EPS and Per-Share Metrics You Can Trace Back to the Filing

Earnings per share sounds like the simplest number in a filing. Two lines, basic and diluted, printed right on the face of the income statement. In practice it is one of the easier numbers to get quietly wrong, because it looks like every other figure on the page and it is not. EPS is a per-share decimal, not a monetary amount, and if your extraction pipeline treats it like one, you get a value scaled by a thousand or a thousandth for no reason anyone would notice at a glance.

Rexfin treats EPS as its own thing, extracts it as a citable atom, and checks it against the filing’s own math before it ever shows up on a page.

Why finance teams care about a number this small

Because it is not small to the people who read it. EPS drives valuation multiples, feeds into covenant language, and is one of the first things an analyst or a board member checks against consensus. If your reporting layer gets a per-share figure wrong, or worse, silently right by coincidence, nobody notices until someone reconciles it against the actual filing and finds a mismatch. At that point the conversation is no longer about EPS. It is about whether anything else on the page can be trusted.

The honest failure mode with a lot of extraction tooling is unit confusion. A parser tuned for monetary cells sees “0.356” and looks for a scale to apply, thousands, millions, because that is what it was built to do for every other line item. EPS is never scaled that way. It is the printed decimal, in whole currency units, full stop. Getting that wrong doesn’t produce a wildly implausible number. It produces a plausible-looking wrong one, which is the worst kind.

How it works

Rexfin routes per-share values through a parser dedicated to that shape of number, kept separate from the one used for money cells, so a monetary figure can never be misread as an EPS and a per-share figure can never get treated as one needing unit-scaling. That separation is structural, not a rule someone has to remember to apply.

Once basic and diluted EPS are extracted, Rexfin doesn’t just take the printed number on faith. It looks for the filing’s own EPS note, the tabular breakdown that shows profit attributable to owners divided by weighted-average shares outstanding, and checks that the printed EPS is consistent with those two operands. This is a corroborating check, not a veto: weighted-average share counts involve issuer-specific adjustments that are genuinely hard to reproduce exactly, so the tie is accepted within a tight tolerance rather than demanded to the decimal. When a filing states basic and diluted EPS are the same figure, both are recorded as such. When continuing and discontinued operations are broken out separately, those rows are kept out of the total EPS figure so they cannot get blended into it by accident.

Loss periods matter too. A negative EPS is usually printed in parentheses, and a sign-handling bug in that spot silently flips a loss into a profit. Rexfin’s extraction treats that as a real risk to guard against, not an edge case to hope doesn’t come up.

And when the source document is a scanned image rather than machine-readable text, Rexfin does not pretend it has a verified page reference it does not have. It still computes what it can, but it will not label a figure as citable unless it can actually point you to the page it came from. That is the same fail-closed posture that runs through how Rexfin cites financial statements generally: a number without a provable source is marked as such, not dressed up to look verified.

Who this is for

Anyone building a view of a company that includes per-share figures, whether that is a comp set, a quarterly tracker, or a board pack, benefits from EPS that ties back to the actual filing rather than a scraped table nobody re-checked. It matters most for teams who get asked to defend a number on the spot: an analyst fielding a question about diluted EPS trend, a controller reconciling external reporting against internal figures, anyone who has been burned once by a per-share figure that turned out to be off by three orders of magnitude.

If you want to see how a claimed number gets weighed against competing sources before it is shown at all, that logic lives in trust-chain evidence ranking. And if a later filing restates a prior period’s earnings, the same reconciliation discipline is what catches it, covered in restatement detection. EPS is a small line on the page. It gets the same scrutiny as the large ones.