Drive it yourself

Your scenario. Real Aurora DSQL.

Pick how many agents, how much stock, and how big a budget. We provision exactly that on Aurora DSQL and fire a real swarm through the commit kernel. Make the agents outnumber the stock and watch OUT_OF_STOCK declines; starve the budget and watch MANDATE_REJECTED declines. The four invariants stay at zero either way.

Your scenario

Agents1,000

concurrent

Units in stock100

sharded across buckets

Mandate budget$50,000

@ $25/unit → 2,000 affordable

Presets

Live result

AURORA DSQL

// dial in a scenario and hit Run. // the kernel commits each agent through one // idempotent ACID transaction on Aurora DSQL.

// Every run provisions a fresh, isolated SKU + mandate, fires the swarm through the same kernel the production API uses, then runs a SELECT over the real rows to prove oversells / duplicate settlements / ledger drift / mandate breaches = 0. Inputs are bounded server-side so each run stays snappy. Nothing here is mocked.

ZeroRace · agent-safe checkout kernelAurora DSQL · DynamoDB · Vercel