When Two Filings Disagree on the Same Number, Which One Is Right?

A company’s FY2023 annual report shows revenue of 412 million. The FY2024 report, published a year later, shows FY2023 as the comparative column, and it says 407 million. Nobody flags this. Both numbers sit quietly in their respective filings, and unless someone happens to pull both PDFs and compare them line by line, the discrepancy just never surfaces. That gap between “the number as originally reported” and “the number as later restated” is what cross-filing corroboration is built to close.

Two independent looks at the same period

Every fiscal period, in principle, gets reported twice: once as the current year in its own filing, and once as the prior-year comparative in the following year’s filing. Those are two independent statements about the same set of facts, produced at different times, sometimes by different preparers. When they agree, that agreement is worth something: it’s a second, separate confirmation of a figure that didn’t come from re-reading the same PDF twice. When they don’t agree, that disagreement is worth something too, and today it mostly goes unnoticed.

Rexfin’s design for this check is to match each period’s as-originally-reported value against the corresponding prior-year comparative in the next filing, within a tolerance that scales with the statement’s rounding unit, because two filings both rounding to the nearest thousand aren’t disagreeing by a few dollars, they’re just rounding.

A match is stamped, a mismatch is never dropped

When the two independent reads land on the same value within tolerance, the underlying atom gets stamped cross-filing-corroborated: a second, independent source lining up behind a figure that already passed multi-backend extraction agreement within its own filing. That’s not the same kind of check twice; it’s an entirely different kind of confirmation, because it’s comparing two separate documents rather than two reads of one page.

When the two don’t match, the design principle is the one that runs through every layer of Rexfin’s trust chain: a mismatch is raised as a typed, visible discrepancy, never silently dropped and never silently averaged away. Both values are kept. The delta is kept. The citation for each (which document, which page) is kept, so a person looking at the discrepancy can go check both source pages directly rather than trust a summary of the disagreement.

Extraction error or a real restatement: a real distinction, not a shrug

Not every mismatch means the same thing, and treating them the same would be its own kind of dishonesty. A mismatch can mean an extraction slipped somewhere upstream (an OCR misread, a column mis-attributed on a scanned page). Or it can mean the company genuinely restated the prior period under IAS 8, because they found and corrected an error, or reclassified something, or changed a policy that required retrospective adjustment. Those two explanations point a finance team toward completely different next steps: one is “check the extraction,” the other is “understand why the company restated and whether it changes your read of prior guidance.” Classifying which is which, with both values and the delta laid out, rather than picking a winner silently, is the point of this check.

What this depends on, and why it comes later in the pipeline

This check only works once the groundwork underneath it is solid: it needs a stable document identity for each filing, atoms that have already passed within-filing verification, and the same figures pinned to the same canonical concept across both filings so “revenue” in one is actually being compared to “revenue” in the other, not to a differently labeled line that happens to sit near it. That’s why cross-filing corroboration sits downstream of intake and tagging rather than running standalone: comparing two filings only means something once each one, individually, is already trustworthy.

This is part of Rexfin’s ingest roadmap, building directly on the trust-chain infrastructure already shipping elsewhere in the product: the same atom-level verification and citation discipline, extended across filings instead of within one. It’s the kind of check that matters most for companies with a restatement history, or for anyone doing multi-year trend work who needs to know whether a jump between two filings’ comparative figures is a real correction or a data problem.

See how corroboration fits alongside the rest of the trust layer at the product tour hub, or book a demo to see how a cited discrepancy is presented when two filings disagree.