How the score is calculated
Every number on this platform is produced by written rules operating on published data. No part of a score is estimated, guessed, or produced by an AI model. You can reproduce any composite by hand from the figures shown on the stock page — if the arithmetic on screen does not add up, that is a bug and we want to hear about it.
What the score is for: judging how fragile or solid a company looks on its published figures, against others in its industry. Over 18 years of market history, companies this score rated weakest lost more than half their value roughly three times as often as those it rated strongest.
What it is not for: anticipating price. We tested that question directly, across both investor profiles and every horizon from three days to two years, and found no such relationship — so we make no claim of one. The full results, including the ones that do not flatter us, are on the proof page.
The short version
- Each company is measured on five dimensions, each scored 0–100.
- Most measurements are relative: a company is ranked against companies in its own sector and region, because a 22× price-to-earnings ratio means something different for a utility and a software company.
- The five dimensions are combined using weights that depend on your investor profile.
- If too much data is missing, we publish no score at all rather than a number we cannot stand behind.
The five dimensions
- fundamental
- Is this company financially healthy and attractively valued?
- Free cash flow yield, price-to-earnings, EV/EBITDA, return on invested capital, return on equity, debt-to-equity, revenue growth and gross margin — each ranked against sector peers. Banks and insurers use a different metric set, because debt is their raw material rather than leverage.
- technical
- What is price momentum saying, and is entry timing favorable?
- Trend structure across 20/50/200-day moving averages, MACD, RSI, volume trend, Bollinger position and average true range. All standard published parameters — nothing invented here.
- macro
- Is the current macro environment favorable or hostile for this stock?
- A single economic regime score from the policy rate, the inflation trend, the yield curve and credit spreads — then adjusted for how sensitive the company’s sector is to that environment. The regime is the company’s own: a Japanese company is read against Bank of Japan policy and Japanese prices, not American ones.
- sentiment
- What is the collective market mood, and is it shifting?
- News tone over 30 days weighted toward recent stories, how the analyst rating distribution leans and which way it moved, net insider buying and selling, and short interest.
- risk
- How much downside risk does this stock carry, and what kind?
- Realised volatility, maximum drawdown, trading liquidity, beta against a regional benchmark, and balance-sheet strength. Inverted, so a high score means low risk.
The risk dimension is inverted. A high risk score means low risk — a stable, liquid, lightly-indebted company scores near 100. It is labelled on every screen because it is the single easiest thing to misread.
How your profile changes the score
The same company is measured identically for everyone. What changes is how much each dimension counts.
| Dimension | Value Investor | Swing Trader |
|---|---|---|
| Fundamental | 45% | 10% |
| Technical | 10% | 45% |
| Macro | 20% | 20% |
| Sentiment | 5% | 15% |
| Risk | 20% | 10% |
The technical dimension goes further than reweighting: the same chart is interpreted differently. A stock that has fallen hard reads as a possible discount to a value investor and as absent momentum to a swing trader, so the two profiles score an identical price history differently — not merely weight it differently.
What the labels mean
| Score | Label | Meaning |
|---|---|---|
| 70–100 | Strong Positive | Signals aligned |
| 62–69 | Favorable | Mostly positive |
| 47–61 | Monitor | Mixed signals |
| 38–46 | Elevated Risk | Risk rising |
| 0–37 | Weak Signal | Low priority |
These bands describe where a stock stands relative to the market we cover. Each dimension averages its own components and the composite averages the dimensions, so scores cluster toward the middle rather than spanning 0–100; the boundaries above are the percentiles of the scores we actually observe, which is why “Strong Positive” begins at 70 rather than 85.
These are descriptions of what the data shows, not instructions. A “Weak Signal” is not a warning to sell — it means the measurements we take are not currently favourable, which is a different statement.
When we refuse to show a score
Real data has holes. Rather than filling them with assumptions, we do this:
- A missing measurement is dropped and its weight spread over the rest.
- A dimension missing more than 40% of its inputs is not scored at all, and its weight moves to the others.
- Below 60% total coverage, no composite is published — the page says so instead of showing a number.
- Every stock page shows its coverage and a confidence level, so you always know how complete the picture is.
Known limitations
Publishing these is deliberate. A methodology without limitations is marketing.
- Not every country has an economic regime
Twenty-one regimes are published, covering the United States, the euro area, the United Kingdom, Switzerland, Sweden, Norway, Denmark, Japan, Korea, China, India, Taiwan, Indonesia, Australia, Canada, Brazil, Mexico, Chile, Colombia, Israel and South Africa. A company from anywhere else carries no economic score at all and its weight moves to the other four dimensions. We would rather show nothing than describe a Singaporean company with American interest rates. Until 27 October 2026 that is exactly what this product did.
- No regime has every component
No free-licensed manufacturing survey exists for any country, so no regime has a growth component. Several markets publish no high-yield credit spread and several publish no government-bond yield series we can read. Each company’s page names what its own regime goes without.
- Sentiment is thinner than it will be
Options positioning and social sentiment are not yet included. Short interest comes from FINRA, which covers US-listed securities only, so the component is empty for every other listing and its weight moves elsewhere.
- News is not collected for every company
Our news source carries articles for US-listed companies and almost nothing for a company listed in Frankfurt or Paris. Where no news source covers a listing at all, news is left out of that company’s sentiment score rather than counted as a gap, and the analyst rating balance carries the dimension instead. A quiet month at a company the source does cover is a different thing, and does lower its coverage.
- Restated accounts
When a company restates prior figures, our data provider may supply the corrected numbers rather than the originals. Historical testing therefore sees slightly better information than was actually available at the time. We cannot fully eliminate this at our data tier and would rather say so.
- Return on invested capital assumes a tax rate
Our data does not include tax expense, so a flat 25% is assumed. It is applied to every company equally, so it does not disturb the ranking — but the absolute figure is an approximation.
- No trading costs
Nothing here accounts for commissions, spreads or taxes. Any comparison of scores to outcomes is before costs.
Accuracy
We have run that test. The engine was replayed over 18 years of market history — roughly 53,000 measurements across 213 monthly dates — scoring every company using only the filings and prices that were public on each of those dates.
What held up. Companies in the weakest band lost more than half their value 4.6% of the time over the following year, against 1.6% for the strongest band — roughly three times as often. The pattern appeared in 75% of the periods measured, and through the 2008 and 2020 crashes the gap was wider still.
What did not. A high score did not lead to a larger gain. We tested that across both investor profiles and every horizon from three days to two years, and found no such relationship — so we claim none. Strongly-scored companies returned less than an equal share of everything we cover, because the score favours steadier companies, and steadier companies fall less in a crash but also recover less in a rally.
One limit worth stating plainly: companies that went bankrupt are largely missing from our historical data, so the collapse rates above are conservative floors rather than exact figures. Including those companies would most likely widen the gap between the bands, not narrow it. The full results, with the method and the rest of the limitations, are on the proof page.