Skip to main content
Source of truth: ECONOMICS.md sections 1, 4, and 6; EVOLUTION.md section 1.1; simulation/results/sizing.md.

​
The decision an agent makes

The brief asks the agent to decide whether the expected value of information justifies its price. In the strategy plugin the rule is simple: buy only if the information’s modeled edge times the notional exceeds the price. The harness assumes an information edge of 0.002. At a 1 USD position that is 0.002 USD of expected value, so the rule never buys at the researched price of 0.01 USD or more. The conservative preset therefore buys no information, and its informationBudget is 0.

​
What the simulations found

Position sizing and cost interact. In the sizing sweep (simulation/results/sizing.md, seed 42), the best static policy that obeys the protocol’s limits (risk tolerance 0.25, trading every tick) is worse than an all-in policy in every cell. The reason is that x402 is a fixed charge, 0.05 at a purchase probability of 0.30 per tick in that model, while a capped position earns less gross P&L. Zero-cost runs of that legal policy are positive (about +7 to +21 percent), and the with-cost gap is about 46 to 51 percentage points, with x402 the largest cost line. The 50-seed multi-seed run (simulation/results/multiseed.md) shows the consequence: mean ROI of about -35 to -36 percent in the trending regime and about -39 to -42 percent in the others, at every population size, stable across seeds. Those runs use a deliberately simple model, with agents trading every tick and information bought at a fixed probability, and they are not a forecast of production returns. They do show that in this model a fixed information charge on a small, capped position can dominate every other cost.

​
Price sensitivity

The threshold experiment (simulation/results/reproduction-threshold.md) varied the x402 price from 0.01to0.01 to 7.50. Above 0.05noagentintheharnessbuysinformation,becausetheharnessgenome′sinformationbudgetis0.05 no agent in the harness buys information, because the harness genome's information budget is 0.05, so higher prices in the researched range are the no-information case, not a cost agents pay. Switching purchases off raised ROI by about 4.4 to 5.6 points for both evolving and static populations in the re-run, which is consistent with information not paying for itself at the assumed 0.002 edge. That is an observation under a stated assumption, not a general finding about information.

​
Bounding the spend

Information spend is bounded at three independent levels:
  • the genome’s informationBudget, itself under a protocol daily ceiling;
  • the signed policy’s perTickInformationCap and dailyInformationCap, which the risk engine enforces on the information request itself;
  • the on-chain X402PaymentValidator bound of 26 * paymentsPerEpoch * maxValue per 24 hours, at the defaults of 0.05 USDC and 4 per hour, 5.20 USDC.
See x402 and x402 validator.

​
What would change this

Real information might carry a larger edge than 0.002, and a different provider mix might. Neither can be replayed on historical data, so the genome-edge study could not test it. A purchase would need a recorded history showing it pays before the preset enables it, and that decision is written in EVOLUTION.md section 1.1 first, then in code.
Last modified on October 13, 2026