The 3% Rule Meets The 30-Second Round - What Two 2026 Papers Say About Overconfidence And CCTV Games

Two 2026 papers just landed on the same conclusion from opposite directions. Gomez-Cram, Guo, Jensen and Kung looked inside Polymarket's transaction universe and found that only about 3% of traders make money consistently. A Journal of Gambling Studies paper published on 18 February 2026 tested seventy individual investors and found that overestimation - one of three overconfidence components - is the single cognitive marker that separates low-risk from moderate-to-high risk of excessive trading. Put those two findings next to a 30-second CCTV game round and the design argument writes itself. The game does not give overestimation anywhere to hide.

What the Polymarket data actually shows

SSRN's summary of Prediction Market Accuracy: Crowd Wisdom or Informed Minority? lays out the numbers. Accuracy comes from a persistent 3% who react instantly to public news, correct pricing violations, and trade against the crowd's mistakes. The other 97% do not trade any better than a coin flip. Theis Jensen puts it plainly - if you do not have an informational or programming edge, prediction markets are not a long-term investment vehicle.

The favorite-longshot bias makes it worse. Prediction Frontier's July 2026 guide and the Burgi, Deng and Whelan Kalshi analysis of 300,000+ contract outcomes show cheap longshots at 5 cents resolve Yes closer to 2-3% of the time. Takers - traders who accept the best quote - lose about 32% on average, while makers lose about 10%. That gap is the price of immediate certainty in a market where you have already been mispriced by prospect theory.

The overestimation finding matters more than the crowd number

The Journal of Gambling Studies paper tested seventy active retail investors using a simulated technical-analysis task. Overestimation - believing your absolute performance is higher than it is - predicted moderate-to-high risk of excessive trading. Overplacement - believing you rank higher than others - did not. The pattern matches the classic illusion-of-control literature in gambling research and gives a specific cognitive lever to design around.

Where CCTV games sit relative to these biases

Three properties of the format matter. Round length is fixed at roughly 30 seconds - the outcome resolves before overestimation can find room to grow through a session of anchoring on unrealised losses. The payout is a discrete count from a bounding-box detector, as the Vision Evals piece from 30 July argued - there is no interpretive gap where a player can convince themselves the near-miss was really a signal. And the game does not price its own probabilities. There is no maker-taker gap, no market depth for herding to hide in, no favorite-longshot wedge because the odds table is not set by other players.

The disposition effect - holding losers to avoid crystallising a loss - is the closest bias that still applies. That is why the responsible play stack matters. Session limits set before a session starts remove the in-the-moment decision the bias exploits. A pre-committed exit is the operator-side answer to the same behavioural finding.

What this argues about the category

The 3% rule is not a marketing win for CCTV games because these are not prediction markets - the format does not promise you can outperform anyone. It is a design floor. When even a data-rich market with public news, five-minute settlement windows and skilled programmatic traders bleeds 97% of accounts to zero-alpha outcomes, the case for a format that resolves in 30 seconds on a numeric count with a fixed odds table starts to look less like a limitation and more like a feature. The EU AI Act piece from Monday made the transparency argument. This is the same argument from the psychology side.

What to watch

Whether the next round of prediction-market research isolates round length as a specific behavioural variable. Whether operators publishing CCTV games start reporting session length and stake-per-round distributions, which would let researchers test the same biases in the live product. Whether the UKGC and MGA add cognitive-load benchmarks to their game certification frameworks in 2027 - a natural next step after the transparency rules that have already landed.

Where to play remains Stake, Roobet, Shuffle and Razed. Set a session limit before you press start.

Gambling involves risk - never bet more than you can afford to lose.