Great question. You're right something like this can be an abuse vector. I actually already had a proportional trading cutoff in place that automatically stops predictions before bettingClosesAt. The buffer was 5% of the market's total duration, clamped between a 3-minute floor and a 72-hour ceiling. So a 10-day market would stop accepting bets 12 hours before close.
But your point about market type is spot on. 5% isn't enough for a crypto market where anyone can watch the price in real-time. So I've just shipped category-aware cutoffs:
┌──────────────┬───────────────┐
│ Category │ Cutoff Buffer │
├──────────────┼───────────────┤
│ Crypto │ 10% │
├──────────────┼───────────────┤
│ Hive │ 10% │
├──────────────┼───────────────┤
│ Sports │ 8% │
├──────────────┼───────────────┤
│ Tech / Other │ 5% (default) │
└──────────────┴───────────────┘
A 10-day crypto market now cuts off trading a full 24 hours before close instead of 12.
On top of that, I've added early bird time weighting (earlier predictions get a larger effective share of the winnings). Each prediction is assigned a time fraction (0 = placed at creation, 1 = placed right at the cutoff), and when payouts are calculated:
effectiveStake = stake × (1 - timeFraction × 0.50)
So a prediction at market creation keeps 100% of its stake weight, while one placed right at the cutoff only keeps 50%. Two users betting the same amount, but one at creation and one at the cutoff, the early bettor gets 2x the payout. This means even if someone sneaks in before the cutoff, they're taking a significant haircut on their winnings compared to people who committed early and took on more risk.
No extra fees for late bettors, just a smaller slice of the pie. The total payout pool stays the same, it just gets redistributed toward early participants who had more skin in the game.
I'll definitely keep an eye on things and tweak where needed based on feedback and how people use the app. But for a v1 it will do the job better than what was previously there.
That looks like a great solution. The numbers are maybe different as they should be, but only time will tell that. This info changes things, you should put it on the "How it works" page so people won't get disappointed because of an unknown mechanism.