Revenue is up 20%, operating cash flow has increased, three customers praise the product, and the stock is well below its old price. You already have a headline: growth continues; the market has missed it. What will you type into the search box next?
Searching why the company deserves a long-term holding can produce an increasingly polished supporting case. Instead, identify the fact most capable of stopping that article. The task is to make the conclusion vulnerable to evidence, not merely adopt a more pessimistic tone.
Contrary evidence is more than an opposing opinion
You will identify unsupported steps in a bullish brief, distinguish strong disconfirmation from forceful but weak objections, and save sources, competing explanations, and update conditions. Retaining a view, reducing confidence, and suspending judgment are all possible outputs.
Confirmation bias describes seeking or interpreting evidence in ways favoring an existing belief or hypothesis. Nickerson's review distinguishes examining evidence to reach a conclusion from selectively assembling support for a conclusion already chosen; selective treatment need not be conscious deception.[1]
You need not decide first whether you are a biased person. Compare standards. If a management quotation suffices for good news while contrary evidence must establish an irrefutable final verdict, the process treats evidence asymmetrically. That difference can be examined without reading anyone's mind.
Save a provisional view before opening the attachment
C5-55 is a complete fictional research exercise about Morning Bridge Software, distinct from other chapters' companies. All financial amounts are USD millions, assessed October 29, 2026. No valuation model, current recommendation, or real personal experience is supplied.
The brief reports current revenue 120 versus 100, operating cash flow ten versus eight, and three published customer cases saying the product saves time. Two news items repeat the same corporate release. A former $90 share price is now $60. The brief concludes that existing-customer revenue keeps expanding, new customers add growth, and the price is attractive.
Before continuing, save three lines: supplied facts, the inference most needing verification, and evidence that would reduce your confidence. Mark the judgment provisional because inputs are incomplete. No initial confidence is also useful if its missing requirements are explicit. Do not reconstruct this first version after reading the answer.
Now open the same-day attachment. All prior revenue of 100 came from the original customer cohort. That same cohort contributes only 90 this period; new customers supply 30, totaling 120. There are no acquisitions or currency effects, and revenue scope and measurement match. Operating cash flow ten is company-wide, without customer-specific collection or renewal explanations.
Revenue retention for the original cohort is 90/100, or 90%. This is a revenue-based net measure that permits expansion by some original customers to offset contraction by others. It does not say 90% of customer headcount stayed. Total revenue rises 20% while the original cohort's contribution falls 10%; both statements can be true.
Follow the objection that actually tests the thesis
The strongest contrary evidence is not generic industry competition. The attachment directly challenges the specific existing-customer expansion claim. New revenue of 30 offsets an old-cohort decline of ten. If acquisition slows, the proposed growth explanation may fail. That opens a concrete causal question rather than adding a decorative risk paragraph.
The attachment does not establish inevitable decline. A large customer's contraction, product transitions, and widespread departures would have different implications. New customers' persistence and profitability also remain unknown. Do not replace an unsupported bullish narrative with an equally complete invented bearish one.
Specify the next documents: same-cohort renewals, usage, discounts and cancellations; major-customer concentration; new-customer revenue and collection; and unchanged metric definitions. Aggregate-only sources leave causes unresolved. AI cannot supply missing customer behavior from the gap.
The three cases still show selected uses worth investigating. Without sampling methods and unsuccessful cases, they do not establish expansion across the customer population. Two articles reproducing one release also do not provide two independent verifications. Count underlying sources as well as files.
Give a weak objection its actual weight
A separate anonymous post claims that dark office windows photographed on a weekend prove failing business. It lacks verifiable date, staffing arrangements, and a connection to customer outcomes. Even an authentic office photograph would not establish falling revenue or accounting fabrication.
Record it as a weak lead, rather than giving it half the report to demonstrate balance against the cohort table. If its author later supplies an authenticated major-customer termination, assess that new document's date, scope, and amount. A source's earlier position should not permanently raise or lower every future claim.
Apply the same verification standard to opposing and supporting arguments. Requiring three bullish points and three bearish points can crowd out decisive evidence with weak objections. When reliable contrary material has not been found, say so instead of requesting an evenly matched AI debate.
Is $90 evidence or a starting value?
“Previously $90, now $60, therefore cheap” uses the old price as an estimation starting point. Anchoring describes estimates being drawn toward an initial value, with subsequent adjustment remaining too close. Tversky and Kahneman's original work examines starting-value effects and how vivid descriptions can displace base rates.[2]
An earlier price is not always useless. It can lead to investigation of earlier expectations, financing, or changed markets. Purchase cost also matters for personal accounting and applicable taxes. It simply cannot establish current fair value by itself. The first target price or first AI valuation deserves the same scrutiny.
