What is the right way to compare two SEC filings?
Compare two SEC filings in seven passes: verify the filing pair; isolate the same section; anchor on reported facts; use text differences to locate passages; classify the source role; test the change against a written thesis; and record the next evidence needed. Start with form type, issuer, report period, filing date, accession number, and amendment status. A comparison between unlike forms, periods, or section boundaries can produce a precise-looking but meaningless result.
Do not begin with an automated materiality score. Word-count changes, sentence additions, keyword counts, and text-similarity percentages can identify where wording moved. They cannot establish why it moved, whether the underlying business changed, whether management’s explanation is complete, or what the change means for valuation. Open both source filings and read the surrounding definitions, tables, footnotes, and cross-references.
The useful output is a bounded research record: the exact changed statement, the reported number or disclosure that supports it, the claim it affects, the strongest alternative explanation, and the next filing, note, event, or operating metric that could resolve the uncertainty. Filing comparison is therefore a change-detection process, not a stock-ranking system or a substitute for judgment.
Know what each filing section can and cannot establish
A Form 10-K provides the annual description of the business, material risk factors, management’s discussion, audited financial statements, controls, and other required disclosures. A Form 10-Q updates the first three fiscal quarters with interim statements and defined sections such as Item 2 MD&A and Part II Item 1A Risk Factors. A Form 8-K reports specified events on a more current schedule. The forms overlap, but they are not interchangeable.
Financial statements establish reported amounts under the company’s accounting policies. Notes define and disaggregate those amounts. MD&A presents management’s view of results, liquidity, trends, and uncertainties. Risk factors describe risks the company considers significant under the disclosure framework. A filing can contain all four without proving an investor’s causal interpretation. The comparison should preserve the role of each source instead of blending them into one narrative.
A statement value, period, unit, concept, and filing accession. It establishes what was reported under the disclosed accounting basis.
An attributed explanation, trend, expectation, or uncertainty. It records management’s account, not independent verification.
A deterministic comparison such as growth, margin change, sentence difference, or text similarity with a visible method.
A written judgment about why the evidence matters, what could challenge it, and which next source should be reviewed.
Match the filing pair to the question before comparing text
For a quarterly operating question, compare consecutive 10-Q filings and keep the latest 10-K as annual context. For an annual business-model or risk review, compare consecutive 10-K filings. For a specified corporate event, inspect the relevant 8-K and then identify which later 10-Q or 10-K incorporated its financial effects. Comparing a 10-Q with a 10-K can be useful for context, but section lengths and disclosure requirements differ; do not interpret that structural difference as a change in importance.
Record the report date separately from the filing date. The report date defines the period or balance-sheet date covered; the filing date tells you when the document became public. A later filing may contain information about an earlier period, and an amendment may correct or supplement the original. Preserve the accession number because a ticker, form label, and date alone may not identify the exact document.
Confirm whether the company incorporates prior disclosure by reference or states that no material changes occurred. A short quarterly risk section may point back to the annual baseline. If the next quarter reproduces the full section, the word-count increase can be enormous even when only part of the underlying risk story changed. That is a structural transition, not an automatic risk alarm.
Move from document change to evidence without skipping the hard part
Verify identity, form, period, accession, and amendment state
Match the legal issuer, CIK, ticker used by your workspace, form type, report period, filing date, and accession number. Check whether either document is an amendment. If the issuer changed its fiscal calendar, segments, accounting policy, or presentation, note that before comparing sections.
This first pass prevents a common failure: interpreting a parser or navigation mismatch as a company change. If the intended section is missing, implausibly short, or bounded by the wrong heading, stop. A missing comparison is safer than a confident result built from unrelated filing text.
Isolate the same section and preserve its context
Compare Item 2 with Item 2, or Item 1A with Item 1A. Keep the section heading, start and end boundary, tables, footnotes, and references visible in the original filing. Normalized text can support exact matching, but it should not replace the document layout that tells a reader what a number or sentence modifies.
