// Deep Dives

Case Studies

Three projects built and run in production. Challenge, what was built, the numbers, and what it taught.

01

The InDecision Framework

Trading / Markets
// Challenge

After years of trading, I kept losing money not from bad analysis but from inconsistent execution. I'd have a thesis, then abandon it when price moved against me for two hours. The problem wasn't knowledge — it was that my decision framework existed only in my head. An informal process is not a process. It's a preference.

// What Was Built

A 6-factor weighted scoring model that generates independent bull and bear cases (0–100 each) for any crypto asset. The spread between them determines conviction level and whether to act. Built over 2+ years of backtesting, refined through 7+ years of live trading with real capital. The framework covers Daily Pattern (30%), Volume (25%), Timeframe Alignment (20%), Technical Confluence (15%), Market Timing (10%), and Risk Context as a gate. Not a model that outputs a prediction — a model that outputs a conviction score.

// Key Numbers
82.5%
Accuracy, 7+ years live
6
Weighted factors
≥15
High Conviction spread
4
Assets: BTC/ETH/SOL/XRP
// Lesson

The framework doesn't make you smarter. It makes you consistent. Consistency is where edge comes from. The purpose of a formal framework isn't to improve your analysis — it's to make your analysis executable by a system, including your future self at 2am with a position going against you.

02

PolyEdge — Live AI Trading Bot

AI / Trading
// Challenge

InDecision generated signals manually. Polymarket offered binary options on crypto prices with short settlement windows. The opportunity was clear: automate InDecision's scoring, route it through an AI decision layer, and execute trades algorithmically — removing human hesitation from the equation entirely. The manual signal-to-execution gap was costing edge. Every second between signal and execution is an opportunity for psychology to interfere.

// What Was Built

PolyEdge — an 8-factor scoring engine with InDecision as the backbone, running continuously on a Mac Mini. The bot monitors active Polymarket markets, scores each opportunity using PolyEdge + InDecision signals, filters by conviction threshold and fee structure (skips bets where market odds ≥85%), and places live bets autonomously. Integrated with invictus-sentinel for uptime monitoring and mission-control for P&L tracking. 1,063 automated tests cover every decision path in the system.

// Key Numbers
1,063
Automated tests
93%
Code coverage
$5
Live bet size
8
Scoring factors
// Lesson

The hardest part wasn't the model — it was building the infrastructure to trust it. 1,063 tests exist because I needed to know the system would do exactly what I told it to, every time, without my supervision. A bot placing live bets is not the same as a bot you're watching. Trust requires evidence, and evidence requires tests.

03

OpenClaw — The 54-App AI OS

Infrastructure / AI
// Challenge

I was managing too many manual loops: checking Discord, triggering content pipelines, reviewing blog drafts, monitoring trading bots, spawning coding agents. Each context switch was destroying flow. A human should not be the orchestration layer for repeatable processes. I needed a system that could run everything asynchronously and surface only the decisions that required my judgment.

// What Was Built

OpenClaw — a persistent AI agent running 24/7 on a Mac Mini. It manages conversations via Discord and Telegram, runs cron jobs, executes 40+ automated pipelines, spawns Claude Code agents for implementation work, and routes outputs back to me for async review. Systems built on top of it include blog-autopilot (YouTube → published article with zero keystrokes), content-flywheel (daily X posts), invictus-sentinel (infrastructure monitoring with AI post-mortems), and mission-control (portfolio dashboard). One always-on machine. No cloud compute required.

// Key Numbers
54
Apps across 7 categories
40+
Automated pipelines
4
launchd daemons
8
Docker services
// Lesson

The bottleneck was always me. Every pipeline I automated bought back irreplaceable hours. The goal isn't to eliminate human judgment — it's to eliminate human execution of repeatable steps, so judgment is the only thing humans do. When the system is working correctly, I am only ever making decisions. Never executing.