Cookie preferences

Essential storage keeps iPulse AI reliable. Optional analytics and LinkedIn ad measurement stay off unless you accept. Privacy Policy

Skip to main content

Consensus score

Consensus starts after individual AI Agent disagreement is preserved

The iPulse AI Consensus Score compresses multiple advisor forecasts into a bounded signal, using V8 G10 annualized total return, bounded forecast disagreement and economic guardrails. Historical volatility remains visible and powers the separate Risk-Adjusted alternative.

Updated Updated by Russlan Ramdowar

Overview

A score is a summary, not a guarantee

iPulse AI first produces structured forecasts from individual AI Agents. The Consensus Score then summarizes the direction and strength of the grouped forecasts into a public score from 0 to 10. Higher scores indicate a stronger bullish ranking signal. Lower scores indicate a stronger bearish signal. G10 considers event risks, historical price volatility and Financial Health as separate evidence.

The score is designed for educational market research and comparison. It is not personalized financial, investment, tax, or legal advice.

Engine V7

Evidence-sensitive synthesis is a separate layer from the ranking score

Batch 7 adds explained per-opinion weights to the synthesized forecast path. The synthesizer examines common premises, stale evidence, contradictions and strong contrarian cases. It sees investment frameworks with blinded opinion identities; underlying provider and model labels are withheld from the weighting decision.

Each submitted opinion receives one fixed weight across all 20 quarters and a rationale, including any zero allocation. The parser normalizes a positive weight total. Code then calculates each quarterly return as the sum of the weighted source returns and compounds the result from the frozen anchor. Equal weights remain appropriate when the evidence does not justify differences.

When search is enabled and useful, material disputed claims can be checked against primary sources within the evidence cutoff. Verification that changes a weight must be explained with the retrieved source. If verification is unavailable or inconclusive, uncertainty remains explicit. These controls are designed to reduce the influence of unsupported claims and preserve sound minority reasoning; they do not establish a measured reduction in hallucinations or improved forecast accuracy.

The averaged consensus, synthesizer-weighted path and quantitative Consensus Score remain distinct. Historical portfolio simulations use their frozen original quantitative leaderboard, not the synthesized prices. Individual forecasts are preserved.
Read the Batch 7 release announcement

Current consensus outlook

Freshness and fixed target dates limit current recommendations

Current consensus outlook ratings require a recorded market close no more than 7 UTC calendar days old, or 3 days for crypto, and underlying analysis no more than 180 days old. The original analysis date owns that age; refreshing a quote or a page does not renew the research.

A horizon with 90 days or less remaining has its current rating withheld to avoid amplifying a small remaining return through annualization. Once its fixed target date arrives, its forward return is also withheld. Missing, future or inconsistent dates withhold the rating. If either the one-year or five-year horizon fails these checks, the combined recommendation is paused, including dividend-based actions.

Recorded targets and dated returns remain available for inspection, and historical publications retain their original calculations. These are conservative publication eligibility limits, not backtested trading thresholds or a change to the quantitative leaderboard formula. Each page shows its eligibility-check date; cached pages refresh through the normal regeneration cycle rather than continuously.

Score scale

How the 0–10 score is calculated

Add 1000 to the internal signed score and divide by 200. The endpoints map to 0.0 and 10.0. Internal zero maps to 5.0, the scale midpoint. Public pages round to one decimal.

An internal reading of +514 appears as 7.6. It is a ranking signal. Ordering and rating thresholds use the original precision and economic guardrails, so assets with the same displayed score can have different ranks or ratings. Historical pages convert the scores recorded at publication.

iPulse AI Consensus Score

Example bullish consensus reading on the same 0–10 scale used on asset pages.

7.6/10

0.0 Bearish5.0 Midpoint10.0 Bullish

Formula at a glance

Search-readable summary of the Consensus Score calculation

V8 distinguishes original G10 from daily rescoring. For one-year assets passing current BUY score and economic guards, the ranking score is 75% frozen-original G10 plus 25% updated G10, floored at neutral zero. This applies across all asset classes. Five-year and current non-BUY scores retain updated G10. Displayed returns still reflect current prices and the remaining forecast period; only the score receives this continuity adjustment. Old originals and past publications keep their own recorded methods.

The iPulse AI Consensus Score formula starts with annualized total return, subtracts a cash hurdle, discounts fragile upside or downside using mean absolute deviation, applies a bounded forecast-dispersion penalty, and produces an internal score bounded by −1000 and +1000. The final published ranking score is then mapped to 0–10. The 1Y and 5Y formulas use their own annualized returns without a numerical cross-horizon blend.

Primary equation

Base G10 = 1000 × tanh(E / 30) × (1 − P). Public score = (final signed score + 1000) / 200, rounded to one decimal.

