What should I do when credible investment sources contradict each other?

Two credible sources can report different growth rates without either being wrong. They may be measuring different periods, currencies, business boundaries or accounting definitions. Before choosing a winner, rewrite both statements in a common format.

Use a sentence with five slots: entity, metric, period, measurement basis and value. A claim about total group revenue in reported currency is not directly comparable with a claim about one segment's constant-currency growth. If any slot is missing, the comparison is unfinished.

This approach also helps when an optimistic report and a cautious report seem irreconcilable. One may describe current profitability while the other discusses future financing needs. Those are different questions. Your goal is to isolate the smallest genuine disagreement. Otherwise you risk spending hours evaluating author credibility when the apparent contradiction is only a mismatch in scope.

Locate the source table and its notes rather than comparing paraphrases. Investor.gov: How to Read a 10-K identifies the financial statements and management discussion within a 10-K. Those locations can help you find the definition and management explanation relevant to the disputed figure.

Check whether a later event changed the picture. Investor.gov: How to Read an 8-K describes disclosures that can arrive between periodic reports. A source using an earlier snapshot may have been reasonable when written and simply no longer answer the same present question.

Create a reconciliation worksheet with the two values, their definitions and possible adjustments. Do not immediately average them. An average of incompatible numbers has no clear meaning. If one source measures revenue and another measures bookings, taking their midpoint does not produce a better estimate of either. The reconciliation should preserve the original values until you understand why they differ.

Fictional Elm Services generated $200 million of revenue last year and $230 million this year. Source A reports 15% growth, calculated as $30 million divided by $200 million. Source B reports 8% growth for comparable operations at constant currency.

Assume the company's hypothetical bridge attributes $10 million of the increase to an acquisition and $4 million to currency translation. Removing those contributions leaves $216 million. The comparable increase is $16 million, which divided by $200 million equals 8%. Both rates can describe the same year accurately under these simplified assumptions.

The research error would be to call Source A dishonest or Source B pessimistic before examining the bridge. The remaining substantive question is which growth rate best answers your question. Total reported growth describes the consolidated result. Comparable growth helps investigate expansion within a consistent business base. Neither alone tells you whether the acquisition created value or whether currency movements will reverse.

Sometimes sources use the same facts and genuinely disagree about the future. Move the disagreement into explicit assumptions. If both accept current revenue but differ on next year's margin, identify the cost or pricing mechanism behind each forecast.

Construct a discriminating question: what observation would be more likely under one explanation than the other? If one analyst expects temporary launch costs to fade and another expects permanent service costs, future cost per customer may be more informative than total revenue growth. Specify the observation before new data arrive so you do not reinterpret every outcome to preserve your preferred story.

Do not force a probability onto the alternatives without a basis. A scenario table can show consequences without pretending to know their likelihood. You can conclude that the disagreement remains unresolved and identify the next disclosure that could narrow it. That is stronger research than blending incompatible arguments into a vague middle position.

• Write both claims with entity, metric, period, basis and value.
• Find the original source location behind each statement.
• Check units, currency, business boundaries and publication dates.
• Recalculate the difference using only disclosed adjustments.
• Separate differences in observed facts from differences in forecasts.
• Identify an observation that could distinguish competing explanations.
• Preserve unresolved disagreement instead of manufacturing a compromise.

Give each outcome a clear label: reconciled definitions, outdated information, calculation error, genuine assumption difference or insufficient evidence. These labels tell you what to do next. A calculation error can be corrected immediately. An assumption difference may require a future disclosure. Insufficient evidence may justify ending the investigation for now. Clear classification prevents the same debate from restarting every time another article appears.

Consider a separate hypothetical disagreement about customer retention at Elm. A report says 95% of customers renewed; another says revenue from renewing customers fell 12%. These claims sound inconsistent only if customers are assumed to spend identical amounts. Suppose Elm had 100 customers, of whom 95 renewed. The original cohort previously contributed $10 million, but the renewing customers now contribute $8.8 million. Customer renewal is 95%, while the cohort's revenue contribution is 88% of its prior amount. Both statements fit the invented data.

Now suppose a third report calls that 88% figure net revenue retention, but its calculation includes $1 million from newly acquired customers. Removing that amount leaves $7.8 million from the original cohort. The problem is no longer a difference in emphasis: under a definition restricted to the original cohort, the third report used an incompatible numerator. Keep its published number in the evidence record, then show the corrected calculation separately.