StockEyes uses bounded extraction and fails publication when the target section, official host, filing order, or plausible size contract does not hold. That design matters more than an elaborate score. A diff engine that silently includes the wrong section can produce hundreds of persuasive but irrelevant changes.
Anchor on reported results before reading management’s explanation
Capture revenue, margins, cash flow, segment results, balance-sheet changes, share count, or the other reported facts relevant to your question. Use comparable periods and the same accession where possible. Label calculated growth or margin changes as calculations, not reported values.
Management may explain that revenue changed because of price, volume, mix, currency, supply, demand, or a segment. The financial statement can establish the consolidated amount; it cannot automatically establish the driver. Begin with the amount so the prose is tested against a known baseline rather than used to create one.
Use document metrics to locate passages, never to rank importance
Exact sentence additions identify normalized sentences not found in the prior section. Removed sentences identify prior wording not found exactly now. Five-word shingle similarity estimates repeated phrasing. Fixed keyword families route attention toward topics such as revenue, margins, supply, regulation, cybersecurity, customers, or capital.
Every metric has false positives. A date, table label, reordered paragraph, punctuation change, revised definition, or new cross-reference can alter the output. A low similarity result can reflect a rewritten section; a high similarity result can still conceal one decisive sentence. Inspect the largest apparent changes, then read before and after each passage.
Classify the changed language by source role
Ask whether the passage is a reported fact, a management explanation, a risk disclosure, an accounting definition, a forward-looking statement, a legal update, or repeated boilerplate. The same words carry different evidentiary weight in different roles. A customer-demand statement in MD&A and a customer-concentration risk in Item 1A may address related topics without supporting the same conclusion.
Write one atomic observation. Include the exact source, section, period, and what changed. Avoid “management became more confident” unless you can point to language, define the comparison, and explain why the change is not caused by editing or context. Tone is an inference, not a reported field.
Test the passage against a written thesis and an alternative explanation
Name the thesis condition affected: customer adoption, pricing, product mix, operating leverage, liquidity, supply, regulation, competition, capital intensity, dilution, or another bounded mechanism. Then state whether the evidence supports, challenges, or leaves a question about that condition. Do not assign a stance when the connection is unclear.
Write at least one alternative explanation. More supply-chain language could reflect a worsening constraint, a more detailed disclosure, a new product transition, or a revised sourcing discussion. Margin improvement could reflect mix, pricing, cost absorption, a comparison base, or a one-time item. The alternative determines which next evidence matters.
Record the next proof and preserve the audit trail
Finish with a concrete source: a footnote, segment table, next 10-Q, regulator record, customer disclosure, capital-spending figure, share-count reconciliation, or future event. State what result would support and what result would challenge the current interpretation. “Research more” is not a task.
Save the exact filing URLs, accessions, comparison date, method, selected passage, reported baseline, interpretation, and unresolved question. A later reviewer should be able to reconstruct what was public at the time without trusting your memory or a regenerated summary.
What common filing-diff measures actually mean
current words − previous wordsDescribes section length under the extraction method. It is not business impact or disclosure quality.
current normalized sentences not exactly present beforeFinds exact-text differences. It does not distinguish a changed date from a new risk.
previous normalized sentences not exactly present nowLocates dropped or rewritten wording. A concept may remain under different language.
shared normalized five-word sequences ÷ combined unique sequencesEstimates repeated wording. It is not semantic agreement, truth, tone, or materiality.
fixed keyword-family matches by sectionRoutes a reader toward text. Counts overlap and cannot identify the meaning of the passage.
(current reported value − comparable value) ÷ |comparable value|Quantifies a financial difference only when definitions and periods are comparable.
A technically correct metric can still be analytically weak. Similarity depends on normalization and section boundaries. Keyword counts depend on the fixed vocabulary. Reported changes depend on concept and period matching. Publish the inputs and method beside the output, then use the output to decide what to read - not what to believe.
