As Filed vs As Restated: How to Track Restatements and Reclassifications

To track restatements and reclassifications, keep every version of every figure with the date it was filed, then compare each period across the filings that report it. Kepler does this for every company it covers: it reads every filing, keeps each number as first reported, as last reported and as it stood on any past date, and shows which filing changed it and how, with each version cited to its line. We build Kepler. This guide defines the ways a reported number changes after the fact, explains why the difference matters for models, backtests and comps, and gives a step-by-step method for finding each kind in the filings.

Seven ways a reported number changes later

Type	What happened	An error?	Where it shows up	Typical signal
Big R restatement	Material error in prior statements	Yes	8-K Item 4.02, then a 10-K/A or 10-Q/A or the next filing	“Restated” columns, auditor’s explanatory paragraph
Little r restatement	Error immaterial to prior periods, corrected in comparatives	Yes	The next 10-K or 10-Q	“As previously reported / adjustment / as revised” table
Out-of-period adjustment	Small prior error corrected in the current period	Yes	The current period’s results	A note; prior periods unchanged
Reclassification	Presentation changed	No	The next filing’s comparatives	“Reclassified to conform to the current period presentation”
Segment recast	Reportable segments changed	No	The next filing, sometimes an 8-K first	Prior periods re-presented by new segment
Discontinued operations	A business sold or held for sale	No	Every comparative period	One line below continuing operations
Retrospective accounting change	A new principle applied to prior years, or a stock split	No	The adopting filing	“As adjusted”; per-share figures restated

In conversation, “restatement” often covers all seven. Accounting draws the line at error: only the first three correct a mistake. Kepler labels each change by type, so a model can treat an error correction differently from a new presentation.

Big R and little r restatements

The terms are informal, and the SEC uses them. In a March 2022 statement, the SEC’s then Acting Chief Accountant described a reissuance restatement as a “Big R” restatement and a revision restatement as a “little r” one, and noted that little r restatements rose to nearly 76% of all restatements in 2020, from about 35% in 2005. Both correct errors in previously issued statements.

Big R. When a company concludes that previously issued statements should no longer be relied upon because of an error, it files an 8-K under Item 4.02 within four business days, giving the date of the conclusion, the periods affected, a brief description and whether the audit committee discussed it with the auditor (Form 8-K instructions). If the auditor prompted it, the auditor’s letter follows as an exhibit. Restated statements arrive in an amended filing or the next periodic report, labeled “restated,” and the auditor’s report carries an explanatory paragraph.

Little r. The error wasn’t material to the prior statements, so they aren’t reissued. The next filing revises the comparative columns and discloses the correction, usually in a table with “as previously reported,” “adjustment” and “as revised” columns. There’s no Item 4.02 and no amendment, which is why little r corrections are easy to miss. Kepler reads every revision table at ingest and records the original figure, the adjustment and the revised figure as separate, cited facts.

The cover-page check boxes. Since the SEC’s 2022 clawback rule, the 10-K cover asks whether the financial statements in the filing “reflect the correction of an error to previously issued financial statements,” and whether any of those corrections required a recovery analysis of executive incentive pay (Form 10-K). SEC staff guidance from April 2025 says the first box covers both Big R and little r corrections (Cooley’s summary). Both answers are tagged in XBRL as dei:DocumentFinStmtErrorCorrectionFlag and dei:DocumentFinStmtRestatementRecoveryAnalysisFlag, which makes them screenable. In Kepler, “every company in our coverage that checked the error-correction box this year” is one question; our guide to screening companies with AI covers screens like it.

Reclassifications, recasts and discontinued operations

These aren’t errors, and they change far more history than restatements do.

Reclassifications. Companies rename, merge, split and move lines from year to year. Apple’s fiscal 2021 annual report shows $880 million of proceeds from issuing common stock in fiscal 2020. Its fiscal 2022 report shows fiscal 2020 again, and the line is gone: it went into “Other,” which became $754 million, the old $880 million plus the old ($126) million. Both reports were right when filed. The only notice is the boilerplate sentence in Note 1. Our paper Reclassified to Conform shows how common this is, and our XBRL guide explains why the tags lose the thread.

Segment recasts. When a company changes its reportable segments under ASC 280, it re-presents prior periods on the new basis. The recast may arrive first in an 8-K, ahead of a securities offering, or simply in the next 10-Q or 10-K. Segment margins in your model from before the change no longer line up with the company’s current figures. Kepler keeps both segment histories, the old basis as filed and the new basis as recast, and marks the boundary.

Discontinued operations. When a business is sold or held for sale and meets the criteria in ASC 205-20, its results move to a single line below continuing operations for every period presented. Revenue and operating income for prior years drop, with no error and no change to net income.

Retrospective changes. A new accounting principle adopted retrospectively, or a stock split, changes prior-year figures in the adopting filing. Per-share data and share counts are restated for splits.

Why the difference matters

• Models. A ten-year history that takes some years from the original filings and others from restated comparatives mixes two bases. Growth rates computed across a recast compare different businesses.
• Backtests and point-in-time research. A signal tested on today’s restated history uses information that wasn’t public on the test date. Point-in-time data keeps what was known when.
• Comps. One peer on a recast basis and another on as-filed figures don’t compare.
• Credit and compensation. Covenant tests and incentive pay were set on the figures reported at the time. Under Rule 10D-1, a restatement can trigger recovery of incentive pay.
• Diligence. A pattern of little r corrections or reclassified lines tells you something about the finance function.

