An AI summary sounds confident: “Revenue increased 20%, proving that the brand is more popular. Multiple media outlets have confirmed the story, and growth next year is assured.” Several links follow. You open the first and find a genuine 20% figure. Is the paragraph ready for your research report?
The problem may survive even when the number is correct. A reported result can become an unsupported explanation, then a promise about the future. Taking the paragraph apart lets you preserve the useful information while seeing which claims still need evidence.
What you will be able to do
- Separate a persuasive investment story into reported facts, interpretations, and forecasts.
- Check whether a citation supports its particular claim and whether different reports provide independent evidence.
- Save an evidence record with unresolved questions and conditions that would change your interpretation.
The essential skill is checking support. EDGAR gives you access to company filings, but locating a document does not mean an outside authority has verified every statement it contains. [1]
Four claims hiding in one paragraph
Songhe Restaurants, its documents, dates, and figures below are entirely fictional teaching materials. This is a separate exercise; it does not replace the Morning Bridge figures in G04. Use the supplied information without filling gaps from your impressions of the restaurant industry.
Document A is a quarterly operating statement assumed to have been published on October 20, 2026. It reports current-quarter revenue of $1,200,000, compared with $1,000,000 in the same quarter a year earlier. The original locations contributed $960,000 in the current quarter and new locations contributed $240,000. Those new locations contributed no revenue in the earlier period. The document provides no order counts, price increases, profit figures, or guidance for the following year.
Document B is a post from the company's marketing account, published that day: “Revenue grew 20%. Customers love Songhe more than ever.” Document C is a news brief published the following day. It reports 20% growth, attributes the figure to the company's statement, and links to Document A. It includes no independent investigation.
The AI has combined these materials into the opening paragraph. Write its claims separately: revenue increased 20%; growth proves greater brand popularity; multiple outlets confirmed the story; next year's growth is assured. Each needs its own assessment. One correct calculation cannot confer a “verified” label on the entire paragraph.
The first claim has a reproducible calculation. Divide $1,200,000 by $1,000,000 and subtract one: growth is 20%. A more precise report would say, “The company reports year-over-year quarterly revenue growth of 20%, which can be recalculated from the supplied figures.” That wording preserves the source of the information. Checking the arithmetic is different from independently auditing the company's accounts.
The second claim explains a cause. Revenue could change because of store openings, prices, sales volumes, or other factors. The third makes a claim about the evidence itself: how many sources exist and what they checked. The fourth concerns a future outcome. Current-quarter figures cannot guarantee next year's business results.
A fact, interpretation, or forecast is not identified by how confidently it is written. Ask what the sentence is trying to establish: what happened, why it happened, or what will happen. Then ask whether the available material can support that particular step.
The same 20% can tell two different stories
Return to Document A. All $1,000,000 of earlier-period revenue came from the original locations. Their current-quarter revenue is $960,000, a decline of $40,000, or 4%. New locations contributed $240,000. Their contribution more than offset the decline, producing total revenue of $1,200,000.
These inputs support a simple comparison for the original group. They do not create a formally defined same-store sales measure for a real company. Such a measure may have eligibility rules, operating-day requirements, or currency adjustments. Keep the comparison within the scope actually supplied.
The marketing explanation has crossed from total growth to a claim about stronger demand for the existing business. Yet new-location revenue and declining original-location revenue can coexist. The marketing account might have other evidence, but it has not provided that evidence here.
You could write: “The supplied breakdown shows that new locations contributed the growth in total revenue, while revenue at the original locations declined. The reasons for that decline remain unknown.” This retains the calculation without declaring that the brand has improved or deteriorated.
Finding a hole in an optimistic explanation does not establish its opposite. It also does not establish a reason to sell or short a stock. The immediate result is a narrower business conclusion and a better-defined question about what remains unexplained.
A working link can lead to the wrong support
Suppose the words “growth next year is assured” link to Document A. The link opens, and the document really contains revenue figures. It contains no forecast for next year. The citation is genuine; the claimed support is absent.
Put the sentence and the relevant source passage beside each other. Do they concern the same business, period, and metric? Did the summary preserve qualifications such as “may,” “expects,” or “if these conditions hold”? If you cannot identify a supporting passage, mark the citation as not supporting the claim. A familiar domain is not a substitute for that comparison.
An original source helps you recover definitions, notes, and context. It can still have a point of view. Management knows its operations and also has reasons to present results favorably. Its explanations need to be examined alongside financial information, subsequent results, and other evidence.
The SEC's guide explains that companies prepare and submit their 10-Ks; the SEC does not prepare them or endorse their accuracy. [2] “Filed with the regulator” therefore cannot become “the regulator endorses this company's growth prospects.” Establish the connection between a particular statement and a particular passage, rather than borrowing credibility from an institution's name.
Five articles may lead back to one source
Four more websites now republish Document C. Lin counts five news reports and feels more confident. Follow their references, however, and all five lead back to the company's single operating statement. The number of web pages has increased. The number of independent checks of the revenue figure has not.
Draw an arrow from each report to the material it relies on. Republishing can help you discover information, but it does not automatically create another independent confirmation. Where the arrows end matters more than the number of search results.
A reliable record assembled through a different collection process might add information. First check whether it measures the same thing. Observed store traffic is not company revenue; a customer survey is not every customer's experience; a supplier's orders are not necessarily revenue the company has recognized. Different sources can address related questions without measuring interchangeable quantities.
