Methodology
Last updated September 2026
This page explains how Prism.AI computes True Exposure and the Diversification Score: what the number means, what it does not measure, and how the look-through calculation works under the hood.
What it measures
The Diversification Score is a single number from 0 to 100 that summarises how evenly your capital is spread across individual companies and sectors, after looking through your ETFs, funds, and other holdings.
The score is not a recommendation and it is not a verdict. If you hold a concentrated portfolio on purpose (a high-conviction bet on a single sector, a barbell strategy, a single-stock position), a low score is telling you something you already know. If you thought you were diversified, a low score shows you where you actually stand.
The score is composed of two components:
- Position component (60%): how evenly your capital is spread across individual companies, after look-through.
- Sector component (40%): how evenly that same capital is spread across the 11 GICS sectors.
Descriptive band labels:
| Range | Label |
|---|
| 0–25 | Highly concentrated |
| 26–50 | Concentrated |
| 51–75 | Moderately spread |
| 76–100 | Broadly spread |
What it does not measure
The score measures only spread. It says nothing about:
- Quality or valuation. A portfolio spread evenly across 500 poor-quality companies scores 100.
- Expected return. Concentration can be deliberate and rational.
- Correlation. Two "different" sectors can move together in a crisis. HHI does not know this.
- Volatility or drawdown risk. The score has no awareness of beta or historical vol.
- Fees. ETF expense ratios are not modelled.
- Suitability. What is right for one investor is wrong for another. Prism does not know your situation.
The score is a description of your portfolio's structure as it stands today. It is not a grade and it is not advice.
How look-through works
When you hold an ETF or fund, you do not own it outright; you own a proportional share of every company inside it. If you hold NDQ and also hold NVDA directly, your true NVDA exposure is higher than your allocation chart shows.
For each holding in your portfolio, Prism does the following:
- Single stock: the full stated weight goes to that company.
- ETF or fund in our data file: the fund is decomposed. Each constituent gets
fund_weight × constituent_weight_within_fund. Contributions from multiple sources are summed.
- ETF or fund not in our data file: treated as a single opaque position; the detail block notes which holdings could not be resolved.
- Fund-of-funds (e.g. VDHG, DHHF): one level of nesting is resolved, then stopped.
The result is a single map of company → true weight across your entire portfolio.
Example only
If a portfolio held the NDQ ETF at 40% and also held NVDA directly at 10%, and NDQ's weight in NVDA is approximately 9%:
True NVDA exposure = (40% × 9%) + 10% = 3.6% + 10% = 13.6%
This is higher than the 10% shown in the allocation chart. The overlap callout surfaces this difference. This is an illustration of how the calculation works. It is not a comment on NDQ, NVDA, or either as an investment.
The residual
Our data file holds the top 25 constituents per fund by weight. For a broad fund like VTI (approximately 3,700 holdings), the top 25 cover roughly 40% of the fund's weight. The remaining 60% is modelled as equally-weighted pseudo-positions summing to the residual weight.
This residual model is not optional. Without it, a total-market ETF would behave like a 25-stock portfolio and the score would be badly wrong. The residual pseudo-positions carry no ticker, sector, or country identity; they contribute to the position component but not to the sector component.
The maths
Herfindahl–Hirschman Index (HHI)
The underlying measure is the Herfindahl–Hirschman Index, a standard measure of concentration borrowed from competition economics. For a set of weights summing to 1:
HHI = Σ(wᵢ²)
HHI ranges from 1/n (perfectly even across n positions) to 1 (fully concentrated in one position). Its inverse, effective N, is the number of equally-weighted positions that would produce the same HHI:
effective N = 1 / HHI
Position component
The position score maps effective N to 0–100 on a logarithmic curve, targeting 50 effective positions as the "broadly spread" reference:
positionScore = clamp(100 × ln(effN) / ln(50), 0, 100)
Expected values:
effN = 1 → score ≈ 0 (single position)
effN = 5 → score ≈ 41
effN = 10 → score ≈ 59
effN = 25 → score ≈ 82
effN ≥ 50 → score = 100
The curve is logarithmic because the difference between 3 and 6 effective positions is meaningful; the difference between 40 and 50 is not.
Sector component
The same HHI maths applied to sector-aggregated look-through weights, targeting 8 effective sectors (an even split across all 11 GICS sectors almost never occurs in practice):
sectorScore = clamp(100 × ln(effSectors) / ln(8), 0, 100)
Tickers whose sector is unavailable are excluded and the remaining weights are renormalised. An "unknown" bucket is never created; that would penalise ASX-heavy portfolios for a data gap, not a real property of the portfolio.
If sector data covers less than 5% of total portfolio weight, the sector component is suppressed and the final score equals the position component alone.
Final score
score = round(0.6 × positionScore + 0.4 × sectorScore)
If sector data is unavailable: score = round(positionScore)
