meridian — RWA market-making agent

Mid (USD)

—

Effective spread (bps)

—

Inventory

—

Realised PnL

—

Inventory PnL

—

Fill rate

—

Adverse selection

—

Sharpe-ish

—

Mid + quotes

PnL trajectory

Order ladder (live)

Recent fills

tick	side	px	qty	edge (bps)

Strategies in this repo

constant_spread live

Baseline. Symmetric quotes around the oracle mid at a fixed bid/ask offset.

filesrc/strategies/constant_spread.py handlesvol regimes? no handlesinventory skew? no good forliquid pairs, calm regimes

adaptive_spread live

Widens with realised vol; skews bid/ask to unwind inventory. Avellaneda-Stoikov inspired.

filesrc/strategies/adaptive_spread.py handlesvol regimes? yes handlesinventory skew? yes good forRWA, long-tail, thin books

compliance_gated roadmap

Per-counterparty quoting based on jurisdiction + ONCHAINID claim. Required for ERC-3643 RWA.

trackedsee issue #11 (RWA tokenomics) depends onkcolbchain/rwa-toolkit

geo_priced roadmap

Different mid per jurisdiction (US vs EU vs APAC) — same asset, different regulated price.

trackedsee issue #5 (jurisdiction transfer module)

Simulation model (open)

A simplified single-asset book runs in the browser. Each tick the mid takes a Brownian step (vol slider). The agent posts a bid + ask using the selected strategy. Each side has a per-tick fill probability; the toxic-flow share assumes the post-trade mid moves against the maker that fraction of the time, modelling adverse selection. Inventory PnL marks the current position to mid.

This is a teaching toy — the production agent in this repo runs a richer book replay against historical RWA prints. See src/backtest/engine.py.