How Autonium works

The Algo Score is a single 0–100 conviction gauge that summarizes many signals into one number. It answers one question: is this thesis strengthening, weakening, or breaking right now?

Higher scores mean more signals currently align in the same direction with enough magnitude to matter. Lower scores mean signals conflict or point down. The score is designed to be comparable across tickers within the same asset class and horizon.

• It is not a price target or forecast.
• It is not a buy or sell recommendation.
• It is not personalized to your account, risk tolerance, or holdings.
• It is not a probability of profit.

Autonium is a self-directed research workspace. You decide what to do with the score.

Every score is computed against a selected horizon. Signals are re-weighted so that short-term signals count more on shorter horizons and slower structural signals dominate longer horizons.

• Short (days–weeks) — emphasises momentum, volume, and volatility regime.
• Medium (weeks–months) — emphasises trend, relative strength, and options positioning.
• Long (months–quarters) — emphasises trend persistence, drawdown behaviour, and macro/sector context.

Signals are grouped into categories. Category names are public; the exact formulas and thresholds are proprietary.

• Trend & momentum — direction, strength, and acceleration of the price series.
• Relative strength — performance versus benchmark and versus sector.
• Volatility regime — realized and implied volatility, regime detection.
• Volume & flow — participation, breadth, and options flow context.
• Smart money & institutional — filings, aggregate positioning, unusual activity.
• Sentiment — social and news signal weighted by source quality.
• Risk penalty — drawdown severity, concentration, and macro headwinds.

• Algo Scores recompute daily at the US session close.
• Model portfolios are evaluated daily and rebalanced only when the model's rules trigger — typically a few times per week.
• Sentiment and smart money data refresh multiple times per day.
• Price data is delayed per market data licenses (typically ~15 minutes for US equities).

Crypto and equities are scored on the same 0–100 scale but with different normalization to account for 24/7 markets, higher realized volatility, and different liquidity structures.

• Crypto uses a wider volatility band and different volume normalization.
• Crypto benchmarks are asset-class specific (e.g. BTC) rather than SPY.
• Options-flow signals apply only to equities.

Each model portfolio has a rule-based universe of eligible assets.

• Autonium Core — US large-cap equities that meet liquidity and score-history requirements.
• Autonium Prime — a narrower momentum sleeve drawn from the Core universe, plus a defensive fallback when the regime turns unfavourable.
• Assets are added and removed by rules, not discretion. Universe changes are versioned.

• Positions are sized by rule, with per-name caps to limit concentration.
• Cash is held when fewer eligible names meet the ruleset.
• No leverage. No short positions in the public model portfolios.

• The model evaluates every eligible asset daily.
• A rebalance is triggered only when a rule condition changes (entry, exit, or resize).
• Trades are executed at the next available close in the model.
• All rebalances are logged and time-stamped.

• Equity portfolios are benchmarked against SPY (S&P 500 ETF, total return proxy).
• The benchmark is chosen for investability, not just correlation — it's what a passive alternative would return over the same window.

Public performance shown on the site is the live model track(walk-forward), not a curve-fit backtest. When historical backtests are referenced elsewhere, they follow these rules:

• Point-in-time data only. No hindsight leakage from restated fundamentals or later universe changes.
• Assumed 10 bps per side transaction cost on equities and 25 bps on crypto.
• Assumed 5 bps slippage against the next available close.
• No survivorship bias filtering beyond documented universe rules.

• Signals can lag around unpredictable single-day events (earnings, macro shocks, headlines).
• Model portfolios are hypothetical — no orders are executed at a broker and no fund is managed. Real execution can differ.
• Institutional data (13F, insider filings) is delayed per SEC filing windows.
• Universe changes and rule updates are versioned but can affect comparability over long windows.
• Past performance does not guarantee future results.