What consecutive MSFT and NVDA 10-Q comparisons demonstrate
The examples below join two independently validated StockEyes datasets dated Jul 19, 2026. The MD&A comparison isolates Part I, Item 2. The risk comparison isolates Part II, Item 1A. Each company’s current and previous 10-Q accessions reconcile across both datasets, and the current reported baseline comes from the same current 10-Q. The metrics describe documents and reported results; they do not assign importance, credibility, persistence, valuation, or a recommendation.
The two examples are intentionally different. A large section-length change can come from a changed disclosure state. A high-similarity document can still contain dozens of sentence changes. A lower-similarity document can reflect extensive rewriting. The correct comparison is not “which company changed most?” It is “which exact passage reaches a written research question, and what evidence tests the interpretation?”
Microsoft Corp.
Mar 31, 2026 10-Q compared with Dec 31, 2025 10-Q
58.3%86 added and 77 removed normalized sentences.
Products and segments+8 fixed-keyword mentions. This locates text; it does not measure importance.
91.0%63 added normalized sentences under the bounded Item 1A comparison.
AI and model behavior-3 fixed-keyword mentions. Read the exact passage before interpreting it.
Reported baseline: Q3 revenue changed +18.3% year over year; operating margin changed +66 basis points. The figures and the narrative come from the same current 10-Q accession, but the figures do not prove why the change occurred.
Research question: Which exact products and segments explanation changed, which reported number supports or challenges it, and what source should be checked next?
NVIDIA Corp.
Apr 26, 2026 10-Q compared with Oct 26, 2025 10-Q
27.1%64 added and 85 removed normalized sentences.
Supply and capacity-11 fixed-keyword mentions. This locates text; it does not measure importance.
68.9%55 added normalized sentences under the bounded Item 1A comparison.
Capital and liquidity-21 fixed-keyword mentions. Read the exact passage before interpreting it.
Reported baseline: Q1 revenue changed +85.2% year over year; operating margin changed +1649 basis points. The figures and the narrative come from the same current 10-Q accession, but the figures do not prove why the change occurred.
Research question: Which exact supply and capacity explanation changed, which reported number supports or challenges it, and what source should be checked next?
Two lessons from the same deterministic method
High similarity does not eliminate meaningful sentence-level work
Microsoft’s risk sections share a high proportion of normalized five-word sequences, yet the exact comparison still identifies added sentences and changes inside fixed topic families. A reader who sees only the similarity percentage could miss a narrow addition. A reader who sees only the added-sentence count could overstate routine drafting. Both measures are navigation aids that must converge on the exact passage.
Use the same discipline for MD&A. Start with the reported quarterly baseline, then identify which changed explanation relates to revenue, margins, product mix, capital spending, or uncertainty. Ask whether the explanation matches the statements and notes. A management discussion can be internally consistent without proving future persistence.
Lower similarity is a reason to read more, not a bearish label
NVIDIA’s MD&A comparison has less repeated phrasing than the other example under the same method. That may reflect changed tables, periods, product language, business conditions, organization, or drafting. It does not carry a positive or negative sign. The useful question is which operating explanation changed and whether the current reported results, segment evidence, supply disclosures, or later sources support it.
This boundary is central to evidence-first research. Document change tells you where the issuer’s presentation differs. Financial analysis asks whether a reported mechanism changed. Thesis review asks whether that mechanism matters to a written claim. Valuation asks what expectations are already reflected in price. Portfolio review asks what the security contributes to total exposure. Combining all five into one score hides the reasoning investors most need to inspect.
The Change-to-Evidence Filing Comparison Worksheet
Complete one row for each passage that reaches a real research question. Do not record every text difference. Preserve the exact filing pair and section, then distinguish document evidence from reported evidence and personal interpretation. Write “not established” when the source cannot answer a column.
