply to it as much as to PvE. A result is worth only as much as the checks behind it, and the checks belong where the player's own device cannot reach them.