Last week we noticed an influx of users all voting for the same answer repeatedly, leading to statistically impossible outcomes (e.g., 100% accuracy over 30 votes). 

So today we launched CollusionGuard 9000 🛡️ 

It uses ML to identify statistically anomalous clusters and adjust vote weights accordingly. For example, if you voted alongside 50 other people on all of the right answers for ten questions, your vote diminishes in value until your responses normalize.

People who fall into those clusters can still participate in boosted pools and win an equal amount by selecting the right answer. No banning required (for now). 

There’s risk in this approach. We expect a drop off in numbers from colluders - and some won’t consider this decentralized - but the overall integrity of the game should increase.

Co-founder @ponder | @survey | Ex   | colinjohnson.io

but what if those answers are just what seems right and it’s a coincidence

If it’s just a coincidence a few times, nothing changes. All stays the same. 

If it’s the same group of people 10+ times when there are four answers…almost statistically impossible. Cluster analysis is really good these days.