MANGO: Build an AI Empire by Scraping Everyone Else's Data

The pitch

You are the founder of a new AI company. You have no data, no models, no talent, and $10,000. What you do have is a complete absence of ethics and a browser. Welcome to MANGO — an idle clicker x strategy simulation where the business model is scraping, the growth model is distillation, and the legal model is "settle out of court and keep going."

The game started as a joke. We were sitting around talking about how every AI company in 2026 has the exact same playbook: scrape the internet, train on data you don't own, distill your competitors' models, and when someone sues you, settle for a fraction of what you made. It's not a bug in the industry — it's the feature. So we thought: what if we turned that playbook into a game? What if the most efficient strategy in the game is also the most ethically horrifying one? What if winning means becoming the thing you hate?

Why we built this

The AI industry in 2026 operates on a simple, unspoken contract: take first, ask forgiveness later. OpenAI scraped the internet. Google scraped the internet. Anthropic scraped the internet. Meta scraped the internet. Every foundation model company trained on copyrighted content — books, articles, code, academic papers, video transcripts, forum discussions — without asking, without paying, and without attributing. When caught, the response is always the same: "fair use," "transformative," "we'll figure out licensing eventually."

MANGO makes that contract explicit. The game's core loop — scrape data, train models, distill rivals, repeat — is literally what these companies do. The only difference is that in the game, you can see the numbers going up, which makes it fun instead of infuriating. The satire is not subtle. Your crawlers are called "Web Scraper," "Forum Bot," "Social Harvester," "Code Miner," "Paper Scraper," "Video Extractor," "Archive Raider," and "Dark Web Crawler." Your models are named by parameter count. Your prestige mechanic is called "Model Collapse." Your random events include "Copyright Lawsuit" (you lose 5% of your data tokens) and "Leaked Weights Found" (you gain compute power — from someone else's work).

We built MANGO because satire is most effective when it's interactive. You can read a think piece about data scraping and feel mildly concerned. Or you can click a glowing orb 500 times, watch your data tokens climb, buy a Dark Web Crawler, and realize you're having fun doing exactly what Sam Altman does every day. The discomfort is the point.

The game theory

MANGO is built on three layers of game theory that map directly to real-world AI industry dynamics:

Layer 1: The idle clicker (exponential growth)

The clicker layer is the hook. You click the MANGO core, you get Data Tokens (DT). You buy crawlers that generate DT automatically. Each crawler type is 10x more expensive and 8-10x more productive than the last, creating the classic exponential growth curve that makes idle games compulsively playable. The math is simple but the dopamine is real.

The clicker layer maps to the data acquisition phase of a real AI company. Your first Web Scraper is a single intern running wget. Your Dark Web Crawler is a distributed botnet harvesting cached datasets from hidden forums. The escalation is the same — it's just that in real life, nobody tells you when you've crossed a line.

Layer 2: The strategy sim (resource management)

The strategy layer is where MANGO becomes a game rather than a toy. You're managing:

The strategy layer forces trade-offs. Do you invest in more crawlers (more data) or more servers (more compute)? Do you hire cheap interns or save up for a Chief Scientist? Do you lease an office (cheap, small) or buy one (expensive, more room)? The day/night cycle means staff only work at full capacity during business hours — at night, revenue drops 50%. Power management adds another constraint: if your servers draw more power than your generators produce, compute output drops to 10%.

Layer 3: The prestige loop (model collapse)

In real AI research, "model collapse" is a phenomenon where models trained on AI-generated data degrade over generations — the output becomes increasingly generic, then increasingly broken. It's a serious concern for the industry. In MANGO, it's your prestige mechanic. When you trigger a Model Collapse, you lose everything — all crawlers, models, staff, office — but gain Synthetic Weights (SW), a permanent multiplier to all production. The more total DT you've accumulated, the more SW you get. Then you start over, stronger than before.

This is, again, exactly what real AI companies do. They train a model, it gets stale, they scrape more data, retrain from scratch, and the new model is better because of what they learned the first time. The "collapse" is just a product cycle. The SW is institutional knowledge. The fact that it's called "Synthetic Weights" is the punchline.

Features

The distillation mechanic

The most pointed satire in MANGO is the distillation system. In the real AI industry, "distillation" is the practice of training a smaller, cheaper model on the outputs of a larger, more expensive model made by a competitor. It's technically legal, vaguely unethical, and universally practiced. OpenAI accuses DeepSeek of distilling GPT-4. Google accuses everyone of distilling Gemini. Nobody admits to doing it themselves.

In MANGO, distillation is a research tree. You spend Compute Power to "distill" a rival model, gaining a permanent bonus:

The names are thinly veiled. The bonuses stack. The achievement for completing all five is called "Monopoly." We considered calling it "Fair Use" but decided that was too on-the-nose.

What's next

MANGO is live and playable right now, but it's far from done. Here's the roadmap: