Where should I find primary sources before researching a stock?

Opening ten browser tabs is easy. Knowing which document would settle your question is harder. Begin with a sentence narrow enough to investigate: Did this business generate more revenue from existing operations, or did an acquisition explain the increase? That question gives you a destination. A general search for whether the company is good does not.

A primary source is the original disclosure, agreement or dataset behind a claim. For public-company research, Investor.gov: Using EDGAR to Research Investments identifies annual, quarterly and current reports available through the SEC database. The document type helps you choose where to look; it does not make every statement equally certain.

Keep three categories separate in your notes: what the company reported, what management expects, and what you conclude. A filing can contain all three kinds of material, but your interpretation is still yours. This separation prevents an attractive forecast from quietly becoming an observed result in your research.

Build a small source map rather than a large reading list. For a question about a business model, start with the business description. For a number, locate its financial statement and related note. For why it changed, find the management discussion. For an unusual transaction, identify the relevant announcement and agreement rather than relying on an annual-report summary written months later.

Investor.gov: How to Read a 10-K describes the annual report's business, risk, management discussion and financial statement sections. Use those as navigation aids. Read the section that answers your question first, then follow its references.

Record the company's legal name as well as the familiar brand. Similar names, different share classes and subsidiaries can send you toward the wrong issuer. Keep the report's period end and publication date beside its link. When comparing figures, also note currency, units and whether the number covers the whole group. These details take seconds to record and can prevent an hour of analysis built on mismatched inputs.

Consider fictional Harbor Instruments. An article says annual revenue grew 25%, from $80 million to $100 million, and describes strong customer demand. You locate the original annual report and a hypothetical acquisition note stating that a purchased business contributed $18 million after completion. All figures in this example are invented for learning.

The reported increase is $20 million: $100 million minus $80 million. Removing the disclosed acquisition contribution gives $82 million, or $2 million above the prior year. The rough growth rate excluding that contribution is $2 million divided by $80 million, or 2.5%.

This is a useful reconciliation, not a verified organic growth rate. Currency movements, disposals and sales between the businesses might still affect comparability. Write exactly what the arithmetic supports: most of the reported increase is explained by the acquired revenue contribution. Do not replace the company's headline with an equally unsupported claim that customer demand grew precisely 2.5%.

Create one row per important claim with six fields: claim, original link, location inside the document, reporting period, exact definition and your interpretation. Add a short unresolved-question field when a comparison remains incomplete. A page reference or note heading is more useful than a bare link to a two-hundred-page report.

For Harbor, the ledger should retain the reported $100 million, the $18 million acquisition contribution and the specific calculation you performed. Someone revisiting the work should be able to reconstruct it without trusting your memory. Save both the original figure and your adjusted calculation instead of overwriting the first with the second.

Use secondary sources to discover questions and alternative explanations. If a journalist points out a contract concentration issue, follow that lead into the customer disclosure. Good outside analysis can improve your search dramatically. Its usefulness does not remove the need to distinguish the author's reasoning from the underlying evidence.

• Write one factual question before searching. Name the metric and period you need.
• Confirm the issuer's identity and find the original reporting document.
• Record publication date, period end, currency, units and business scope.
• Read the note defining your metric before comparing it with another figure.
• Reproduce the important calculation and label any adjustments you introduced.
• Look for a later update that could supersede the relevant information.
• End with one supported conclusion and one unresolved question, rather than a broad verdict.

A useful stopping rule is that every material number in your paragraph has a traceable location and every assumption is visible. You do not need to read every filing before understanding one issue. If the answer depends on information the company has not disclosed, say that explicitly and avoid manufacturing a more precise estimate from unrelated data.

For Harbor Instruments, the acquisition question requires several distinct pieces of evidence. Start a hypothetical document chain with the announcement of an agreement, then the completion announcement, then the reporting note describing the acquired business's contribution. Each item answers a different question and belongs in a separate row.

The agreement announcement can establish what the parties intended and which conditions were described. It cannot establish that the deal subsequently completed. The completion announcement can establish the reported completion status, but may still leave the contribution to group revenue unresolved. The later reporting note is where the hypothetical 18 million contribution enters the original example.

Write the question beside each link rather than treating the chain as a stack of confirmations. Did the transaction close? When did the acquired activity enter the reported group figures? Which period does the contribution cover? A document that answers the first question need not answer the third.

If an expected completion date changes, retain that change in the chain. Do not apply the originally expected date to the later accounts simply because it appears in the earliest announcement. Likewise, do not annualize a partial contribution without showing the assumed period and the limits of that exercise.

This method is useful because it makes missing links obvious. You might have a completed transaction and group revenue but no separate contribution disclosure. In that case the acquisition explanation remains qualitative; the missing contribution cannot be manufactured by subtracting unrelated segment totals.

A copied number often loses the surrounding information that made it interpretable. Build your extraction around a miniature record: table title, row label, column heading, unit, footnote and value. Preserve the sign as well, especially when parentheses distinguish an expense or outflow from a positive amount.

Consider a hypothetical statement headed amounts in thousands, showing revenue of 82,000. The value represents 82 million in the stated currency. Copying 82,000 into a worksheet labeled millions would inflate the amount a thousandfold. The digits alone are not the data; the heading supplies part of their meaning.

Now imagine two adjacent columns labeled six months and three months. A search result might display both values without their headers. Before calculating growth, return to the original table and establish which column belongs to which period. If the document uses cumulative amounts, the latest column is not automatically the latest standalone quarter.

