Today we are switching on a full quality-control layer across the platform. Every submitted response now passes through 14 validation methods - from speed and straight-line detection to device fingerprinting and geo-validation - before it is approved and rewarded.
Why we built it
Bad responses are expensive twice: researchers pay for data they cannot use, and honest participants compete with bots and speeders for quota slots. Industry studies consistently find that 10-15% of unfiltered online survey responses show at least one fraud signal. Our own baseline measurement came in at 12.4% - almost exactly in line.
What researchers get
- •A Quality Control panel on every survey - toggle individual checks (speed, straight-lining, duplicate IP, text quality, pattern analysis, outlier detection) per study
- •An overall quality score (0-100) on every response, with per-check breakdowns and flags
- •Sort and filter responses by quality score, not just by date
- •Automatic manual-review queue for anything scoring below 60
What participants get
- •A transparent trust score that rises as responses are approved
- •Clear rejection reasons, so an honest mistake is a lesson rather than a mystery
- •Faster reward release - clean responses no longer wait behind suspicious ones
The design principle
“No single check decides anything. A response is judged by the portfolio of evidence, and a human reviews anything borderline.”
- Tayqun engineering team
The full methodology - what each check measures and how thresholds were chosen - is documented in The State of Online Survey Quality report.
Coming next
Per-question timing collected client-side, cross-survey device fingerprinting improvements, and an appeals flow so participants can contest a flag directly from the submission page.