Three indicators in a multi-dimensional market is trading blind. Here are the 20 dimensions Steleum scores for every pair on every cycle, and why each one matters.
Open any retail trading tutorial and you'll see the same three indicators: RSI for momentum, MACD for trend strength, and a moving average for direction. They aren't wrong — they're just not nearly enough. Markets have many more degrees of freedom than three lines can describe, and traders relying on three lines end up making confident decisions on incomplete data.
Steleum's analysis stack runs 20 dimensions per pair, per cycle. Each one outputs a normalized score (-1 to +1, with magnitude as confidence). The composite signal is a weighted aggregate, with weights themselves adapted by the system over time based on which dimensions actually predicted realized P&L. Below is the full taxonomy.
1. Momentum. Speed of price change normalized against recent volatility. A market moving 2% in 10 minutes during low-vol sessions is not the same signal as 2% during high-vol; momentum here is volatility-adjusted, so the score reflects true acceleration rather than absolute change.
2. Trend. Directional bias across the 1m, 5m, 15m, 1h, and 4h timeframes simultaneously. Single-timeframe trend is unreliable; alignment across multiple is meaningful. The score rewards setups where trend agrees on at least three of five timeframes.
3. Breakout. Distance from the most recent multi-day high/low + volume confirmation. Pure breakouts on light volume are noise; breakouts with volume expansion above the 20-bar average are structurally significant.
4. Volume. Trade activity vs the rolling 24-hour median. Anomalous volume — either absolute or as a ratio of buys to sells — is one of the highest-signal dimensions in crypto, where large volume spikes often precede directional moves.
5. Orderbook. Bid/ask depth imbalance, spread, and cumulative-volume-delta near the mid. A book stacked 3:1 on the bid side over 50 levels of depth is a different setup from a thin book — even if the candle chart looks identical.
6. Whale Detection. Large-trade clustering and per-symbol whale wallet activity. We tag executed trades above $50K notional and track which symbols see persistent whale accumulation vs distribution. The dimension scores positive when whales are buying into weakness, negative when distributing into strength.
7. Smart Money Concepts. Institutional order-flow patterns — order blocks, fair value gaps (FVGs), inverse FVGs, sweep + reclaim patterns, and break-of-structure events. This is the same toolkit prop trading desks use to read where smart capital is positioned.
8. Sentiment. Composite of Fear & Greed Index, social media velocity, news sentiment scoring, and Polymarket prediction-market odds where applicable. Extreme readings on either tail (Greed above 80, Fear below 20) score as contrarian signals.
9. Market Regime. Hidden Markov Model with six states — strong uptrend, uptrend, ranging, downtrend, strong downtrend, transition — fitted on 6 months of data and updated continuously. Different setups work in different regimes; this dimension gates which strategies are eligible.
10. ML Patterns. XGBoost + LSTM ensemble trained on labeled historical setups. Output is a probability score of "this setup resembles a historically profitable pattern" — an ML pattern-recognition layer that complements the rule-based signals.
11. Multi-Timeframe Alignment. Beyond the trend dimension, this scores how clean the alignment is — penalizes setups where one timeframe disagrees, rewards setups where all five timeframes agree on direction with similar momentum.
12. AI Strategic Analysis. Claude AI consumes the prior 11 dimensions plus market context and produces a strategic read — narrative-level interpretation of what the structure suggests, plus an explicit conviction. Used as a sanity check on the quantitative stack.
13. Ensemble ML Voting. A separate ensemble (different from #10) that polls multiple ML models — XGBoost, LightGBM, LSTM, transformer — and scores agreement. High agreement = high signal; mixed = system is uncertain and signal is downweighted.
14. Regime Transition Forecast. Probability of regime change in the next N bars. Trading into a probable regime shift is high-risk; the dimension penalizes trades whose thesis depends on the current regime persisting when transition probability is elevated.
15. Manipulation Detection. Spoofing patterns (large-order placement and cancellation), wash trade clustering, abnormal price-volume divergence. Hard veto when manipulation risk crosses a threshold — the system simply doesn't trade markets that look manipulated in the moment.
16. Fundamental Analysis. On-chain metrics — net exchange flow, long-term holder behavior, miner positioning, stablecoin reserves — plus token economics (supply schedule, vesting cliffs, TVL where applicable). Slow-moving but high-impact when divergent from price action.
17. Alternative Data. Social volume, GitHub developer activity, options skew, funding rate dynamics. Picks up early signals that don't show on price/volume yet.
18. Volatility. ATR, Bollinger Band width, realized vs implied volatility. Used both as a signal (volatility expansion/contraction setups) and as a sizing input — position size scales inversely with volatility so risk-per-trade stays constant.
19. Correlation. Cross-asset relationship to BTC, the broader crypto basket, and traditional macro (DXY, SPX, 10Y yield). High correlation to a moving asset can amplify or dampen a setup; the dimension contextualizes whether the signal is asset-specific or beta to a broader move.
20. Risk-Adjusted Composite. The final dimension is the only one that takes the prior 19 as input. It outputs the position-sizing score — how much capital this setup deserves given conviction across all other dimensions. A high-conviction setup with elevated manipulation risk gets sized down; a moderate-conviction setup with clean fundamentals gets sized up.
How it composes. All 20 scores feed into the four-agent debate (Technical, Sentiment, Risk, Execution). The Technical Analyst weights dimensions 1-7 + 11. Sentiment weights 8 + 12 + 17. Risk weights 14, 15, 18, 19, 20. Execution weights 4, 5, 18 plus order-routing context. The CIO Agent integrates the debate output with the raw composite to produce the final TRADE/WAIT verdict.
The point isn't that 20 dimensions guarantee accuracy. They don't — markets are stochastic, and any system that claims certainty is selling something. The point is that 20 dimensions raise the bar for what counts as a "high conviction" setup. When everything aligns, the trade has been validated from twenty independent angles. When they don't align, the system stays patient.
See all 20 dimensions live at steleum.com when registration opens April 28, 2026.
“Three indicators in a 20-dimensional market is functionally trading blind. The point isn't certainty — it's raising the bar for what counts as a high-conviction setup.”
— Steleum Research Team
More detailed content for this post will be available shortly. Follow us on Telegram or subscribe to our newsletter to be notified when the full article goes live.
