IHunting
I started by testing strategies other traders sold online, with their screenshots and win rates. I replayed them on real prices, with fees.
- 70+ strategies tested by November
- almost all killed
Most edges are screenshots.
a builder who tests things until they break.
building at 42 ParisI build things, then try to prove them wrong. Most of them die. The ones that survive get real money, real users and a public record. I write the specs and check the numbers. AI agents write a lot of the code.





I started by testing strategies other traders sold online, with their screenshots and win rates. I replayed them on real prices, with fees.
Most edges are screenshots.
Arceus started as a portal and a research lab. Every idea has to beat fees, random entries and data it has never seen before it gets a cent.
If a number can be typed by hand, someone will type it.
Two systems outside trading: a router for crypto options, and a way to pay in digital dollars with no internet.
A demo is a promise. A mainnet transaction is a fact.
The one strategy that survived the lab went live with real money, while I started the Common Core at 42 Paris.
Paper is free. Real money is the test.
Arceus as a portfolio: trading first, then e-commerce and a phone farm for Europe. Each one measured like a trade.
Measure first. Scale second.
A calm sea never made a skilled sailor.an old proverb, written on a wall I saved
LIVEA bot that fades the hype on small crypto perpetuals. A month of paper trading, then real money. Its public page reads every trade from the blockchain.
+19.5% account · 5 of 6 trades won · live since Nov 11 · too early to call it an edge
MAINNETPay in USDC with no internet. Sign in airplane mode, pass the payment by Bluetooth, QR or link. It settles the moment any phone nearby gets signal.
1.01 s from an iPhone in airplane mode to Arc mainnet · 107 commits · iOS, Android, web
TESTNETA best-price router for crypto options. One order, split across venues, without ever holding the user's money. AI agents can call it too.
20 venues · 1.2M price observations · Rust core
NEXTContent accounts run like a portfolio: one real phone and one French SIM each. The tools for this only exist in Asia, for Asian IPs.
0 phones live · autopilot written and tested in dry-run · 428 TikToks of one niche mapped
| Idea | What I tested | Result | Verdict |
|---|---|---|---|
| Darkrai | Fade the hype on small perps, plus funding | passed fees, placebo, unseen data | ● LIVE |
| paste.trade calls | 12,371 public trading calls on real prices | +18.8% → +0.23% | ✕ DEAD |
| Scalping < 1h | Every crypto setup under one hour | 2,160 configs → 0 | ✕ DEAD |
| Funding carry | Collect funding across many coins | 16 / 16 out-of-sample runs negative | ✕ DEAD |
| Pump.fun | Buy new coins at graduation, sell at 2x | −8.3% per trade | ✕ DEAD |
| Post-listing short | Short coins right after they list | −19% out of sample | ✕ DEAD |
| Copy a famous vault | Rebuilt its rules from on-chain data | alpha ≈ 0 vs holding | ✕ DEAD |
| ICT London | London-session ICT setups on futures | −0.19R out of sample | ✕ DEAD |
| Turtle rules | The original Turtle breakout rules | no better than random (p = 0.23) | ✕ DEAD |
| GEX levels | Gamma levels from a paid tool | died on the placebo | ✕ DEAD |
| Options alerts | Paid unusual-flow alerts | 0 of 112 with an edge | ✕ DEAD |
Made this site at night with AI agents. There's also a film version of it.
The Darkrai page showed paper results as if they were live. Rewired it to read the chain directly, then redesigned it four times in one night.
Cut an interview with a crypto trader for people who are tired of crypto. 11 minutes down to 8:47.
First real trade: a short. I checked the stop was on the exchange before going to sleep.
Built my own harness for Claude Code: model routing, project cards, and a decision queue that reaches me on WhatsApp.
Darkrai goes live with real money. No number on its page is typed by hand.
Four agents audited a levels strategy I liked. No lookahead, but a real edge of 0 to 0.3 points. Killed it.
Started ArcDrop. Question: can a payment survive with no signal at all?