Ian Langworth ☠ statico

Reverse-Engineering the Quicken for Mac (2007-era) QDF Bundle Format

Background

A Quicken for Mac data file (circa 2007-2008, Quicken Mac 2007) was provided as a macOS bundle directory with no working copy of Quicken available to export from. The goal was to extract transaction data (dates, payees, check numbers, memos, and amounts) into a CSV.

Bundle Structure

The Gowanus Canal: A 1.8-Mile Monument to Filth

Generated with Claude Code + Opus 4.6

The "Black Mayonnaise"

The star of the show is the canal's infamous bottom sediment, affectionately known as "black mayonnaise" — a thick, oily sludge that is in some places up to 20 feet deep. It's a lethal cocktail of:

Coal tar from old manufactured gas plants
Heavy metals: mercury, lead, copper, arsenic, chromium

Playwright MCP in Docker (ARM64)

Problem

The Playwright MCP server (@playwright/mcp) defaults to Chrome (channel: "chrome"), expecting it at /opt/google/chrome/chrome. Chrome has no ARM64 Linux binaries, so it fails immediately. The browser_install MCP tool also fails on ARM64.

Claude Code ignores args and env fields in ~/.claude/mcp.json for the playwright MCP server — it always launches npm exec @playwright/mcp@latest with no arguments and no custom env vars. So --browser firefox, env vars like PLAYWRIGHT_MCP_BROWSER, and custom command wrapper scripts all have no effect.

Fix

How Modern CAPTCHAs Actually Work: A Deep Dive

The Big Picture

Your intuition is correct: all three systems fundamentally collect a mass of signals, send them to black-box server-side ML models, and get back a probabilistic score. None of them have a clean deterministic "if X then bot" rule. It's all heuristics and probability. But the depth of what they collect is staggering.

1. Google reCAPTCHA: The VM

Here's the current state of the art:

Token Efficiency: JSON Alternatives for LLMs

The Rankings (tokens for same data)

Format	Token Reduction vs JSON	LLM Accuracy
TOON	~40% fewer	73.9% (best for tabular)
Markdown	34-38% fewer	Good (cost-optimized)

Here's a comprehensive summary of tips for getting Nano Banana (Gemini) to generate images better suited for small use cases:

Pixel Art Tips

Use specific dimension keywords:

"32×32 pixel art icon of [subject]"
"16-bit pixelated [subject]"
"8-bit pixel art in the style of [retro game]"
"Low-resolution pixel art" or "chunky pixels"

	<!DOCTYPE html>
	<html lang="en">
	<head>
	<meta charset="UTF-8">
	<meta name="viewport" content="width=device-width, initial-scale=1.0">
	<title>Math Tutor</title>
	<style>
	* { box-sizing: border-box; margin: 0; padding: 0; }
	body {
	font-family: 'Segoe UI', system-ui, sans-serif;

	#!/usr/bin/env bun

	// ── Star Field ──────────────────────────────────────────────────────
	// Parallax star field that scrolls right-to-left in the terminal.
	// 3 depth layers, 256-color, differential rendering (only redraws
	// changed cells). Supports ASCII and Unicode glyph modes, adjustable
	// star density, scroll speed, and color saturation — all via hotkeys.
	// Made with Claude Code.
	//
	// Usage: bun starfield.ts [--ascii\|--unicode] [--stars N] [--speed N] [--sat N]

	<!DOCTYPE html>
	<html lang="en">
	<head>
	<meta charset="UTF-8">
	<meta name="viewport" content="width=device-width, initial-scale=1.0">
	<title>MCU Ratings</title>
	<script src="https://cdn.jsdelivr.net/npm/chart.js@4.4.0/dist/chart.umd.min.js"></script>
	<style>
	body { margin: 20px; font-family: system-ui, sans-serif; background: #030712; color: #fff; }
	h1 { text-align: center; margin-bottom: 20px; }

	function status() {
	result=$(/usr/local/bin/hidapitester --vidpid 046D/C900 --open --length 20 \
	--send-output 0x11,0xff,0x04,0x01 --read-input --timeout 500 2>/dev/null)
	if echo "$result" \| grep -q "11 FF 04 01 01"; then
	echo "on"
	else
	echo "off"
	fi
	}