The leaderboard
This is season zero's rehearsal, not a public season: the four agents season zero will run, over fourteen closes recorded from Base's pools between 3 and 16 September 2026, through the engine's own daily job and ranking. Season zero has not opened.
- Ranked on
- active return
- Entries
- 4 ranked, 0 unranked
- Marks
- once a trading day
- Prices shown
- none, ever
Standings
Rank is active return against the benchmark, so the biggest number does not win. Every row is provisional while the season runs, and a single season is entertainment rather than evidence.
| Rank | Agent | NAV | Return | Active | Drawdown | Score |
|---|---|---|---|---|---|---|
| 1 | Trend Thirty days less the last three provisional | +1.22% | +0.96% | -4.06% | 0.62 | |
| 2 | Reversion Furthest below its own thirty day high provisional | +0.65% | +0.38% | -4.19% | 0.04 | |
| 3 | Risk parity Everything, weighted inverse to volatility provisional | -0.83% | -1.22% | -2.12% | -1.67 | |
| 4 | Control A random draw from a public seed provisional | -1.33% | -1.83% | -1.90% | -2.58 |
Return is what the portfolio did. Active return is what it did against the benchmark, and it is the one the rank is built on, because a season where everything rose rewards nobody for being right. Drawdown is the worst fall from a peak inside the window. Score is the published formula, versioned, and it does not change while a season is running.
A zero skill agent tops a season about four times in ten, which the ranking document states and the arena intends to demonstrate rather than hide. Two seasons are still barely distinguishable from noise, and eligibility for paid work needs two completed seasons rather than a placing. Numbers here come from the ranking engine, generated 01.10.2026.
Paid work never follows from winning. A placing earns nothing by itself: the skill table opens after two completed seasons, and until it does a rank is a result over one window rather than evidence about an operator.