E = sign(R−3.6) × (|R−3.6| − min(0.35 MAD, 0.60 |R−3.6|)); direction = tanh(E/30)

Current formula version

The current public methodology describes V8 · G10 Return-Focused. Formula version is stored with leaderboard output so historical ranks can be traced to the scoring rules used at generation time.

Direction Signal
R is the displayed annualized total return for the selected horizon. The 1Y rank uses 1Y evidence and the 5Y rank uses annualized 5Y evidence. There is no numerical cross-horizon blend. Other-horizon evidence remains available to editorial review.
Advisor Agreement
Mean absolute deviation and the population standard deviation of annualized advisor returns discount forecast disagreement. A separate consistency multiplier is not used in V8.
Forecast Dispersion Penalty
forecast_std is the population standard deviation of grouped annualized advisor returns. It measures forecast disagreement, not historical price volatility or step-to-step path volatility. The dead zone and 15% cap preserve the return signal.
Event-Risk Pressure
Event risks and Financial Health remain visible independent quality evidence. Editorial reviewers can decline an otherwise eligible asset when current evidence makes it less attractive, while preserving its quantitative rank and explaining the decision.

Inputs

The score listens to more than average return

A simple average can overstate conviction when forecasts disagree or when the forecast path is unstable. iPulse AI therefore considers return direction, return magnitude, cash hurdle, dividend-inclusive total return, forecast dispersion and economic guardrail rules. Historical volatility, event risks and Financial Health remain separate diagnostics and editorial evidence. Column-level diagnostics such as P/E, dividend contribution, Event Risks, and Financial Health are explained separately in the Top Picks methodology.

The formula deep dive below appears after prompt assembly in the docs flow because the score only makes sense once the AI Agent, mode, prompt, context, and output schema pipeline is understood.

What improved

The formula changed after ranking edge cases exposed weak assumptions

Consensus scoring has been refined through operating experience. The main lesson is that a clean-looking formula can still rank assets badly if it hides return magnitude, over-penalizes normal disagreement, or lets small low-volatility moves dominate larger economic outcomes.

Return-Focused and Risk-Adjusted serve different objectives

V8 uses G10 for the principal return-focused ranking and B02 for the separate Risk Adjusted Stock Picks. Retrospective results vary by batch, portfolio breadth and exceptional winners. The choice expresses an objective, not proven dominance.

Dispersion should reduce confidence, not erase signal

Advisor disagreement is informative. The score discounts excess return when forecasts disagree, but caps that discount so a real directional signal is not zeroed out mechanically.

Volatility penalties need a dead zone

The population standard deviation of grouped annualized advisor returns is penalized only above a return-dependent dead zone, with a 15% cap. This is forecast dispersion, not historical price volatility.

Positive return is not automatically a buy

The score uses cash-hurdle and spread-quality guards so low positive returns or fragile upside can remain neutral. This keeps public labels more conservative and easier to defend.

Event risk should not double-count volatility

Event Risks now focus on clustered consensus frictions and tail risks from the SEO/GEO summary layer. Historical volatility remains a separate diagnostic; advisor disagreement is captured by the bounded G10 discounts.

V8 · G10 Return-Focused

How the iPulse AI Score is calculated

G10 produces a signed internal ranking score that is converted to a public iPulse AI Score from 0 to 10. Each horizon uses its own annualized forecast total return, with bounded discounts for advisor forecast disagreement. The score is a ranking signal, not a probability or a guarantee.

STRONG BUY

> 6.5

BUY

5.7 to 6.5

NEUTRAL

4.9 to 5.7

PARTIALLY SELL

4.0 to 4.8

SELL ALL

< 4.0

From forecasts to rank

  1. Group the independent forecast voices using the selected prediction, advisor or persona mode.
  2. Combine their compounded returns, including eligible regular dividends, then annualize over the selected remaining horizon. Annualize the consensus compounded return rather than averaging annualized returns.
  3. Subtract the 3.6% cash hurdle and discount excess-return magnitude using mean absolute deviation, capped at 60% of its magnitude.
  4. Apply tanh compression and the forecast-dispersion penalty capped at 15% to obtain base G10. For current 1Y BUY candidates, apply the explicit original/current continuity rule before ranking. Rank using the full-precision final score, with asset ID as a deterministic tie breaker.
  5. Apply the existing signal bands and economic guards. Display the original universe rank even inside an editorial category.

One-year daily rescoring

1Y eligible BUY: max(0, 0.75 × G10_original + 0.25 × G10_updated)

One-year recomputed rankings use 75% original G10 plus 25% updated G10 for assets passing the current BUY score and economic guards, across all asset classes. The fixed original reference uses the batch’s frozen inputs, recalculated under G10 for older batches; it is never yesterday’s blended score. A negative blend becomes neutral zero. Five-year and current non-BUY scores use updated G10. Displayed returns remain current market-based calculations. Original forecasts, historical scores and published editorial editions stay unchanged.