For a practical extraction exercise, ask another reader to rebuild one sentence using only your saved record. If they cannot tell whether the number covers the group or a segment, add that field. If the metric depends on a note, save its location alongside the table reference.

Capture only the context needed for the claim being investigated.

Suppose a hypothetical issuer publishes annual figures and later supplies a corrected comparison table. Your source map now contains versions of related reporting, not two independent estimates to average. The research task is to establish what changed and which claims depend on it.

Imagine Harbor originally showed prior revenue of 80 million, while a later correction changes that comparable figure to 84 million. Against current revenue of 100 million, reported growth changes from 25% to approximately 19.0%. The calculation is 16 divided by 84. Neither version should quietly overwrite the other in a historical research record.

Save a short revision entry identifying the replaced field, the new field, the stated reason and the date you learned of the change. If the issuer explicitly provides a revised comparable series, use that series for the current comparison while retaining the earlier value as an obsolete input. If the relationship between versions is unclear, describe the ambiguity before calculating.

Do not infer a business deterioration from a corrected historical denominator. Nor should you assume that every later document supersedes every part of an earlier one. A narrow correction may leave the other disclosures unchanged.

The practical question is dependency: which sentences, ratios or scenarios used the changed value? Follow those links through your worksheet. This allows a focused correction without rebuilding unrelated parts of the analysis or pretending the original information never existed.

A familiar brand can lead you to the wrong reporting boundary. In a hypothetical example, Harbor's website describes a subsidiary that sells laboratory equipment, while the listed parent also owns an unrelated service operation. A strong customer story about the equipment business does not establish the performance of the whole parent.

Create an identity card with the legal issuer, the business or subsidiary under discussion, and the scope of each disclosed number. If a statement refers to the consolidated group, label it group. If a table covers a particular segment, preserve the segment's exact name. Avoid using a brand name as a universal label for every figure.

Suppose the hypothetical equipment segment reports sales of 40 million and the service segment reports 65 million, while consolidated revenue is 100 million. The difference might be explained by 5 million of transactions eliminated on consolidation in this invented case. Adding the segment figures and calling 105 million external customer revenue would ignore that stated reconciliation.

This example illustrates why the relevant reporting boundary must be read, not guessed. Without an actual elimination disclosure, you could not assume the difference has that explanation.

Your worksheet should therefore retain both the segment totals and the group bridge. If the question concerns product demand, segment detail may be useful. If it concerns the parent's overall revenue, the consolidated figure answers a different question. Keeping both boundaries visible prevents evidence about one business from silently becoming evidence about another.

For each research question, write a retrieval plan before searching. Name the issuer, the event or metric, the period and the likely document section. Then define success as locating the specific disclosure or establishing that it remains unavailable after a focused search.

A completed hypothetical Harbor worksheet might read: question, acquisition contribution to current annual revenue; target, reporting note on the completed acquisition; required context, contribution period and group scope; result, 18 million reported contribution; limitation, comparable organic growth not established. That is enough to explain why the source matters.

Add a location that another person can use: the note heading, table title or clearly identified subsection. A bare homepage link forces the next reader to repeat your search. If a document has printed page numbers that differ from the file viewer's numbers, record which numbering you used.

Track failed searches briefly as well. If you checked the relevant note and found no customer retention disclosure, record that bounded result. Do not write that the company never discloses retention unless your investigation actually supports such a broad statement.

The finish line is a supported answer within the declared scope. You are allowed to finish with an identified disclosure gap. That outcome is especially useful when further searching would only reproduce the same unsupported claim through different outlets. Research quality depends on traceability, not on how many documents the folder contains.

A hypothetical customer announces that it is testing Harbor's equipment at one site. That is primary evidence of the customer's stated test. It is not evidence of a nationwide rollout, a long contract or Harbor's eventual profit from the relationship. Write the narrow statement first, then list the additional claims separately.

Independent documents can also describe different stages without contradicting each other. Harbor might announce shipment while the customer discusses evaluation. Shipment and acceptance are distinct events. Before declaring a contradiction, ask whether the parties are reporting different steps in the same process.

What if the customer document is more specific than the supplier's? Use the specificity for the question it addresses, but do not assume the customer knows the supplier's total costs. What if an agreement is partly unavailable? Identify the missing terms and refrain from treating favorable assumptions as contractual facts.

A useful evidence ladder for this particular example is test announced, equipment delivered, evaluation completed and commercial terms disclosed. These are research milestones you have chosen, not a universal legal classification. Each step supports a different sentence.

The payoff is a more precise conclusion. You can report that a test is underway while leaving commercial scale unresolved. That may feel less decisive than a partnership headline, but it gives the next reader a clear boundary between documented activity and the business outcome still being investigated.

Original disclosure is essential, but it has boundaries. Management chooses language, forecasts can fail, accounting involves estimates, and public documents may not reveal customer-level economics. Reading the original source reduces errors introduced by repetition; it does not eliminate business uncertainty.

Be especially careful when treating silence as evidence. A company that does not publish a retention rate has not thereby demonstrated weak retention. It has left that question unanswered for outside readers. Equally, a polished explanation cannot fill a missing numerical bridge.

The practical tradeoff is depth against coverage. A focused investigation can explain one revenue change very well while leaving valuation or financing untouched. Describe that scope honestly. Your finished note might say that acquisitions explain most reported growth, while the comparable growth rate remains uncertain. That limited statement is more useful than a sweeping assessment whose evidence cannot be traced.