How forecasts, AI advisors, prompt assembly, and consensus scores actually work

Each forecast starts with a task configuration. The task decides which asset is being analyzed, which forecast horizon is requested, which AI advisor persona is speaking, which execution mode is used, which market data is injected, and which structured output schema the model must satisfy.

The prompt is assembled from reusable components rather than written as one loose paragraph. A persona core definition sets the investing identity, a subject-category head adapts that identity to stocks, crypto, commodities, indices, or forex, a mode head controls how deeply and creatively the advisor reasons, and task instructions define the exact forecast job.

The model response is then parsed into structured fields: rating, forecast timeseries, scenario rationale, key drivers, frictions, tail risks, tail opportunities, and investment thesis text. Validation checks make sure the output has the expected horizon, step count, dates, rating logic, and schema shape before the result is packaged for the app.

Investment research is not only about getting one prediction. It is about seeing which assumptions drive the prediction. A value-oriented advisor may care most about owner economics and margin of safety, while a macro advisor may care about rates, liquidity, energy shocks, and policy cycles. A risk advisor may reject a forecast that looks attractive on return alone.

Putting those voices next to each other gives the user more information than a single blended answer. Agreement can become a stronger signal. Disagreement can be even more useful because it tells the user where the forecast is fragile and which assumption deserves more attention.

The formula starts from the forecasted compounded return over the selected horizon. When dividend mode is enabled and a usable dividend history exists, iPulse adds an annual net dividend uplift before annualizing the total return. That keeps high-income equities from being judged only on price appreciation.

The annualized return is scored with a blended signal: 25% risk-adjusted return using historical volatility, and 75% return-anchored signal using a fixed 15% denominator. This keeps volatility awareness in the score without letting very low historical volatility dominate the ranking.

The final score is scaled by advisor consistency and reduced by a non-linear forecast volatility penalty. Buy-side guards can downgrade bullish scores to neutral if the annualized return is too weak or if the lower return band falls below a safety threshold.

Every prediction is tied back to the advisor persona, model, task configuration, output format, generation batch, and forecast horizon that produced it. That lineage matters because performance cannot be improved if the system cannot tell which version created which result.

As more predictions age into measurable outcomes, iPulse can compare forecast paths, ratings, and consensus signals against what actually happened. The goal is to improve prompt composition, advisor frameworks, validation rules, and task design instead of relying on one static prompt forever.

iPulse is currently a market intelligence product, not a portfolio management product. The app lets users inspect asset forecasts, advisor reports, consensus views, leaderboards, and historical prediction context without uploading their private financial life.

User accounts are used for authentication, approval status, subscriptions, and insight credits. Market data and prediction data are served through backend APIs; the frontend does not need direct database access to display the research experience.

In iPulse, Warren Buffett is not treated as a celebrity slogan. The persona is a structured analytical framework. For equities, it asks whether the business is understandable, whether the moat is widening or eroding, whether owner earnings are durable, whether capital allocation is rational, and whether the current market price leaves room to be wrong.

For indices, the same voice adapts to basket-level questions: aggregate moat quality, methodology governance, valuation of the underlying holdings, and whether the basket behaves like a collection of durable compounding businesses or a group of overvalued claims.

That voice is intentionally different from a growth-disruption advisor or a macro-cycle advisor. The value framework slows the forecast down and asks whether the asset deserves long-term ownership, not merely whether the price might move next month.

Coverage includes US and international equities, major crypto assets, spot commodities, selected commodity-linked markets, major forex pairs, and index-tracking markets such as SPY, QQQ, TQQQ, VTI, and IWM.

Forecast workflows can produce ratings, thesis summaries, and timeseries paths across multiple horizons. The goal is to show both the near-term scenario and the long-range thesis rather than one isolated price target.

A clean consensus can be useful, but disagreement is often where the research becomes interesting. If the value advisor is negative and the growth advisor is positive, the real question may be whether future growth can outrun current valuation. If the macro advisor is cautious while company-level advisors are bullish, the risk may sit outside the business itself.

iPulse is designed to help users inspect those fault lines. The goal is not to force agreement. The goal is to make the disagreement legible enough that a user can decide which assumption they believe.

For dividend-paying assets, price-only forecasts can understate total return. iPulse can estimate a historical annual net dividend yield and compound it over the forecast horizon before scoring the annualized total return.

This matters most for equities and income-oriented markets. Crypto, commodities, and assets without distributions generally receive zero dividend uplift, so the toggle does not change their score.

iPulse forecasts are model-generated research outputs, not instructions to buy or sell. They can help users compare scenarios, inspect assumptions, and discover where advisor views agree or disagree, but they do not know a user's full financial situation, tax constraints, liquidity needs, or risk tolerance.

Users should treat iPulse as one research input and make decisions using their own judgment, risk limits, and professional guidance when needed.