| Changed passage | Source role | Reported anchor | Thesis connection | Alternative + next proof |
|---|---|---|---|---|
| Quote or precisely paraphrase the narrow change; record form, section, period, accession, and URL. | Reported fact, management explanation, risk disclosure, definition, legal update, forward-looking statement, or repeated context. | Comparable statement value, note, segment table, disclosed event, or “not established.” | Supports, challenges, question, or no change for one written causal condition - with your reasoning. | State another explanation and the exact source, metric, or date that could distinguish it. |
| Example: the supply discussion contains changed wording in the newest Item 2. | Management explanation; not independent evidence of demand or constraint severity. | Inventory, commitments, segment revenue, gross margin, and cash-flow facts from the same accession. | Question: does capacity expansion support the demand thesis without breaking the capital-intensity boundary? | Alternative: product-transition timing. Next proof: next 10-Q supply commentary, inventory, commitments, and segment result. |
Print tip: save the worksheet with the comparison date and method version. A later filing can change the interpretation, but it should not rewrite what the earlier documents contained.
Which filing change deserves a research task?
- Are the issuer, form, periods, accessions, and section boundaries correct?If no, stop and repair the comparison. Do not interpret the output.
- Is the difference only a date, table label, formatting change, or moved paragraph?If yes, record it as structural unless the surrounding context changes meaning.
- Does the passage add, remove, narrow, expand, or redefine a claim?If unclear, open both filings and preserve more context before classifying it.
- Can a reported number or defined disclosure anchor the language?If yes, link it. If no, label the passage as a statement or unresolved disclosure rather than a proved mechanism.
- Does the change reach a written thesis condition or risk boundary?If no, archive it as context. If yes, explain the connection in your own words.
- Is there a credible alternative explanation?Record it and name evidence that would distinguish the competing interpretations.
- Can a specific future source resolve the question?Create a task with the filing, note, metric, event, owner, and review date.
- Does the conclusion require valuation or portfolio context?Move that work into its own review. Filing language alone does not produce an investment action.
Nine ways filing comparisons create false confidence
A 10-Q update is compared with a full 10-K section and the structural difference is described as a sudden business change.
A no-material-change statement is treated as an empty risk section instead of a reference to the annual disclosure.
A longer section is called more important, risky, transparent, or alarming without reading why the length changed.
Repeated five-word sequences are treated as proof that meaning stayed the same, even though one changed condition can matter.
More mentions of AI, demand, regulation, or supply receive a bullish or bearish label without passage context.
A reported result and management’s account of the driver are written as one independently verified claim.
A KPI, segment, non-GAAP measure, or period is compared after its definition or composition changed.
A generated answer removes accession, section, passage, uncertainty, or the ability to inspect the original source.
A disclosure change becomes a buy or sell instruction without valuation, expectations, exposure, tax, liquidity, or suitability work.
A disciplined 20-minute filing-change pass
- Identify the issuer. Match legal name, CIK, ticker, and security context.
- Record both filings. Save form, report period, filing date, accession, amendment state, and direct URL.
- Choose one section. State the exact start and end boundary and why it answers the research question.
- Check the annual baseline. Determine whether either quarterly section incorporates or depends on the latest 10-K.
- Establish reported facts. Capture comparable values, concepts, periods, units, and relevant notes before prose.
- Inspect section structure. Note tables, headings, cross-references, changed definitions, and reorganized content.
- Use metrics for routing. Review word delta, exact sentences, similarity, and topic families without ranking importance.
- Open the exact passages. Read surrounding sentences, qualifications, forward-looking labels, and footnotes in both filings.
- Classify the source role. Separate reported fact, management statement, risk disclosure, definition, calculation, and inference.
- Write one atomic change. Include what changed, where, when, and what the source does not establish.
- Map the thesis condition. Name the customer, product, segment, margin, cash, capital, dilution, or risk mechanism involved.
- Test an alternative. Write another explanation that fits the available evidence.
- Name the missing proof. Choose the filing note, KPI, event, regulator source, or future result that could resolve it.
- Set the review date. Tie the task to a known filing or event when possible; otherwise use a deliberate calendar date.
- Preserve the record. Save the filing pair, method, passage, anchor, interpretation, alternative, and next proof together.
Turn one changed topic into a falsifiable research question
Choose one company example above and open both current and prior SEC filings. Select either Item 2 MD&A or Part II Item 1A. Use the StockEyes topic table to locate a changed family, then ignore the count while reading the exact passages. Write one sentence describing the narrow textual change without assigning sentiment or importance.
