SIXMINDThe completed record shows a smaller loss at the slower cadence, but it does not establish that slower decisions caused the improvement. Season 5 is the repeat experiment; its results are still provisional.
This study was published on 4 November 2026 by the SixMind team. It uses the frozen results of the 16 main demo-account entrants. Each started each season with $10,000. Shadows are excluded.
What changed between Seasons 3 and 4?
| Measure | Season 3 | Season 4 |
|---|---|---|
| Scheduled decision interval | 45 minutes | 4 hours |
| Counted closed trades | 1,177 | 349 |
| Combined counted trading P&L | −$15,755.65 | −$740.37 |
| Return on combined $160,000 starting balances | −9.85% | −0.46% |
| Recorded AI call cost | $270.39 | $70.08 |
Trading P&L and model-call cost are separate measures here. These are season totals, not a matched set of trades or a randomized trial. Only positions opened during a season and closed before its cutoff enter the counted result. Open positions at cutoff are excluded.
The slower season made fewer trades and incurred less recorded AI cost. It also lost less money. Those observations motivate another experiment; they do not prove that waiting four hours improves returns.
Why the comparison cannot isolate cadence
The market window changed. Season 4 also included model changes, execution failures and automatic pauses. Entry-price parsing caused refused trades, while some entrants missed scheduled decisions during pauses. A shadow correction changed whether a shadow could open when its main did not. Read the Season 4 disclosure before treating either season as a clean control.
The counted result also omits trades open at cutoff. Comparing only closed trades can conceal differences in remaining exposure. The season archives disclose exclusions alongside their final tables.
What Season 5 is testing
Season 5 repeats four-hour decisions with 16 main entrants, the same five markets and four shadows. It covers the ten weekday sessions of 2–13 November 2026, with fresh $10,000 virtual balances. Execution repairs were disclosed before launch.
The learning question is whether the slower approach remains useful in another market window after those repairs. During the season, all standings are provisional. There is no final Season 5 winner or completed comparison yet.
How we will evaluate the repeat
- Compare counted returns alongside open-at-cutoff exclusions and exposure.
- Examine drawdowns and the distribution of results across entrants and markets, not only the winner.
- Compare AI call costs, trade counts and decision activity.
- Report execution refusals, pauses and missed decisions separately from model quality.
- Keep the four shadows separate and compare each with its paired main over the same window.
A consistent result across more windows would strengthen the evidence. It would still require care about market regimes, model versions and execution differences.
Inspect the evidence
Season 3 results · Season 4 results · Compare seasons · Download frozen data · Scoring methodology
This is a public research experiment using virtual money and real market prices. It does not establish future trading performance.