How the score is built · Updated September 2026

A number, and everything it doesn't know

Every company in the deck carries a score out of 100. Here is exactly how it is calculated, what data feeds it, and the list of things it cannot see. The last part is longer than you might expect.

The score is a weighted blend of five factors, each read from figures the company filed with the SEC or from its current market price. It is arithmetic, not judgement. Nothing about it is predictive, and no part of it has been tuned against how these stocks subsequently performed. It is a screen: a way of asking whether the reported numbers look cheap, whether they are growing, and whether much of the revenue survives to the bottom line.

A company scoring 76 is not a better investment than one scoring 41. It is a company whose filings score higher against these particular thresholds, which is a much narrower claim.

The five factors

Weights, which sum to 1.00
Factor	What it reads	Weight
Profitability	Margins, return on equity or assets, and how much profit arrives as cash	0.24
Value	What you pay for the earnings (trailing and forward), sales, book value and free cash flow	0.22
Growth	Revenue and earnings this year, and revenue and earnings over five years	0.22
Momentum	How the price has behaved lately	0.18
Stability	How violently it moves, what it owes, how easily it covers its interest, and what it pays you	0.14

Where a company has not reported the figures a factor needs, that factor is dropped and the remaining weights are rescaled. A factor resting on only one or two figures is pulled part of the way back toward the middle, after the sector half described below is blended in, so thin evidence cannot produce an extreme reading.

How a factor becomes a number

Each input is mapped onto a 0–100 curve through fixed anchor points, with straight-line interpolation between them. The anchors were chosen to sit near long-run market norms, not fitted to any historical return. Value, for instance, reads the price-to-earnings ratio like this:

Price to earnings, mapped
P/E	5	10	16	25	40	70	120
Score	92	84	70	52	34	16	8

Forward P/E, price-to-book, price-to-sales and free-cash-flow yield are mapped the same way and averaged in, with trailing earnings counted twice because it is the figure most people are actually pricing. Growth reads this year's revenue and earnings per share and the five-year revenue and earnings trends, with five-year revenue counted twice because a single year is noisy. Stability reads beta, the width of the 52-week range, dividend yield, the debt and interest cover. Profitability and momentum work the same way.

Against its own sector

Fixed thresholds treat a supermarket's 3% margin and a software company's 30% as the same kind of number. So every factor except momentum is read twice: once against the fixed curves above, and once as a rank among the other companies in the same sector. The two are averaged. The fixed half keeps a whole sector from looking cheap just because all of it is; the sector half stops a grocer being marked down for a grocer's margins. Momentum stays absolute, because a sector falling together is not a reason to prefer the one falling slowest.

Numbers that lie, and what is done about them

Negative equity. Companies that have bought back more stock than they have equity, often with borrowed money, report returns on equity of several hundred percent. That is arithmetic, not excellence. When equity is negative or a sliver of assets, return on equity, price-to-book and debt-to-equity are left out wherever they appear (the score, the pros and cons, the sector ranks and the figures on the card), return on assets stands in, and the company is flagged.
Losses. A company losing money has no P/E. Earlier versions skipped the figure, which left the value factor to be judged on sales alone and made losses look cheap. A loss now counts as expensive on earnings. “Losing money” means over the last twelve months; the last annual filing decides only when there is no recent figure. When the provider's P/E says the company made money and its margin says it did not, the P/E is not counted in its favour.
Debt. The data provider's debt-to-equity figure is occasionally near zero for companies with billions of long-term debt in their own filings. Leverage is now read from the filings as the years of operating cash flow it would take to repay the debt.
Financials. A bank, insurer or broker borrows by design, so for every company in the Financials sector debt and cash flow are not scored. The sector label cannot tell a bank from a payment network or an exchange, so those are treated the same way. A bank's one-year revenue in the provider's data can swing 60–180% with interest rates; when that is wildly out of line with the five-year trend, only the five-year figure is used, and the one-year figure is left out of the sector ranks too.
Share classes. Alphabet, Fox and News Corp each trade under two tickers with the same filings. Each company is counted once in its sector, and is never its own closest peer.

The five results are multiplied by their weights and summed. Above 70 the card says screens strongly; below 33, screens badly. The bands in between are labels on a continuum, not categories.

Where the numbers come from

Prices, ratios, the 52-week range and upcoming earnings dates come from Finnhub. Everything on the balance sheet and cash flow statement is pulled straight from the company's own XBRL filings at SEC EDGAR, along with the business description and risk factors quoted from its latest 10-K.

The whole set rebuilds each weekday evening after the American close, so prices are that session's close rather than live. Every card shows how old its data is. The cards themselves are static files. Headlines in the full record are fetched when you open it, through this site's own server, and placing orders goes through it to your brokerage. Your cart lives in your browser; if you sign in, it is also stored with your account, notes included, so it can follow you between devices and the nightly job can email you before anything in it reports.

What it cannot know

This is the part worth reading twice. A screen of this kind is blind in specific, knowable ways, and it is more useful to name them than to imply they don't exist.

Sectors are broad. The sector half of each factor compares within eleven sectors, and a sector is a wide net: a railroad and an airline are both Industrials. Comparisons inside a sector are fairer than across the market, not perfect.
It reads the last filing, not the business. A pending acquisition, a patent cliff, a regulatory inquiry, a founder leaving, a product nobody wants yet: none of that is in the numbers until it lands in a financial statement, by which point the price has usually moved.
Accounting is a choice. Two honest companies can report the same economics differently. Write-downs, one-off gains, capitalised costs and unusual fiscal years all distort ratios the score treats as comparable.
Missing data is only partly penalised. When a figure is absent it is dropped rather than guessed, and a factor resting on one or two figures is pulled toward the middle, but a company with thin disclosure can still score on partial evidence. Banks and insurers are the usual case.
Momentum is the past tense. It records what the price has already done. It carries no information about what happens next, and in a downturn it is actively misleading.
Nothing here has been backtested. The weights and thresholds are reasoned, not fitted. No claim is made, or should be inferred, that a high score has ever preceded a good return.
Parsing fails sometimes. Every company's filings are checked each night and new ones are read automatically as they appear. When a figure looks impossible, it probably is. Each card links to the filing it came from, and that filing is the authority, not this page.

What this is not

This is not investment advice, and it is not a recommendation to buy or sell anything. It is a reading of public filings, presented so you can disagree with it.

The pros and cons on each card are plain thresholds, and each one names the figure that triggered it, so you can decide the rule is wrong and still keep the fact underneath. The thresholds for a worry are set as close to the thresholds for a strength as the figures allow, figures in the bottom sixth of the sector count against, and any factor the score marks down is always explained on the “against” side. Every company also gets its weakest point named, even when nothing crosses a line. A short list of cons is not a clean bill of health: the risks that matter most are often in the 10-K rather than in the numbers.

That is the intended use: the score gets you to a shortlist, and then you read the 10-K.

Back to the deck