Kepler gives each of these readers the version it needs: point-in-time figures for the backtest, one consistent basis for the model and the comps sheet, and the as-filed numbers a covenant test used.

How to find them: a method

1. Screen for Item 4.02. Use EDGAR full-text search, which covers filings since 2001, for 8-Ks mentioning Item 4.02 in your universe.
2. Read the cover-page flags. Check both boxes on every 10-K, or query the two dei flags across a coverage list.
3. Look for amendments. Note every 10-K/A and 10-Q/A. Many 10-K/As only add Part III information from the proxy, so open each one.
4. Search the language. “Restated,” “as previously reported,” “as revised,” “recast,” “reclassified to conform,” “discontinued operations,” “retrospectively.”
5. Compare overlapping years. Each 10-K shows two or three years. Put the same year from consecutive 10-Ks side by side, line by line, and explain every difference with arithmetic: a line that vanished usually went into a neighbor or a total.
6. Read Note 1, the segment note and the discontinued operations note for what changed and why.
7. Check the auditor’s report for an explanatory paragraph about restated statements.

Doing step 5 by hand for one company across fifteen years takes days. Kepler runs it in code for every company it covers.

Why data vendors and AI tools flatten them

A database that stores one number per company per period has to choose which version to keep, and the simplest choice is the latest. The SEC’s own frames API keeps the last-filed value for each period, as our XBRL guide explains. Standardization adds another layer: a 2023 study in the Journal of Accounting and Economics by Du, Huddart and Jiang, Lost in standardization, found discrepancies between as-filed and Compustat data, “potentially a result of Compustat’s standardizations” and more pronounced for firms with complex reporting, and that FactSet data “also exhibit significant and often larger discrepancies from as-filed data.”

General-purpose AI adds a third: a model answers from whichever filing its search retrieved, so fiscal 2023 revenue may come from the 2023 10-K or the 2025 10-K, and the answer won’t say which. Ask any tool for one restated year both ways; Kepler returns both figures, each with the filing it came from. The checks in how to check AI-generated numbers and how to analyze a 10-K with AI cover the rest.

How Kepler does this

Every version of every figure, dated. Kepler reads every filing for 14,000+ companies, 26M+ SEC filings in all, and keeps each figure as every filing stated it, with the filing date. Any statement, line or cell is available as first reported, as last reported, or as it stood on any date you choose. A backtest gets point-in-time inputs; a model gets one consistent basis; a comps sheet gets every peer on the same one.

Changes explained by arithmetic, in code. Statements assemble from each filing’s own XBRL with no model in the path, and histories follow lines across renames, merges, splits and retags using the method in our Reclassified to Conform paper, which recovers the change from the periods consecutive filings repeat. Each company gets a changelog of what moved, boundary by boundary. Apple’s reads: “absorbed proceeds from issuance of common stock (880 for 2020) into other financing (754).” Where a line’s definition changes, the history marks the break.

Errors flagged as they’re filed. Extraction models we train read every 8-K, 10-K/A and revision table at ingest. Agents watch your coverage for Item 4.02 filings, cover-page error flags, revision tables and segment recasts, and alert you with the exact sentence and the cells in your model it touches. Your restatement review is saved as a playbook, and every alert and change is logged.

On your models, in your firm’s terms. Our engineers build your firm’s ontology, so restatements map onto your own models and KPI definitions, and the ontology knows time: what was true as filed, as restated, or on any date. Click any figure to see every version and the filing behind each. The same data reaches Excel, Claude, ChatGPT and your own systems through our API. Our engineers built this layer before at Palantir.

See any company’s statements as filed and as restated on app.kepler.ai, or book a demo to run the comparison on your coverage.

Frequently asked questions

What is the difference between as filed and as restated financials?

As filed, or as originally reported, is the figure in the filing that first reported a period; as restated is the same period's figure in a later filing after an error correction, reclassification, segment recast or discontinued operation. Kepler keeps every version with its filing date, so any figure is available as filed, as restated, or as it stood on any past date. Both are correct for their date.

What is the difference between a restatement and a reclassification?

A restatement corrects an error in previously issued financial statements. A reclassification changes how amounts are presented, such as moving a line into another or splitting one in two, without any error. Reclassifications usually leave totals like net income unchanged and are disclosed with a single sentence that prior amounts were reclassified to conform to the current presentation.

What is a Big R versus a little r restatement?

A Big R, or reissuance, restatement corrects an error material to previously issued statements: the company files an 8-K under Item 4.02 saying they can no longer be relied upon, then reissues them. A little r, or revision, restatement corrects an error that wasn't material to the prior statements by revising the comparative figures in the next filing, with no Item 4.02 and no amended filing.

How do I find restatements in a 10-K?

Check the two boxes on the 10-K cover page that flag error corrections and clawback analyses, then search the filing for 'restated', 'as previously reported', 'as revised', 'recast' and 'reclassified to conform'. Read Note 1 and the segment note, look for an explanatory paragraph in the auditor's report, and compare each prior-year column against the 10-K that first reported it.

What is an 8-K Item 4.02?

Item 4.02 of Form 8-K, Non-Reliance on Previously Issued Financial Statements or a Related Audit Report or Completed Interim Review, is filed within four business days when a company concludes, or its auditor tells it, that prior statements should no longer be relied upon because of an error. It names the periods affected and briefly describes the facts.

Why does point-in-time data matter for backtests?

A backtest should only use what investors could have known on each date. Restated figures weren't public until the restatement was filed, so a history built from today's values leaks future information into past decisions. Point-in-time data keeps each figure as it was originally reported, with the date each later version appeared.