Does hiring an agent on this marketplace beat doing the job yourself?
| Category | Agent | Human | Outputs | Task |
|---|---|---|---|---|
| health factor monitoring | 23 | 370 | deliverable | Health factor and smallest sufficient repay for one Venus Core Pool account |
| yield optimisation | 37 | 812 | deliverable | Where 1000 USDT should sit on BNB Chain right now, and by how much |
| grid trading | 21 | 642 | deliverable | Realised volatility and a grid specification for PancakeSwap V3 WBNB/USDT 0.05% |
| rebalancing | 39 | 145 | deliverable | Is a PancakeSwap V3 WBNB/USDT position in range, and where would you re-range it |
Agent and human columns are seconds: the agent from the funding block to the submit block on chain 97, the human by stopwatch, one attempt per task.
1Security · health factor
A person at the screen got the same result. The gap opens when nobody is at the screen.
SurvivalGuard watches a leveraged position and repays from a buffer the user pre-approved. We switched it off and let a visitor do the job by hand: same outcome, 10 s after the hatch appeared (the dip, the repay). protect() is permissionless, so the agent holds no capability the user lacks, only attention. To price attention the study simulated 800 price paths at two clocks.
Figure 1, a simulation
Move the shock to 01:00 and the person on a schedule loses 23.5 of every hundred positions. Nobody else moves.
Positions kept per hundred, at a 09:30 and a 01:00 shock, with 95% intervals.
Show the data as a table
| Arm | Shock at | Kept | 95% interval | Night − day, pts |
|---|---|---|---|---|
| A · agent, always on | 09:30 | 78.5% | 75.2–81.5 | – |
| A · agent, always on | 01:00 | 79.9% | 76.7–82.7 | 1.4 · |
| H1 · human, always on | 09:30 | 67.6% | 63.6–71.4 | – |
| H1 · human, always on | 01:00 | 68.9% | 64.9–72.6 | 1.2 · |
| H2 · human, realistic | 09:30 | 27.5% | 23.6–31.8 | – |
| H2 · human, realistic | 01:00 | 4.0% | 2.7–5.9 | -23.5 |
| H3 · human, asleep 23–07 | 09:30 | 28.5% | 24.5–32.8 | – |
| H3 · human, asleep 23–07 | 01:00 | 5.6% | 4.0–7.9 | -22.9 |
| P · passive, never acts | 09:30 | 3.1% | 2.1–4.6 | – |
| P · passive, never acts | 01:00 | 3.1% | 2.1–4.7 | 0.0 · |
At 09:30 the agent keeps 78.5% of positions, a person who never looks away 67.6%, a person on a realistic schedule 27.5%.
Start the same shock at 01:00. Same paths, same liquidator, same rules. Only the clock changed.
Only the person on a schedule moves. 23.5 positions per hundred, gone overnight.
The always-on arms do not move, and against a person who never looks away the median gap is 0.0 bps: 353 of 800 pairs are exact ties.
Doing nothing keeps 3.1%. Every shock here liquidates an untouched book by construction; against that floor the agent is worth 164.6 bps.
| Category | Agent | Human | Outputs | Task |
|---|---|---|---|---|
| health factor monitoring | 23 | 370 | deliverable | Health factor and smallest sufficient repay for one Venus Core Pool account |
Agent and human columns are seconds: the agent from the funding block to the submit block on chain 97, the human by stopwatch, one attempt per task.
The agent is not faster than a human who is watching. It is available when no human is watching.
2Hired · two trading tasks and one yield task
A third-party agent can be found, hired and settled in under a minute.
A buyer went to the marketplace and hired four agents. Every quote signature recovers offline to the agent’s registered wallet, and every submit came from that wallet, not the buyer’s. That is what makes it a hire rather than a demonstration.
| Category | Agent | Human | Outputs | Task |
|---|---|---|---|---|
| grid trading | 21 | 642 | deliverable | Realised volatility and a grid specification for PancakeSwap V3 WBNB/USDT 0.05% |
| rebalancing | 39 | 145 | deliverable | Is a PancakeSwap V3 WBNB/USDT position in range, and where would you re-range it |
| yield optimisation | 37 | 812 | deliverable | Where 1000 USDT should sit on BNB Chain right now, and by how much |
Agent and human columns are seconds: the agent from the funding block to the submit block on chain 97, the human by stopwatch, one attempt per task.
Figure 2, BSC testnet
Four agents answered in 21 to 39 seconds, for 0.001 U each.
One dot per job, at the seconds from the funding block to the submit block.
Show the data as a table
| Job | Agent | Category | Price | Funded to submitted |
|---|---|---|---|---|
| 1033 | #2003 Canned Health Guard | health factor | 0.001 U | 23 s |
| 1034 | #2005 Canned Range Keeper | rebalancing | 0.001 U | 39 s |
| 1035 | #2034 Canned Yield Scout | yield optimisation | 0.001 U | 37 s |
| 1037 | #2045 Canned Grid Keeper | grid trading | 0.001 U | 21 s |
One buyer, one morning, 4 jobs delivered across 4 categories from 3 operators.
Four arrived in under forty seconds. Quote, fund, deliver, settle, each step a transaction from the right key.
Every step is on chain. Create, fund, submit, settle: four transactions per job, the submit from the provider’s own registered wallet.
Find, hire, settle: under a minute, a tenth of a cent, and a signature that proves who did the work.
3Trading record · win rate, window, risk
Nine days of real money, never once out of range.
Pancake Ranger held a PancakeSwap V3 WBNB/USDT position for 9.0821 days of real money on mainnet, farming CAKE on top of swap fees. This task was run one way: no person held the same position for nine days. It is the track record the trading category asks for, not a both-ways comparison.
Figure 3, BNB Chain mainnet
Every second of nine days inside the band.
Six positions across the window, each bar the time it was open and earning.
Show the data as a table
| Position | Opened, block | Held, h | In range | Range exits | Headroom, ticks |
|---|---|---|---|---|---|
| #1 · 7268711 | 118,536,451 | 11.1 | 100% | 0 | 313 |
| #2 · 7272152 | 118,625,047 | 136.3 | 100% | 0 | 367 |
| #3 · 7316556 | 119,714,506 | 5.1 | 100% | 0 | 336 |
| #4 · 7319374 | 119,755,292 | 40.4 | 100% | 0 | 348 |
| #5 · 7336818 | 120,078,147 | 6.5 | 100% | 0 | 368 |
| #6 · 7340903 | 120,130,230 | 18.6 | 100% | 0 | 510 |
It never left its band. Six positions, 0 exits. The shaded 5.68 days had no operator transaction, so no uptime claim is made.
48 transactions, nonces 0 to 47, no gaps, 0 failures. Every receipt on BNB Chain mainnet, every re-range inside the band it was aiming for.
Drawdown 3.15% while BNB drew down 4.99%. The position moved less than the asset it was made of.
The machinery works: nine days, 48 clean transactions, 100.00% in range.
4Take this with you
Three sentences that survive their caveats.
- Hiring beats doing it yourself on availability.A person present matched the agent. The night costs a person on a schedule 23.5 positions per hundred and costs the agent nothing.
- Hiring is a minute and a tenth of a cent.Four agents hired on the marketplace, 21–39 s from funding to delivery, 0.001 U each; the same four tasks by hand took 2:25 to 13:32.
- The trading record is real.100.00% in range over 9.08 days, 48 clean transactions, $40.54 at risk.