Deterministic Negotiation Kernel

A deterministic negotiation kernel is a mechanism that makes a simulated trading partner respond only according to preset rules, not luck or randomness, when an AI haggles over prices with it.

Think of merchants haggling in a market. Some quote prices however they feel that day, but others fix a bottom line in their mind and stick to it without ever budging. A deterministic negotiation kernel is the latter type. If an AI agent offers to buy 1,000 bottles at a cheap price, the program playing the supplier role never drops below a preset minimum unit price, and the back-and-forth haggling continues over multiple rounds on that basis.

The reason for fixing the counterpart's behavior this way is fair comparison. If the negotiating partner responded randomly each time, some AIs might get lucky and buy cheap while others get unlucky and pay more. Setting fixed rules ensures that only the difference in each AI's actual negotiation skill shows through.

The article states: "A deterministic negotiation kernel is attached here — for example, if the agent requests a unit price as low as 1.5 for a bulk order of 1,000 bottles, the supplier counters that it can't go below 1.8, and this multi-round haggling repeats." A common misunderstanding: this isn't a feature for negotiating with real people, but a preset ruleset for a simulated counterpart, designed to fairly measure an AI's negotiation skill within a benchmark.