Inside the MarketPulse Intelligence Engine.
MarketPulse is designed to make the calculation visible: raw market data becomes independent model scores, the current market regime changes their importance, and the ensemble produces one score plus three time-horizon outlooks.
Regime-aware weighting
The engine does not use the same formula in every market.
The model stack
Price versus 20/50/200-day averages, slope and persistence.
5/20/60-day acceleration plus MACD confirmation.
Relative volume, participation and price/volume confirmation.
Performance versus the broad-market benchmark.
Annualized realized volatility and stability.
Drawdown depth and downside behavior.
RSI extremes and distance from trend.
Recent highs, momentum and volume confirmation.
Dollar-volume quality and tradability.
Current setup compared with normalized historical behavior.
Revenue, earnings, cash flow, margins and balance-sheet quality.
Price relative to company fundamentals and peers.
Durability and quality of reported operating performance.
Company strength relative to its sector.
Broad economic conditions and market-sensitive inputs.
Event and sentiment inputs when reliable data is available.
Longer-term company quality layer.
How well the company setup matches the current market environment.
MarketPulse Score
A 0–100 weighted ensemble score. It summarizes the current setup without hiding the underlying model votes.
Confidence
Confidence rises with model agreement and data depth, and falls when signals disagree or volatility becomes extreme.
Three horizons
Short-term, 1–3 month and long-term scores use different combinations of momentum, trend, fundamentals, valuation and risk.
Prediction ledger & calibration
Production predictions are intended to be timestamped and retained. Once enough history exists, MarketPulse can calculate actual 7-day, 30-day and 90-day outcomes by score band, calibrate probability estimates, identify weak models, and improve weights without rewriting old predictions. This is how the platform can eventually publish an auditable historical record instead of marketing claims.
Cost architecture
The prediction engine uses ordinary math and statistics for the core calculation. Market data is cached server-side. AI is optional and should be reserved for cached explanations or news summaries, so visitor growth does not create an equivalent growth in token cost.
Important
Scores, confidence, outlooks, rankings and historical statistics are impersonal research outputs—not personalized investment recommendations or guarantees. A high score can still lose money, and historical performance does not guarantee future results.