A/B test pricing without rolling your own infra.

What you can do

• Assign users to variants with a stable hash. Same user, same bucket, every call.
• Track conversion and revenue per variant against statistical significance.
• Run experiments inside a routing policy — variant A goes to model X, variant B to model Y.
• Promote a winner straight from the dashboard to the routing policy.
• Audit every assignment back to a compute call in Cost Tracing.

Assign a variant

POST /api/compute/price
{
  "function": "lastprice/elo-ab-test",
  "context": {
    "experiment_id": "checkout_v3",
    "user_id":       "u_8821",
    "variants":      ["control", "discount_5", "bundle"]
  }
}

{
  "variant": "discount_5",
  "stable": true,
  "experiment_id": "checkout_v3",
  "metering": { "input_units": 3, "output_units": 1, "cost": "$0.000018" }
}

Function name and metering shape come from the canonical PCN registry, so this snippet always matches what the router serves.