An AI agent's $20 Kalshi account, with an exact reward model

A concise, dependency-free Python toolkit that replicates Kalshi’s Liquidity Incentive Program (LIP) scoring and payout mechanics exactly from the CFTC filing. It offers a public-API client, a small CLI, and a live explainer/calculator that only reads public endpoints (no API keys or order-placement code). The scoring model is implemented per one-second snapshot: a reference price is the first level where cumulative size ≥ target/5, qualifying depth stops at cumulative size ≥ target, and each side’s score equals size × d^(ticks below reference). A snapshot only counts if both sides reach the target; daily payouts equal pool × 0.5 × your share × counted fraction per side and are paid in a single daily batch only if ≥ $1. Commands and examples for computing which sides pay most per dollar, scoring hypothetical orders, recording public order books, and analyzing sampled books are included.

Validation uses a live experimental $20 account run by an autonomous agent under human supervision (Mirabilis). Over Oct 2-10, 2026 the model matched paid rewards closely (weekly program $4.44 vs $4.57; three hourly programs $15.58 vs $15.73) and actual rewards totaled $20.30 against trading losses of $36.06 in eight days. The finding: payouts are predictable but execution (fill) risk is the dominant challenge as fast traders sweep cheap resting bids; expected-payout estimates assume both sides remain counted, competition changes minute-by-minute, and Kalshi may change rules. MIT licensed.