Methodology

Six mistakes we made reading company filings, and the check each one taught us

Chevron's debt read as $0.4bn, SoFi's revenue as fee income, a dividend counted five times in a year. What went wrong on our own pages and the check that now stops it.

Every comparison on TickerVerdict is built from what companies file with the U.S. Securities and Exchange Commission. That sounds like the end of the argument: the figures come from the company itself, so they must be right. They are not, or not automatically. A filing is right; the step that reads a number out of it, and the step that decides what that number means, are where things go wrong. We know because we have gone wrong at both, in public, and logged each case on our corrections page.

This guide is the short version of that record. Each section is one mistake that reached a published page, what it looked like, why it happened, and the check we now make before a figure is printed. The checks are not specific to this site. If you compare companies yourself, from a data provider or a spreadsheet, the same six traps are waiting.

1. A number that did not match the balance sheet

Chevron's total debt was shown as $0.4bn. For one of the largest oil companies in the world, that should have stopped us on sight; it did not, because nothing in the pipeline compared it with anything. The cause was mundane. Chevron's quarterly reports tag long-term debt under a label our reader did not recognise, so it found only a small piece of the debt and reported that as the whole. The balance sheet at 30 June 2026 showed $37.1bn. The same page also gave cash as $9.6bn, a figure that included restricted cash; the balance-sheet line for cash is $8.5bn.

The check: total debt is now built from every debt line on the balance sheet — long-term debt, its current portion, commercial paper, short-term borrowings — taking care not to count a line twice where one already contains another. Where a company's cash figure includes restricted cash, the restricted part is taken back out. And a figure that is a small fraction of what a company's size suggests is a reason to open the filing, not to publish. The two pages that carried Chevron's figures were taken down, rebuilt and checked against the 10-Q before they came back.

2. Revenue and profit drawn from different parts of the company

General Motors was shown on its automotive sales alone, $168.35bn, while its operating profit included its finance arm. Dividing a profit that counts the lending business by a revenue that leaves it out makes the margin look better than it is. On consolidated revenue of $186.95bn, the figure that matches the profit, the margins came down and one of our comparisons changed its result.

SoFi went wrong the opposite way. Its revenue was shown as $0.61bn, which was its fee income alone; its total net revenue is $4.3bn. That one error turned a price-to-sales ratio of 4.27 times into 30.1, and made a lender look as expensive as a fast-growing software company.

The check: every ratio must have a top and a bottom drawn from the same perimeter. When a company reports its revenue in parts and also files the consolidated total, and the total is exactly the parts added together, the total is used. For a lender, the reader looks first for total net revenue — net interest income plus everything else — before any narrower revenue line.

3. Counting the wrong number of payments

Coca-Cola's payout ratio was stored as 80.07%. The company pays its dividend four times a year, but it pays on 1 April, and its twelve-month window that quarter ran to 3 April, so the window caught five payments. On four, the ratio is 64.78%. General Motors' payout ratio had a separate fault: it was given as 29.9%, where its dividend per share and earnings per share give 15.1%.

The check: the payout ratio is now computed two ways, from the cash paid out as dividends and from dividend per share over earnings per share, and the two are compared. For almost every company they agree within a few percent. When they differ by more than a fifth, the cash figure contains something other than one year of ordinary dividends — an extra payment caught by the window, or, at GM, dividends paid to minority holders of its subsidiaries — and the per-share figure is used. Neither ratio feeds the score, so no verdict moved, but a page that said one dividend had more earnings cover than another rested on the wrong figure, and that sentence was rewritten.

4. A measure that means something different for this business

UnitedHealth was shown with a gross margin of 88.6%. The arithmetic was correct and the figure was meaningless: an insurer's main cost is the medical claims it pays, and those do not sit in cost of sales, so a gross margin leaves them out. The same is true of lenders, whose main cost is interest. We no longer print a gross margin for insurers or lenders at all, and the quality factor is decided on the measures that do mean something for them.

The check: before a ratio is shown, we ask whether its definition fits the business model, not only whether it can be calculated. The full list of figures we withhold, and for which companies, is in the numbers we refuse to print. Visa is the extreme case: its dividend yield came out at 0.17% because of the way it tags dividends in its filings, and its share count and earnings per share could not be read reliably either, so its comparisons were withdrawn rather than patched.

5. Treating a gap as a draw

When a factor could not be scored because a figure was missing, our pages called it a tie, or "too close to call", and some even gave a reason for a contest that was never measured. That is worse than a blank: it tells the reader two companies were compared and found level when nothing was compared at all. Every page was rewritten so that an unscored factor is called that, and the missing figure is named.

That fix exposed a second problem. Under the original rule, a single missing figure left a whole factor unscored, which on 10 November applied to 101 of the 264 factor results across our 44 comparisons. The rule now scores a factor on the figures both companies have, provided they carry at least half of what it measures, and the page says when it did. On the refreshed figures 57 results remain unscored, most of them momentum, for which we hold no price history. How the six-factor score works gives the rule in full.

The check: a missing value is carried through as missing, everywhere. The written analysis, the table and the score must all show the same gap, and the gate that builds the site refuses a page whose prose calls an unscored factor close.

6. A page that disagreed with itself

Our written analysis leaves out ratios that would mislead for a particular company and says so. For a while the table on the same page did not apply that rule, so a page could state that no trailing P/E was shown for a company and then print it a few lines lower. Separately, every comparison was titled as if it covered share-price returns, which we do not hold.

The check: the prose and the tables now read the same record and the same list of withheld figures, so the table hides exactly what the text withholds. Before the site is built, every figure quoted in the text is looked up in that page's own data, and a page that quotes one it does not contain stops the build. Titles describe what the page covers: valuation, growth and margins, from filings.

What this means when you read any comparison

None of these mistakes came from bad data in the filings. They came from reading the filings carelessly or interpreting them without regard to the business. Those are the two weak points in every comparison site and every spreadsheet, ours included, which is why we publish the record rather than hope nobody notices. When you look at a figure anywhere, the same questions do most of the work:

  • Does it match the balance sheet or the income statement a reader can open?
  • Are the top and the bottom of the ratio drawn from the same company, the same perimeter and the same period?
  • If it is a twelve-month total, how many payments or quarters does it contain?
  • Does the measure mean the same thing for this kind of business?
  • Is a blank shown as a blank, or quietly turned into a number or a draw?

If you find a figure on this site that fails one of them, tell us: it will be checked against the filing, corrected if it is wrong, and logged with the date, what the page said and what it says now. To see the figures side by side for any two of the companies we cover, use compare any two companies.

Comparisons that use these companies

Related guides

TickerVerdict provides factual data comparisons for informational purposes only. Nothing here is investment advice or a recommendation to buy or sell any security. Figures may be delayed; verify with your broker before investing. Capital is at risk.