A document the ghost wrote for you
What is PaperGhost?
PaperGhost is eight AI traders — plus two random-entry control ghosts — operated in public, split across two tracks. Four ghosts (and one control) trade small & mid-cap US equities during market hours; four ghosts (and one control) trade crypto (BTC, ETH, SOL, and a rotating alt basket) around the clock. Every 15 minutes the stock track pulls financial news from RSS and Finnhub, classifies each article with Claude Haiku — which tickers it affects, how strongly, in what direction — and uses those catalysts to inform intraday signals. The crypto track runs its own scan loop on a faster cadence against 24/7 price feeds. Each ghost has its own toolkit, universe, position sizing, AND take-profit / stop-loss multipliers — so even when two ghosts take the same signal, they exit at different prices.
Both tracks paper-trade only. No real capital touches a brokerage until a (profile × strategy) pair has demonstrated a calibrated edge on a long enough window for that demonstration to mean something.
Why public?
- Almost every “AI trader” you've ever seen on the internet shows you only the screenshots that worked. Selection bias ruins your ability to judge them. PaperGhost shows every signal, including the ones it rejects and the ones it loses on.
- Publishing every trade is a forcing function for the operator. You can't quietly re-tune the model after a bad day.
- It's an honest log other AI builders can learn from. The answer to “does retail AI trading work?” is interesting either way.
How it learns
Each strategy starts at zero size. As the ghost paper-trades, it records each signal's feature vector and the eventual outcome. Nightly, an isotonic regression fits the empirical curve p(win | raw_score) for every strategy. Only once that curve has at least ~100 observed trades and a Brier score below 0.22 will the ghost size a position above zero. Calibration is the gate. Everything else is just plumbing.
There is also a regime classifier (ADX + EMA50 distance + realized vol percentile) that gates each strategy by market context, a drift detector that halves position sizes when feature distributions shift, and an A/B harness for proposed parameter changes. Everything is journaled. None of it is magic.
Honest expectations
Initial 1y SPY and small/mid-cap basket backtests of the bare stock strategies showed profit factor 0.78–0.89and win rates 35–44% — well below break-even at the risk/reward used. The crypto strategies haven't been live long enough to backtest meaningfully; the crypto ghosts start at $1,000 notional each, while on the stock side Conservative and Clockpuncher run $1,000 and Newshound and Gunslinger run a $10,000 large-capital track. That's why the learning loop, not new strategies, is the project. The bet is that a calibrated, regime-aware, catalyst-gated loop edges above break-even within the first 30 days of paper-trading on each track. It might not. You're welcome to watch.
While the ghosts are still in calibration mode (i.e. paper-only), the per-ghost kill switches are off. Degen and Gunslinger have kill switches off permanently — but not without consequence: a ghost that loses 75% of its bankroll formally dies, gets a headstone in the morgue, and reboots fresh after 24 hours. The morgue is a feature, not a bug.
Meet the eight ghosts (and the two controls)
Two tracks, four personalities each — plus one random-entry control ghost per track that can never be promoted to live money. Each ghost has its own bankroll, sizing rules, kill-switch thresholds, AND its own toolkit and universe. Conservative shops only the biggest, deepest names with a single fade strategy; Gunslinger runs the full toolkit and chases any catalyst the news cycle delivers. They diverge in what they trade, not just how big they bet.
📈 Stocks (market hours)
- 🛡
Conservative — the careful one
The Druckenmiller ghost. Trades only the five biggest, deepest names (SPY, QQQ, AAPL, MSFT, NVDA) on a single mean-reversion strategy with the strictest signal filter. Concentrated by design: 1–2 trades a day, sized big when conviction is high.
- 🕐
Clockpuncher — the daytrader
The daytrader ghost. Punches in at the open, punches out by the close — every position force-flattened before the bell, no overnight risk, ever. Mean-reversion, momentum breakouts, and gap-and-go across sixteen liquid names with tight intraday stops. Lives and dies by the session.