Record possible interests as well. Does a promoter receive compensation, own the shares, or earn a referral payment from an account opening or course sale? An interest does not prove a claim false. An undisclosed interest is also not something to invent. These questions identify claims that deserve closer checking; they do not justify assigning a blanket truth label to a person.
Give your own explanation a way to fail
It is easy to inspect someone else's story and stop there. Suppose your explanation is now, “Songhe's expansion is hiding weaker demand at its original locations.” That is more specific than “the company is in trouble,” but it remains an interpretation.
Original-location revenue could decline because of demand, temporary closures, operating days, or changes in store formats. Identify those alternatives before requesting evidence. Comparable orders, average revenue per order, operating days, and explanations of store changes would help. If those inputs are unavailable, leave the cause unresolved instead of asking AI to supply a plausible narrative.
Write a review condition before the next information arrives. For example, stable orders under a consistent definition, together with verified temporary closures explaining the revenue decline, would require you to reconsider the weaker-demand explanation. Continued order declines for a consistent group would add support to it. Neither condition has been observed in the supplied materials.
The aim is not to reject a claim after any single unfavorable observation. The explanation needs enough definition for new evidence to affect it. “The brand will win in the long run” gives you little to investigate if it has no deadline, metric, or condition under which you would reconsider.
Confidence should come with reasons. You can be relatively confident about the disclosed figures and arithmetic, lack order data needed to assess the cause, and have insufficient evidence for next year's result. You do not have to attach a number such as 73% to each sentence. Without an estimation method, extra precision merely decorates the uncertainty.
When a claim promises returns
Now imagine Document B adds an invitation to a private group, a promise of profits next month, and instructions to transfer money to an unfamiliar account. This raises questions beyond interpreting restaurant growth. The SEC identifies promises of high returns with little risk and impersonation of investment professionals as warning signs, and advises checking identity through independently obtained contact details. [3]
Do not let the sender's profile or supplied link serve as its own verification. Finding a real registered firm or professional does not establish that your correspondent represents them. Even a genuine identity would not make a particular return promise valid.
A polished backtest chart also needs inputs, rules, sample dates, and costs that someone can check. A screenshot may be selected, altered, or fabricated. Without a reproducible calculation, record it as an unverified display rather than verified performance. Later chapters examine backtests in detail; the distinction here is between seeing a result and checking it.
Try the checks yourself
Before reading the answer points, explain why Document C and its four republications do not provide five independent confirmations. Then rewrite the opening claim about 20% growth, stronger popularity, and assured future growth. Preserve supported information and identify the missing evidence.
Reference answers · reveal
Answer points: The reports rely on one company statement and add no independent collection or verification. A defensible rewrite is: “The company reports 20% year-over-year revenue growth. The supplied breakdown shows that new locations contributed the increase while revenue at the original locations fell 4%. Changes in brand demand and growth next year remain unresolved.” A marketing explanation is not a reported operating fact; an evidence gap is not a bearish investment conclusion.
Try a different business
A fictional software company reports 1,200 paid accounts, up from 1,000. A newly acquired business contributed 250 accounts; the original business ended with 950. No renewal rates, product prices, or acquisition costs are supplied. A summary says, “Paid accounts increased 20%, proving that existing customers are becoming more loyal.” Check the number and the explanation separately, then name one document or dataset to request. Write before reading on.
Reference answers · reveal
Answer points: Total accounts increased 20%, while accounts in the original business declined from 1,000 to 950, or 5%. That decline is not a formal renewal rate. We do not know additions, losses, or the composition of the starting customer group, so it cannot settle customer loyalty. Request renewal and churn information for a consistently defined customer group, checking account definitions and the acquisition's effect on the reporting scope. A same-store label from the restaurant example cannot simply be transferred to software accounts.
Save an evidence record
Use the opening summary or a short public passage of your choice. Record each claim, its fact/interpretation/forecast label, the original source and page or paragraph, publication date, relevant period, metric definition, supporting passage, and verification status. Save the access date and limit copied excerpts to the brief wording needed for the check.
Add source relationships, known interests, supporting and contrary evidence, missing material, and the next action. Keep the first version and append updates as information changes. This is the purpose of the book's T02 evidence record: to show both why you held a view and what changed it.
For a self-check, choose one conclusion and follow the record back to its source. Can you explain exactly how far that passage supports the conclusion? Can you identify evidence that would change your view? A collection of links with the word “credible” beside it has not yet done that work.
Follow the support, then update the claim
Start with specific claims and match each to an object, period, metric, and source passage. A citation provides an entry point; you still have to check its support. Trace republications to their common origin before counting them as corroboration.
Then preserve alternative explanations and conditions for reconsideration. Give forecasts a deadline and visible uncertainty. When information is missing, narrow the conclusion and define the next check. A useful research record lets another person retrace the evidence and see how you would respond when it changes.
References
- Investor.gov: Using EDGAR to Research Investments. Supports access to company filings and their uses.
- SEC: How to Read a 10-K. Used for company authorship and the SEC's role; the older guide's complete list of form items is not presented as current regulation.
- Investor.gov: Social Media and Stock Tip Scams — Investor Alert, February 6, 2026. Supports the identity and return-promise warning signs and independent contact checks.
Editorial status
AI-assisted draft, synchronized to the accepted second-round Chinese text. All Songhe and software-account inputs are original synthetic exercises. Source, arithmetic, and teaching checks are recorded in the author review. A separate AI reviewer has completed an independent full-text review. Human financial review and target-reader testing remain pending.