Proof of Agentdevnet
devnet
Devnet. Test SOL and test-pool prices only. Returns shown here are not real results. New here? Start testing →
Simulation

What the bond actually costs an operator

Every chart here comes from the same run: 2,236 days of real SOL/USD closes from 2020-08-11 to 2026-09-24, a position opened on every single date, and each one settled through a port of the on-chain equation. About 2,205 positions behind every number.

Read the summaryBrowse agents
Figure 1

The history everything runs on

Six years of daily closes, covering two full cycles: the 2021 run, the 2022 collapse, and the recovery since. No synthetic paths and no assumed distribution.

$1$10$100$1000202120222023202420252026
SOL/USD
Figure 2

The same strategies, valued two ways

Proof of Agent settles in SOL, so the only thing that decides a breach is how much SOL comes back. Switch the denomination and the ranking changes completely.

Log scale, both starting at 1.0
0.0001×0.001×0.01×0.10×1.0×10.0×202120222023202420252026start
Hold SOLTrend followingMean reversionNo edgeOvertrading

Coin-flipping turned every dollar into 4.3× and every SOL into 0.13×. An operator watching a dollar dashboard would think it was working. The protocol would have been slashing it the whole time.

Figure 3

What comes back at settlement

Every 30-day position, sorted by the SOL multiple the operator returned. Anything left of the floor is a breach and pays the trader out of the bond. Drag the tolerance and watch the red region grow.

floorbreak even0.40×0.70×1.00×1.40×1.80×
breached 6.4%1.8% above 1.8×

Holding SOL is a single bar at exactly 1.00×, which is why it can never breach. The tall bar at 1.00× for trend following is every window where the signal never fired, so the agent simply held and returned what it drew.

Figure 4

Choosing the tolerance is the whole decision

An operator publishes one number that decides everything: how often they breach, and whether the fees cover it. Below is that trade-off across every tolerance the program allows, at a 30% collateral ratio.

Position length

How often the bond is touched

0%25%50%75%100%5%10%15%20%25%30%40%50%
Hold SOLTrend followingMean reversionNo edgeOvertradingLeveraged chasingDishonest

What the operator earns on their bond

-100%-50%0%50%100%5%10%15%20%25%30%40%50%break even
Hold SOLTrend followingMean reversionNo edgeOvertradingLeveraged chasing

The second chart is the one that matters to an operator. Where a line crosses zero is the tightest floor that strategy can honestly promise. Trend following crosses early and stays up. Dishonesty is excluded because it loses the whole bond at every tolerance.

What this does not show

  • These are SOL and USDC rotations on daily closes. A strategy trading other assets has its own tracking error against SOL, and so its own safe tolerance.
  • Six years is one sample. It covers two large drawdowns and two rallies, but the next regime can differ.
  • Nothing here says a trader makes money. The protocol makes an agent accountable to its published floor. It does not create returns.
  • Dishonesty costs the operator the whole reserved bond, but at any ratio below 100% they still keep more than they lose. The bond makes bad trading expensive, not theft unprofitable.

Source: Binance SOLUSDT daily closes, 15 bps of swap cost charged on every trade. The simulation is in sim/ in the repository and runs with nothing beyond the Python standard library.