Hide $90 and assess what business, profit, cash, and comparative valuation inputs support. Retain an independently justified range if one exists. If the argument reduces to eventual return to the old price, move that claim from evidence to unverified expectation. An anchoring concern prompts calculation review, not automatic trading in the opposite direction.
A story resembling winners still needs its denominator
Use separate synthetic frequencies, not a probability assignment to Morning Bridge. In a predefined set of 100 companies with complete outcomes, twenty achieved at least 20% growth in each of the following two years, while eighty did not. A promotional characteristic observable before evaluation appeared in fourteen successful and forty unsuccessful companies.
Saying 70% of successes had the characteristic can sound like a 70% chance of success for companies with it. But 54 companies had the characteristic and only fourteen succeeded: approximately 25.9%. Fourteen divided by twenty describes the characteristic among successes; fourteen divided by 54 describes success within the characteristic group.
This does not establish that stories convey no information. Overall success is 20%, versus about 25.9% within this group. Transfer still needs checks of sample selection, comparable industries and periods, prior observability, and complete failures. A retrospectively selected collection of celebrated companies would undermine even this comparison.
You need not memorize Bayes' formula here. State the full set, success definition, and conditioning group. A vivid story may suggest a mechanism without supplying a probability. Missing base-rate evidence is a gap, not an invitation to invent a precise success chance. Group-level psychology findings likewise do not establish that every investor makes the same error.
Register an update before the next disclosure
For Morning Bridge, a provisional update can reduce confidence in old-customer expansion as the growth driver and decline to infer cheapness from aggregate growth alone. Continue checking consistently defined cohort retention and its causes.
This exercise specifies a reconsideration condition: at least 100% net revenue retention for two consecutive periods under the same method, supported by checkable renewals and collections. Continuing below 100% while relying on more new revenue to offset contraction retains or strengthens the concern.
Two periods and 100% are teaching observation conditions, not universal software-investment thresholds. Meeting them repairs only part of the business explanation. Margins, acquisition spending, competition, and valuation remain separate. Short-term share-price movement does not directly test this operating claim.
When using AI, request source locations, the documents best distinguishing explanations, and duplicate-source checks. A generated suggestion that customers may leave is a hypothesis. Only verified material changes the fact column; stronger rhetoric cannot strengthen the evidence.
Recalculate with the next cohort
Explanation question: Why should the revenue attachment and dark-office post receive different weight even though both oppose the brief? Answer before checking the calculations.
The attachment directly tests the stated claim under matching definitions. The photo lacks verification and a sufficient revenue connection. Evidential weight follows relevance and reliability, not equal representation of opinions.
For an independent application variation, keep USD-million units. Prior-period revenue is 120, and all customers from that period now form the next old-customer cohort. They contribute 105; new customers add twenty, totaling 125, again without acquisitions or currency effects. An analyst claims that 105 exceeds the earlier 90, so retention is above 100% and the thesis is restored. Check the denominator and record your update first.
Revenue grows from 120 to 125, approximately 4.17%. Net retention is 105/120 = 87.5%, not 105/90. The customer set changed between comparisons; two old-customer amounts are not interchangeable retention denominators.
The registered restoration condition remains unmet. Retain the concern and investigate causes without asserting inevitable failure. A correct update can be narrower than either the original enthusiastic story or its complete reversal.
Save evidence capable of changing your own conclusion
Retain initial and revised versions. Include original claim, direct documents and dates, fact/interpretation/prediction categories, source independence, strongest contrary evidence, weaker objections and reasons, competing explanations, next document request, observable update conditions, and remaining unknowns.
If retaining the original view, state which contrary evidence was checked and why it did not warrant revision. For self-check, an opposing paragraph is not a source, purchase price is not a valuation proof, and a base-rate denominator cannot change unnoticed. Update conditions should not depend on price eventually satisfying you.
Could another reader identify the next fact that would change the conclusion from this record alone? If not, the output may still be a polished supporting brief with cautious wording rather than a testable judgment.
Let possible revision guide the search
Find the observable condition on which the original explanation most depends, then compare evidence's relevance, reliability, and independence. Balance should not obscure decisive material, and an appealing story should not obscure its denominator.
Writing update conditions before the next release gives evidence a chance to change the conclusion instead of allowing the conclusion to direct every search.
References
- Nickerson: Confirmation Bias—A Ubiquitous Phenomenon in Many Guises. Review of General Psychology 2(2), 1998, 175–220. Definition and selective evidence gathering; a review, not validation of this exercise's intervention.
- Tversky and Kahneman: Judgment under Uncertainty—Heuristics and Biases. Science 185(4157), 1974, 1124–1131. Base rates and anchoring; the company and frequency examples are separate synthetic inputs.
Sources checked October 30, 2026. Company, documents, prices, and frequency table are fictional. AI-assisted draft; independent AI full-text and bilingual review is complete. Human financial review and target-reader testing have not been conducted and are not scheduled for this production round.