YK yongkangc

Market Maker Spread Tightening

flowchart LR
    subgraph before["Before MM Enters"]
        direction TB
        B1["Best Bid: $0.45"]
        A1["Best Ask: $0.70"]
        S1["Spread: $0.25 (25%)"]
 end

Parameter Sweep Report - 2 Hour Backtest

Date: 2026-01-30
Time Window: Last 2 hours
Total Runs: 100

Key Findings

All Runs Unprofitable

P&L Range: -$112.63 to -$82.21

Reth Sparse Trie Guide

A comprehensive guide to understanding Ethereum's Merkle Patricia Trie as implemented in reth.

Paper Trading Fill Logic - pm-mm-rs

Overview

The paper trading system simulates realistic order fills using PaperVenue which wraps MockVenue. Fills are triggered when real-time Polymarket book updates cross resting order prices.

Two Fill Types

1. Aggressive Fills (Crossing the Spread)

ParallelSparseTrie::prune() Parallelism Analysis

Summary

Benchmarking revealed that rayon parallelism in ParallelSparseTrie::prune() causes significant overhead without any performance benefit. The parallel path is always slower than serial, even at the maximum of 256 lower subtries.

Benchmark Results

Serial vs Parallel (forced paths)

Polymarket 15-Minute Binary Market Making Strategy

The Market

Polymarket offers 15-minute binary prediction markets:

"Will BTC be higher than $100,000 at 3:15 PM?"

YES token: Pays $1 if BTC > strike at expiry, else $0
NO token: Pays $1 if BTC ≤ strike at expiry, else $0
YES + NO always = $1 (complements)

RocksDB Multi-Database Consistency Architecture Analysis

Problem Statement

When reth uses multiple storage backends (RocksDB for history indices + MDBX for changesets), a commit order race condition causes execution divergence during block validation.

The bug manifests as "nonce 917 too high, expected 0" because:

RocksDB history indices commit first, saying "account X was modified at block N"
MDBX changesets haven't committed yet
Read path follows RocksDB's pointer to a non-existent changeset

RocksDB Benchmark Methodology & Bug Analysis

Executive Summary

Bug Found: RocksDB pending batch writes are not visible to concurrent history index reads, causing EVM execution divergence during engine_newPayload processing.

	[package]
	name = "rocksdb-blocksize-bench"
	version = "0.1.0"
	edition = "2021"

	[dependencies]
	rocksdb = "0.22"
	clap = { version = "4", features = ["derive"] }
	hdrhistogram = "7"
	serde = { version = "1", features = ["derive"] }

	[2m2026-01-20T13:13:15.734451Z[0m [32m INFO[0m Running benchmark using data from RPC URL: https://eth.drpc.org
	[2m2026-01-20T13:13:15.772700Z[0m [32m INFO[0m Connecting to Engine RPC at http://localhost:8551/ for replay
	[2m2026-01-20T13:13:16.002256Z[0m [32m INFO[0m [3mcombined_result[0m[2m=[0mBlock 24262068 processed at 0.9734 Ggas/s, used 20307563 total gas. Combined: 0.9143 Ggas/s. fcu: 1.348709ms, newPayload: 20.862554ms
	[2m2026-01-20T13:13:16.112979Z[0m [32m INFO[0m [3mcombined_result[0m[2m=[0mBlock 24262069 processed at 1.0036 Ggas/s, used 19394546 total gas. Combined: 0.9371 Ggas/s. fcu: 1.37066ms, newPayload: 19.32462ms
	[2m2026-01-20T13:13:16.210811Z[0m [32m INFO[0m [3mcombined_result[0m[2m=[0mBlock 24262070 processed at 0.9430 Ggas/s, used 20225346 total gas. Combined: 0.8675 Ggas/s. fcu: 1.8666ms, newPayload: 21.448162ms
	[2m2026-01-20T13:13:16.333041Z[0m [32m INFO[0m [3mcombined_result[0m[2m=[0mBlock 24262071 processed at 1.6818 Ggas/s, used 49123279 total gas