HENRY BOY ETH CANDLE TRADER · PAPER GAME

How Henry thinks (plain English)

Every time Henry forecasts, he looks at the last closed candles and combines a whole panel of classic technical readings into one projected move. He forecasts walk-forward: he predicts candle #1, feeds its OHLC back into every indicator, then predicts #2 from that updated picture — repeating through #10, so each ghost candle's own metrics genuinely shape the next. Each reading gets its own tunable weight — they all live in one WEIGHTS object in the page code, and Henry nudges them himself based on which readings actually called it right:

• Momentum — the average move of the last 5 candles. Drifting up? Henry leans up.
• Mean reversion — the pull back toward the 20-candle average (EMA20). Stretched far from it? Expect a tug back.
• MACD (12/26/9) — the histogram: is trend momentum accelerating or fading?
• Bollinger %B (20, 2σ) — price pressed against a band tends to snap back toward the middle.
• Stochastic %K (14) — overbought/oversold on a 0–100 scale; extremes get faded.
• CCI (20) — unusually hot or cold readings tend to cool off.
• Williams %R (14) — another overbought/oversold lens; deep negative readings are washed-out.
• ROC (3/5/10) — the speed of the move at three lookbacks; fast moves carry further.
• OBV flow — is volume actually flowing with the price move, or is it hollow?
• VWAP deviation (20) — distance from the volume-weighted fair price; gravity applies.
• Range position — where price sits inside the last 20 candles' high/low range; tops and bottoms of the range invite reversals.
• Live sentiment (composite) — Henry's read of the crowd, refreshed every minute from three ingredients: the daily Fear & Greed mood (slow regime backdrop, read contrarian — extreme fear tilts bullish, extreme greed tilts bearish), real-time taker flow from the last 500 Binance trades (are traders lifting offers or hitting bids right now), and the Level 2 order-book imbalance. Each is normalized to −1…+1 and blended 30/35/35; if any feed is down the blend reweights among the live ones, and if all are down the factor is skipped entirely.
• Level 2 walls & spread — the biggest visible buy/sell walls within 0.3% of mid-price, acting as support and resistance the whole market can see, plus the spread in basis points (tight = healthy). The raw bid/ask imbalance itself feeds the live sentiment composite instead. Skipped entirely if the depth feed is down.
• BTC lead-lag — Bitcoin usually moves first and ETH follows within minutes, so Henry watches BTC's 15-minute momentum on the same Binance feed. When BTC breaks, ETH's forecast leans the same way. Skipped entirely if the BTC feed is down.
• Regime filter — before forecasting, Henry reads the tape: the efficiency ratio over the last 20 candles tells him whether the market is trending (straight-line moves) or choppy (back-and-forth chop). In a trend he trusts the momentum family (momentum, MACD, ROC, OBV, BTC lead, crowd) more and damps the fade-the-extreme readings; in chop he does the reverse. The badge above the chart shows which regime he's in. Wrong model for the day is how most forecasts die — this is Henry's guard against that.
• Polymarket crowd odds — real money betting on where ETH's price lands: Henry pulls the nearest open "Ethereum price on …?" market, turns its price bands into the crowd-implied chance ETH ends above the current price, and tilts with it. It's the market forecasting itself. Skipped entirely if the feed is unreachable or no live market exists.

How Henry emulates the machines 🤖 — most crypto volume is algorithms, so Henry copies the public, documented parts of their playbook:

• Microprice — the volume-weighted fair price at the very top of the book. When the last trade sits below it, pressure points up. This is the single most replicated short-horizon finding in market-microstructure research.
• Tiered imbalance — levels 1–3 are urgent money at the touch; levels 4–10 are patient money deeper. When they disagree, Henry notes the split (urgent vs patient).
• Absorption (VPIN-lite) — wildly one-sided trade flow with a flat price means somebody big is absorbing it. Pressure is building for a pop with the flow.
• Book fragility — thin book plus wide spread equals fragile tape: Henry expects bigger moves and sizes his steps up, without picking a direction from it.
• Cross-venue — the same ETH trades on Binance, Bybit and OKX, and the machines arbitrage across all three, so dislocations show up cross-venue first. When one venue's mid drifts from the others, the laggard tends to catch up; Henry tilts that way and notes which venue is leading.

What's not emulated: the proprietary part — microsecond latency, the full depth feed, colocated servers, and the exact models of the top HFT firms, which are trade secrets worth millions. Henry copies the textbook; the firms keep the answers.

• RSI brake — when RSI(14) screams overbought (>70) or oversold (<30), Henry dampens the projected move instead of chasing it.
• Volume conviction — a move on heavy relative volume gets projected bigger; thin-volume drift gets shrunk.
• Volatility regime — when the ATR regime runs hot, Henry scales his steps with it instead of pretending it's a calm tape.
• Decay — the further out the candle, the less Henry trusts himself: each step's move shrinks, and the confidence label drops a few points per candle.

How he learns: Henry always keeps 10 predictions ahead of the latest real candle. When a new real candle closes, it takes the oldest prediction's place on the chart — Henry rates it ✓/✗ against reality, checks which of his readings called it right, and nudges their weights (small steps, clamped to sane bounds). A fresh 10th prediction is appended at the far end, so the window slides forward forever. It's a toy adaptive loop — not machine learning, just an honest scoreboard with consequences.