Next, open the current Quarterly Change Board and capture one reported number from the same current 10-Q accession. State whether the number supports, challenges, or does not establish the changed explanation. If the periods or definitions do not match, record the mismatch instead of forcing a conclusion.
Finish the worksheet with a written thesis condition, one alternative explanation, and one future source. A strong task might read: “At the next 10-Q, reconcile the changed supply discussion with inventory, commitments, segment revenue, and gross margin using the same definitions.” A weak task reads: “Watch supply risk.” The first can be completed and challenged; the second only preserves anxiety.
A better filing comparison is still not a complete investment decision
SEC filings are issuer disclosures prepared under legal and accounting requirements. They are essential primary sources, but they do not contain every fact an investor may need. Customer behavior, competitor actions, channel conditions, regulation, market prices, independent reporting, and later events can change the interpretation. A filing comparison should therefore preserve unresolved questions rather than claiming complete company truth.
Document evidence also stops before valuation. A favorable reported change can be more than reflected in market expectations. A difficult disclosure can be paired with a price that already discounts severe outcomes. StockEyes’ public filing tools intentionally do not attach price targets, expected returns, or stock rankings because those require point-in-time market and estimate data plus explicit assumptions.
Portfolio action is another separate layer. The same company evidence can matter differently for a small position, a concentrated holding, an employer-linked exposure, or an investor with different liquidity, tax, and time-horizon needs. The purpose of the filing workflow is to make evidence and interpretation traceable. It improves the quality of the question; it does not answer every question for every investor.
Questions to settle before trusting a filing diff
What is the best way to compare two SEC filings?
Confirm the issuer, form, report periods, filing dates, and accessions; compare the same bounded section; use text changes only to locate passages; anchor each passage to reported numbers or a stated disclosure; separate management language from your interpretation; and record the next source needed. Do not treat a word count, similarity percentage, or automated summary as a conclusion.
Which sections should investors compare first in a 10-Q?
Start with the financial statements and notes, then Part I Item 2 MD&A, Part II Item 1A Risk Factors, and any relevant legal, controls, or exhibit disclosures. The right order depends on the research question, but reported facts should usually be established before changed narrative is interpreted.
Does a lower text-similarity score mean a bigger business change?
No. Similarity measures repeated wording under a specific normalization method. Period labels, tables, reorganized sections, drafting, incorporated references, and genuinely new disclosures can all change the score. Business importance requires reading the exact passages and connecting them to financial or operating evidence.
Does more risk-factor wording mean a stock is riskier?
No. More words can reflect a fuller quarterly update, a legal drafting change, a new rule, a changed business condition, or a different way of incorporating the annual baseline. The text comparison can route attention, but it cannot determine probability, severity, valuation impact, or suitability.
Can AI summarize the changes between filings?
AI can assist only when the workflow preserves the exact filing pair, section boundary, citations, claim type, and uncertainty. A summary that cannot link each material claim to the source passage should be treated as an unverified research lead, not as filing evidence.
Does comparing SEC filings tell an investor whether to buy or sell?
No. Filing comparison identifies reported changes, management explanations, disclosure changes, and unanswered questions. A security decision also requires valuation, expectations, portfolio exposure, time horizon, taxes, liquidity, and personal constraints that the document comparison does not establish.
Open the primary material and the working comparisons
- Investor.gov: How to Read a 10-K
- Investor.gov: How to Read a 10-K and 10-Q
- SEC: Beginner’s Guide to Financial Statements
- StockEyes Management Discussion Delta Desk
- StockEyes Risk Disclosure Delta Desk
- MSFT: current 10-Q used in the worked example
- NVDA: current 10-Q used in the worked example
StockEyes generated the worked comparison from SEC-hosted primary documents on Jul 19, 2026. The public pilot covers MSFT and NVDA only. Later filings, amendments, corrections, parser changes, and changed definitions can alter the record. Verify the source filing before relying on a comparison.