How should a stock screen handle amended filings? | Quantery Blog

October 30, 2026 · 7 min readstock-screeningsec-filingsbacktesting

An amended filing is a new public event, but Quantery doesn't store it as a second fundamentals vintage today. The engine keeps the first-reported value for each fact and leaves the existing company-period row unchanged. That protects a historical test from being rewritten. Comparing the original with an amendment is a manual research step outside the current screen engine.

The form name alone doesn't settle what to compare. A 10-K/A can correct financial statements, add a missing exhibit, or supply governance disclosures that were left for a later filing. Read the explanatory note first. Then compare the fields your thesis uses. That's the shortest route from a scary form suffix to a defensible research decision.

A 10-K/A doesn't automatically mean restated earnings

The /A means amendment. It doesn't tell you why the company amended the report.

Tesla's annual filing history gives a clean example. The company filed its annual report for the fiscal year ended December 31, 2025 on January 29, 2026, then filed a 10-K/A on April 30, 2026. The amendment's explanatory note says it was filed to add the Part III material covering directors, executive compensation, ownership, related-party matters, and auditor fees. It also says the amendment didn't revise the financial statements or otherwise update the original disclosures beyond the specified changes (Tesla 2025 Form 10-K/A).

That distinction matters to a quantitative thesis. If your rules use revenue, debt, or operating cash flow, this amendment changes none of those inputs. If you're researching executive pay or beneficial ownership, it contains information that wasn't in the original annual report. Same form suffix, different consequence.

Start every amendment review with three questions:

What does the explanatory note say changed? Filers usually put the purpose near the front.
Do the changed sections feed this thesis? A governance addition doesn't alter a free-cash-flow rule.
When did the new information become public? EDGAR's filing-detail record shows the filing date and acceptance time. The SEC says its submissions data is updated throughout the day as filings arrive (SEC EDGAR API documentation).

Don't infer a restatement from the form code. Don't infer that an amendment is harmless either. Open it.

Why replacing the original creates look-ahead bias

Suppose a company files an annual report, your screen qualifies it, and the company corrects a relevant number later. If a backtest replaces the original fact everywhere in history, the simulated screen gets knowledge that investors didn't have on the original decision date.

That's look-ahead bias: allowing future information into an earlier decision. It can work in either direction. A later correction might remove an apparent bargain, or it might make a weak company look stronger. The direction isn't the core defect. The timestamp is.

Quantery takes a conservative approach to that sequence today:

Before the original filing, the facts aren't available.
After the original filing, the engine uses the first-reported values.
A later amendment doesn't overwrite those values or add a second fundamentals vintage.
To study the correction, compare the original and amended documents outside the engine.

This is one reason point-in-time fundamentals matter. A conventional database often gives you one polished value for a fiscal period. That value may be correct today and wrong for a simulation of what was knowable then. Quantery asks a narrower question: what value was reported first? The filing-date discipline behind a sound backtest explains why the original date matters.

This policy keeps an amendment from changing an old run underneath you. It also has a limit: the current lake can't run the same thesis on an original vintage and an amended vintage as two separate in-app states. A researcher who needs that comparison has to extract the affected inputs from both filings and calculate the rule manually.

Materiality isn't a percentage shortcut

When an amendment does correct the accounts, the next temptation is to ask whether the change is "big enough" and reach for one percentage cutoff. The SEC staff has rejected that shortcut. Staff Accounting Bulletin No. 99 says a numerical threshold can be an initial step, but exclusive reliance on a percentage has no basis in the accounting literature or the law (SEC Staff Accounting Bulletin No. 99).

A small error can matter because of what it changes. The same bulletin lists cases where a misstatement masks an earnings trend, turns a loss into income, affects covenant compliance, or changes management compensation. None of those judgments can be reduced to one universal ratio.

For a screen, "material" has a narrower operational meaning too: could the correction change the rule's output? A revision can be small relative to total assets and still move a company across your debt gate. Another can be large in a line your thesis never reads and leave the score untouched.

