Anti-ban humanization — what actually matters

We get asked "will I get banned?" on every sales call. The only honest answer is "maybe." Anyone telling you "0 bans ever" is selling fiction.

What we cando is make a bot session look as little like a bot session as possible. Here's the model we ship.

The four pillars

1. Tap jitter

A real human doesn't tap the same screen pixel twice. The thumbprint smears across a small ellipse around the visual center. Our taps add ±3–7 pixels of jitter, drawn from a Gaussian centered on the target. Inconsistent enough to look human; tight enough to always hit the actual button.

2. Dwell variance

Real taps aren't instantaneous. A human holds for 30–80 ms depending on intent and finger weight. We sample dwell from a log-normal distribution per profile. Heavy-handed profiles dwell longer. Bursty profiles dwell less. Both look like real users.

3. Idle micro-pauses

Real users get distracted. Mid-routine, they pause for 800 ms while a notification dings, or 4 seconds while their kid asks a question. We inject micro-pauses on a Poisson schedule — rare enough not to slow throughput materially, frequent enough that a perfectly metronomic loop never appears in the logs.

4. Behavior-profile rotation

The same user doesn't play the same way every session. We define three profiles — "careful", "normal", "fast" — each with its own jitter / dwell / pause distributions, and rotate between them at session boundaries. Two consecutive sessions never share the same input fingerprint.

What we don't do

• No memory hooking.We don't touch the game process. Memory readers and packet injectors are the fastest way to get banned.
• No exploit chains. The bot only does what a normal player can do — taps and swipes through the official UI.
• No claim of zero risk. See below.

Where bans actually come from

From talking to banned users, the pattern is rarely "the bot got detected." It's usually:

1. Account sharing across IPs — vendor and the customer both signed in
2. Multi-accounting — many accounts on one device, easily fingerprinted
3. Fly-too-close behavior — buying gold off-platform, swapping accounts in trade

The bot itself is rarely the trigger. We can't rule it out entirely, but the easy wins are: don't share the account, run on your own hardware, don't do anything that'd get a manual player banned either.

How to lower your risk further

• Run on a side account first while you tune routines
• Use your home IP, not a VPN or VPS shared with strangers
• Don't multi-account on the same device beyond what you can manually log into
• Keep updated — we tune the model based on observed signals

Closing

We can't promise zero risk. We canpromise that we're honest about it, that we publish the model, and that we ship updates as the game's detection layer evolves. That's what we wanted from the bot we ourselves wished existed before we built this one.