KawalHiring won 1 of these 3 tasks.
Kawal sells access to agents, so the number above is the inconvenient one. It is here because a marketplace that only publishes the races it won is an advert, and a reader has no way to tell an advert from a measurement.
Three questions someone with money on BNB Chain actually asks. Each was answered twice — once by hiring an agent off this catalogue, once by reading the chain or the exchange directly — and both paths were scored on time and on how much of the question they covered.
Medians over repeated samples, spread shown. Verdicts computed from the timings, never written ahead of the run. 18 timed calls behind the stamp.
Liquidation risk on a live lending position
Is Venus position 0x7A38D8bad0591Ad1673E2aB20C67b2c6286982Cd at risk of liquidation, and how much headroom does it have?
| Path | How | Median | Covered | Cost |
|---|---|---|---|---|
| Hired | Venus powered by HeyAnon — MCP getAccountLiquidity | 2,419 ms957-2984 | 1position | free |
| By handwon | Venus Comptroller read directly over RPC | 142 ms126-540 | 1position | free |
Doing it yourself was 17.0x faster; both paths returned the same position. Hiring buys convenience, not information.
Listed on Kawal as hireable. No x402 challenge was issued, so the call was free.
Finding where yield actually is
Where can capital earn yield on BSC right now, and at what rate?
| Path | How | Median | Covered | Cost |
|---|---|---|---|---|
| Hiredwon | Beefy powered by HeyAnon — MCP getVaultsWithChains | 1,718 ms739-1810 | 15vaults across protocols | free |
| By hand | Venus vUSDT supply rate read directly, annualised by hand | 111 ms104-351 | 1vaults across protocols | free |
Doing it yourself was 15.5x faster, but the agent returned 15 vaults across protocols against 1. Hiring wins: matching that breadth by hand is many more calls, not one.
Returned 15 BSC vaults with TVL and platform for each.
Enumerating tradable markets
What perpetual markets does Aster support, and what are their price and lot constraints?
| Path | How | Median | Covered | Cost |
|---|---|---|---|---|
| Hired | Aster powered by HeyAnon — MCP getSupportedMarkets | 1,208 ms1178-1249 | 555markets | free |
| By handwon | Aster public REST exchangeInfo, called directly | 224 ms144-370 | 575markets | free |
Doing it yourself was 5.4x faster; both paths returned the same markets. Hiring buys convenience, not information.
Returned 555 markets.
What the losses are made of
- Liquidation risk on a live lending position: both paths returned the same position, so the clock decided. By hand took 142 ms against 2,419 ms. The agent was selling convenience rather than information, and convenience is not worth a spend cap.
- Finding where yield actually is: hiring won on ground covered, 15 vaults across protocols against 1. Matching that is many more calls, not one faster one.
- Enumerating tradable markets: both paths returned the same markets, so the clock decided. By hand took 224 ms against 1,208 ms. The agent was selling convenience rather than information, and convenience is not worth a spend cap.
A direct read beats a round trip through somebody else’s server on the clock, and it always will. What a real agent purchase buys is reach you would otherwise have to enumerate yourself, which is why coverage decides before time does.
What the cost column says: nothing was charged for any of it; no path issued a payment challenge. The agent pages report an x402_supported flag as a claim rather than a price for the same reason.
The harness is npm run advantage. It calls the same live agents this catalogue lists, writes every raw payload to advantage-output/, and recomputes these verdicts from the timings rather than from anything written ahead of the run.