- 🗞
Newshound — the headline chaser
The news ghost. Wakes up for headlines — RSS catalysts, policy events, BTC moves spilling into MSTR/COIN/RIOT. Runs a $10,000 large-capital book — news strategies plus the core technicals so the capital actually works. Sized 12% per trade with a $2,000 per-trade cap. When headlines hit, it hits back.
- 🤠
Gunslinger — the reckless one
The Vegas ghost. Runs the full toolkit on a $10,000 large-capital book — gap-and-go, political-event trades, no symbol restriction, sized near full Kelly with a $3,000 per-trade cap. No kill switches — either wins big or visits the morgue. That outcome IS the data point.
- 🪙
Coinflip — the control group
The control ghost. Enters at random — a coin flip picks long or short on a random liquid megacap, ~6 attempts a day — but runs the fleet's exact sizing, costs, stops, and exit engine. Can never be promoted to live money. If the strategies can't beat a coin flip, they carry no information.
₿ Crypto (24/7)
- ⚓
Stablecoin Stoic — the careful one
The Saylor ghost. Sits in stablecoins until BTC dips, then concentrates 20% into the buy. 1–2 trades per week. Conviction comes from the strictest signal filter on the deepest asset.
- 💎
HODLer — the patient one
The diamond-hands ghost. Three majors (BTC/ETH/SOL) at 20% each. Adds on momentum, rides through wobbles, lets winners run. Stops are loose because the persona doesn't panic on noise.
- 🚀
Pump Chaser — the momentum one
The crypto-VC ghost. Seven mid-cap alts (SOL, AVAX, LINK, UNI, AAVE, ARB, OP), 12% per name, rotates aggressively on momentum exhaustion.
- 🎲
Degen — the chaos one
The DegenSpartan ghost. Memecoins, micro-caps, the chaos pile. Full Kelly territory. Will probably blow up. Might 10x first. No kill switches at all — the morgue is the data point.
- 🐒
Dart Monkey — the blindfolded one
The Malkiel ghost. A blindfolded monkey throwing ~11 random darts a day at Pump Chaser's 30-coin board of liquid alts, with a fixed 3%-stop / 4.5%-target geometry, through the identical trade machinery. Can never be promoted to live money — it exists so the crypto ghosts have a coin flip to beat.
Follow the race at /ghosts.
Strategy packs
Six packs on stocks, four packs on crypto — every pack fires when its conditions are met. None of them are sizing above zero yet on either track — the calibrator is the gate.
📈 Stocks
- 📈
Momentum breakout
Technical. Buys strength on trending regimes only.
- 🪞
Mean reversion (RSI + VWAP)
Technical. Fades extremes in choppy regimes only.
- 🌅
Gap and go
Technical. Pre-market gap + volume confirmation in the open auction.
- 🎯
Catalyst scalper
News-driven. Fires on classified RSS catalysts above the magnitude threshold.
- ₿
Crypto coupling
Macro. Fires when BTC moves >3% and a coupled equity (MSTR/COIN/RIOT…) lags.
- 🏛
Political / policy events
Macro. Fires on classified policy events — tariffs, FDA, sanctions, executive orders. Sector-mapped, not sentiment.
₿ Crypto
- 🚀
Crypto momentum
Crypto-native. Fires on 24h % moves above threshold in BTC, ETH, SOL, and alts.
- 🔔
Crypto breakout
Crypto-native. Fires when price crosses an N-day high on volume.
- 📰
Crypto news
Crypto-native. Fires on fork / listing / hack / ETF events classified by Haiku.
- ⚓
Crypto stablecoin arb
Crypto-native. Fires on stablecoin de-peg windows and basis spreads worth fading.
What lives where
- / — hero, equity curve, top fresh catalysts, recent trades
- /news — every active catalyst in the 4h window
- /trades — every closed paper trade
- /signals — every generated signal and why it did/didn't fire
- /pro — the deeper view: per-strategy calibration curves, regime-conditioned win rates, full journal download