So compare the amendment against the actual machinery:

Did a required field move from missing to present?
Did a denominator cross zero?
Did a value cross a gate or scoring boundary?
Did the composite score or verdict change?
Did the amendment affect several related inputs from the same statement?

This boundary check is more useful than arguing about a generic materiality threshold. Your thesis already states which differences matter to its output. Use it.

How to build amendment-resistant rules

A screen can't read an explanatory note through arithmetic. It can, however, avoid making fragile decisions when a later filing changes one input.

The first defense is to guard denominators and missing values. If an amendment supplies a field that was absent before, keep the earlier version null. Don't fill it backward. A null is a fact about what the machine could support on that date. Treating it as zero invents a different fact.

The second defense is to score ordinary variation and reserve hard gates for real dealbreakers. If a tiny revision can flip the entire result, ask whether the cutoff expresses an economic belief or just a convenient line. Strict gates and score gates behave differently, and amendment sensitivity is one more reason to choose deliberately.

The shape below is illustrative, and the bundled templates are the reference for exact fields. The thresholds are proposed research parameters, not claims about any company:

params:
  debt_assets_max: 0.50
  debt_assets_watch: 0.40
  margin_floor: 0.05

quality:
  required:
    assets_now: no_assets
    debt_now: no_debt

features:
  assets_now: newest(total_assets)
  debt_now: newest(total_debt)
  debt_assets: if(assets_now > 0, debt_now / assets_now, null)
  operating_margin: if(ttm(revenue) > 0,
    ttm(operating_income) / ttm(revenue), null)

criteria:
  leverage:
    rules:
      - { when: "is_null(debt_assets)", score: 0, flag: no_ratio }
      - { when: "debt_assets <= $debt_assets_watch", score: 2 }
      - { when: "debt_assets <= $debt_assets_max", score: 1 }
      - { else: 0 }
  profitability:
    rules:
      - { when: "operating_margin >= $margin_floor", score: 2 }
      - { else: 0 }

gate:
  mode: strict

newest() is doing important work here. It reads both balance-sheet values from the newest period, so the ratio doesn't splice debt from one filing into assets from an older one. If an amended document changes only one field, the source review still matters, but the formula won't fabricate a mixed-period balance sheet.

After an amendment arrives, copy the affected original and amended inputs into a small worksheet and apply the same formulas to each set. Quantery won't perform this two-vintage comparison for the same company-period today. Don't change the thesis at the same time. One moving part is enough. If the result changes, inspect the exact boundary crossed. If it doesn't, record that too. A correction that leaves the decision unchanged is useful evidence about the rule's stability.

What an amendment test can establish

A manual amendment comparison can tell you whether your result depended on data that the filer later changed. Repeating that work across historical amendments can show whether a rule is unusually exposed to fragile fields or knife-edge thresholds.

It can't tell you that the latest filing is the final economic truth. Companies can amend again. Accounting judgments remain judgments. A corrected figure can also preserve a poor proxy: precise reported debt doesn't make debt-to-assets the right measure for every industry.

A backtest adds another limitation. Price depth depends on the user's market-data plan, and any backtested return is hypothetical and excludes costs. More important here, Quantery's first-reported policy preserves the original backtest input but doesn't preserve an amendment as a second runnable vintage. The app can protect the old result. It can't grade the correction against that result for you.

The practical standard is less dramatic than "ignore amended companies" or "always trust the latest filing." Read what changed. Keep both source documents in your research record. Recompute the affected rule manually, then decide whether the thesis survived the difference.

That turns an amendment into a research event without pretending the current engine stores something it doesn't. If one changed input breaks the thesis, you now know exactly which assumption to revisit.

Research tooling, not investment advice. Nothing here is a recommendation to buy, sell, or hold any security. Screens, scores, and backtests are informational only; backtested results are hypothetical, exclude costs such as commissions and slippage, and do not guarantee future results. Verify against primary filings and make your own decisions.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles