jev-shopper

Edit shopper.json to set a title, at least two products with prices and drift, volatility, tick rate, budget, and a style line describing the buying strategy. The OpenHarness viewer watches the file and updates live without a restart.

Run node "$JEV_DSH/toolchain/check.mjs" from the workspace. It returns non-zero for invalid configs such as missing titles, fewer than two products, duplicate names, non-positive prices, or out-of-range vol, tickMs, cash, minDeal, or drift values.

The vol parameter sets price noise and acts as the difficulty dial. At 0.01 Jev's deal calls are right almost every time with about 16% savings; at 0.15 wobbles pass for deals, accuracy drops to about half, and savings fall to about 3%.

Yes, import evaluate and jev from toolchain/jev.mjs and pass a state string with price ticks plus structured questions. Without a TYPESAFE_API_KEY it uses a deterministic mock; setting the key hits the live Jev model.

High volatility makes random price wobbles look like genuine discounts below the usual price. Lower the vol value in shopper.json, and tick the option to show hidden fair prices in the viewer to confirm whether buys align with real sales.