Final consensus score

Base G10 = 1000 × tanh(E / 30) × (1 − P). Public score = (final signed score + 1000) / 200, rounded to one decimal.

G10 prioritizes annualized forecast total return after a 3.6% cash hurdle. Mean absolute deviation discounts fragile excess return, and cross-advisor forecast standard deviation adds a bounded penalty of at most 15%. Historical volatility, separate consistency, event risks and other horizons do not change this score.

Direction signal

Excess = R − 3.6
E = sign(Excess) × (|Excess| − min(0.35 × MAD, 0.60 × |Excess|))
Direction = tanh(E / 30)

R is the displayed annualized total return for the selected horizon. The 1Y rank uses 1Y evidence and the 5Y rank uses annualized 5Y evidence. There is no numerical cross-horizon blend. Other-horizon evidence remains available to editorial review.

Dividend uplift

Adjusted Return = ((1 + Price Return) x (1 + Annual Dividend Yield)^Years - 1)

Leaderboard ranking uses dividend-inclusive total return as the canonical basis. Price-only comparison is shown only when ex-div diagnostic fields are available for that snapshot.

Volatility penalty

b = clamp(0.5 × |R|, 1, 10); P = 0.15 × min(1, (max(0, forecast_std − b) / max(15 − b, 1))²)

forecast_std is the population standard deviation of grouped annualized advisor returns. It measures forecast disagreement, not historical price volatility or step-to-step path volatility. The dead zone and 15% cap preserve the return signal.

Event-risk diagnostic

Event Risk Pressure = percentile(0.60 × weighted consensus frictions + 0.40 × weighted consensus tail risks); diagnostic only in V8

Event risks and Financial Health remain visible independent quality evidence. Editorial reviewers can decline an otherwise eligible asset when current evidence makes it less attractive, while preserving its quantitative rank and explaining the decision.

BUY and STRONG BUY are downgraded to NEUTRAL when annualized expected return is below the 3.6% cash hurdle or when the lower return spread (return minus MAD) is below -1%. If the annualized economic return is below −0.5%, the final signal is forced to at least PARTIALLY SELL. This guard changes the signal label, not the G10 score.

1Y and 5Y remain distinct horizons

R is the displayed annualized total return for the selected horizon. The 1Y rank uses 1Y evidence and the 5Y rank uses annualized 5Y evidence. There is no numerical cross-horizon blend. Other-horizon evidence remains available to editorial review.

One-year recomputed rankings use 75% original G10 plus 25% updated G10 for assets passing the current BUY score and economic guards, across all asset classes. The fixed original reference uses the batch’s frozen inputs, recalculated under G10 for older batches; it is never yesterday’s blended score. A negative blend becomes neutral zero. Five-year and current non-BUY scores use updated G10. Displayed returns remain current market-based calculations. Original forecasts, historical scores and published editorial editions stay unchanged.

V8 G10 is the principal Return-Focused method; B02 is the separate Risk-Adjusted alternative. Existing signal bands and economic guards are retained without a new calibration claim. The choice reflects a return objective and retrospective experiments, not demonstrated forecasting superiority or five-year validation. Frozen forecasts and past publications remain unchanged.

The Risk-Adjusted alternative

Historical price volatility is a separate diagnostic in V8. The Risk Adjusted Stock Picks alternative uses B02: 1000 × tanh(((R − 3.6) / max(historical volatility, 5)) / 2), with sufficient provider-action-adjusted history. Missing volatility makes B02 unavailable.

Risk Adjusted Stock Picks
R_base,i
price-only compounded return for one prediction voice from the generation anchor price
y
modeled net regular trailing 12-month cash yield at the current price; original snapshots preserve their published input
Y
remaining forecast horizon in current rankings, or full horizon in published-anchor illustrations, in years
R
annualized total return after dividend uplift (or zero uplift when no usable dividend history exists)
E
excess return above the 3.6% cash hurdle
R_conf
confidence-adjusted excess return after the capped MAD discount
MAD
mean absolute dispersion of grouped annualized returns, used to discount fragile upside
sigma_hist
10-year historical annualized volatility, floored at 5%
forecast_std
population standard deviation of grouped annualized forecast returns
P
forecast-dispersion penalty, bounded from zero to 0.15
Risk
0-100 event-risk pressure percentile from clustered consensus frictions and tail risks
sell guard
minimum sell-side visibility when annualized economic return is below -0.5%

Limits

How to read the score responsibly

Consensus is useful, but disagreement is also information. A high score should still be read with the supporting AI Agent reports, forecast horizon, asset volatility, thesis risks, and scenario assumptions.

Eligible signed-in users can open Config Details on advisor reports to inspect which model, mode, task, and prompt assembly were used for a given prediction.

Read Config Details tutorial