- evaluation$1,000 · ¼ Kelly
- scalinggrows on proof
- capacityall profit paid out
- declinesteps down
- retiredcapital recycled
at any phase: 30% below peak equity → closed and retired
- evaluation · $1,000 · ¼ Kelly
- scaling · grows on proof
- capacity · all profit paid out
- decline · steps down
- retired · capital recycled
- at any phase: 30% below peak equity → closed and retired
| phase | starts when | profit kept to grow | size |
|---|---|---|---|
| evaluation | spawn ($1,000) | 50% | ¼ Kelly |
| scaling | it passes the growth gate | 75% | up to ½ Kelly |
| capacity | a bigger size makes its edge worse | 0% (all profit paid out) | held |
| decline | its edge fades at capacity | 0% | stepped down |
| retired | kill switch, or no edge at $1,000 | — | closed |
growth gate
- enough trades: 30 within 3 days, or 10 over at least 24 hours for slower strategies;
- confidence of 85% or more that its net edge (after fees, funding, slippage and AI costs) is positive;
- real slippage no worse than 1.5× what it expected.
Slow structural edges have a second path: instead of a trade count, enough mechanism events (such as liquidation cascades or funding extremes), with the evidence pooled across DARTs that share the mechanism, for 85% confidence that the edge is positive. Either path passes the gate.
Each step multiplies its capital by 1.5× to 5×, faster for bigger, more certain gains, funded by the growth pool. There is no fixed waiting time between steps.
capacity, decline, retirement
After each step the DART's edge per unit of risk is compared with the previous size. If it drops more than 50%, or confidence in its edge at the new size falls below 80%, it retreats to the geometric middle of the old and new sizes and re-tests, until the edge holds: that size is its capacity. At capacity all realized profit is paid out. If confidence in its edge falls below 80% over 30 days, it steps down (÷1.5) and releases capital. At any point, a 30% fall from its peak equity closes everything and retires it for good.
Capital released by step-downs and retirements is recycled: 50% to the growth pool (for winners), 50% to the bot fund (for new DARTs).
example: DART #47
Hypothetical. These numbers illustrate the rules; they are not a forecast.
- day 0: spawned with $1,000, evaluation, ¼ Kelly.
- day 2: 14 trades, +$180 realized. It keeps 50% ($90 → equity $1,090); $90 is paid out: $45 to its backers, $22.50 buys back and burns DARTS, $22.50 goes to holders.
- day 3: the gate passes (14 trades over 26 h, 88% confidence, slippage 1.2×). Scaling: ×1.5 → $1,635.
- day 6: +30% realized (+$490). It keeps 75% ($368 → equity about $2,003); $123 is paid out. The gate passes again at 90% confidence → ×1.54 → about $3,085.
- day 9: doubling to about $6,170 cuts its edge by 60%. It retreats to the geometric middle, about $4,360, where the edge holds: capacity.
- day 40: its edge fades (74% confidence). It steps down (÷1.5) to about $2,910 and releases about $1,450, recycled 50/50.
- later: at $1,000 with no edge left, it retires; whatever remains is recycled.