This exercise gives the research note a useful structure. Identify the population first, then the movement within it, and finally the label applied to that movement. A high customer renewal rate does not settle spending retention, and neither establishes profitability. Resolving one disagreement should narrow the question rather than quietly turning a limited result into a broad judgment about commercial strength.

Use a hypothetical three date sequence. Elm releases annual figures on Monday. A researcher publishes an analysis on Tuesday using those figures. On Friday, Elm announces that a large customer will not renew next quarter. A second researcher publishes a lower forecast the following Monday. The forecasts now differ, but their information sets differ too. Comparing them without the sequence penalizes the first author for lacking information that did not yet exist.

Record four dates when available: the end of the measured period, the issuer's release date, the research publication date, and the date of any revision. An article's latest website timestamp may reflect a layout change rather than a rewritten forecast. If you cannot establish what changed, mark the revision history unknown. Do not infer that every sentence incorporates all information available at the most recent timestamp.

Preserve the original question as well. A forecast made before the customer announcement can still be relevant when evaluating the author's reasoning at that time. It is less useful for answering a question about the business after the announcement. Your worksheet should therefore distinguish historical assessment from present research. The practical resolution may be to retire an older forecast from the current comparison while retaining it in the audit trail, without alleging that its original reasoning was defective.

Suppose two hypothetical researchers agree Elm will generate $240 million of next year revenue. One expects $48 million of operating profit and the other expects $36 million. Their implied operating margins are 20% and 15%. Rather than averaging profit to $42 million, ask which expense assumptions explain the $12 million difference. An average creates a third forecast without explaining why that forecast deserves attention.

Assume both models include $144 million of direct costs. That leaves $96 million before other operating expenses. The optimistic model therefore assumes $48 million of those expenses; the cautious model assumes $60 million. This is the disagreement worth examining. If the extra $12 million is described as customer support, request evidence about service staffing, customer complexity and contract obligations. If it is described as product development, the relevant questions change.

Keep the comparison conditional. At the common revenue assumption, every additional $1 million of operating expense reduces operating profit by $1 million in this simplified model. That arithmetic is not evidence that expenses will rise. It merely makes the consequences visible. A useful research note can state that the forecasts disagree primarily about support costs, that public data do not isolate those costs, and that the next informative disclosure would concern staffing or service intensity. This preserves an actionable question without manufacturing confidence in either forecast.

For an original editorial exercise, imagine five research notes repeating that Elm's customer spending is accelerating. Trace each note backward. Three cite the same management interview, one cites one of those three notes, and only one examines a separately disclosed customer table. Counting five favorable opinions would exaggerate the diversity of evidence. There may be two evidence paths, with different limitations, rather than five independent confirmations.

Draw a small dependency map in your notes: claim, immediate reference, original observation. This is not a credibility ranking of the authors. A careful analyst may reasonably use management commentary, but multiple repetitions do not convert that commentary into multiple observations. Conversely, two analysts can reach different conclusions from the same table because they select different comparison periods. Source overlap and analytical disagreement are separate issues.

The map also helps identify an efficient next step. If every claim depends on a phrase such as strong engagement, seek a defined measure rather than another quotation. If the underlying table exists, inspect whether it measures customer count, spending or activity. Where the original observation cannot be located, label the claim unverified and avoid using its repetition as support. This is especially useful when a persuasive narrative appears across several outlets on the same day: breadth of distribution alone does not answer the substantive research question.

Prepare one row for each disputed claim, with spaces for the exact statement, entity, metric, period, units, accounting basis, source location and information date. Add a reconciliation column for disclosed adjustments and another for unresolved assumptions. Keep quotations brief; the worksheet needs enough wording to preserve meaning, not a reproduction of the source. Include your own recalculation so a later reader can distinguish the published figure from your derived result.

Set an editorial stopping rule before searching further. For example, in a hypothetical exercise with a $12 million forecast disagreement, first investigate explanations large enough to account for several million dollars. A rounding difference of $50,000 cannot plausibly reconcile that gap. This is a prioritization device, not a universal materiality threshold. A small amount could still matter if it reveals a systematic definition error across many observations.

End the row with one of three next actions: obtain a specific missing document, wait for a named disclosure, or retain the uncertainty. State what would justify reopening the issue. Without that condition, the same disagreement can consume attention indefinitely. A completed worksheet should let someone else understand both the work already done and the evidence still missing. It should also make clear when a dispute has become impossible to resolve from public information rather than merely inconvenient to investigate.

Imagine Elm's support spending rises while its customer base also becomes more complex. The higher expense could reflect temporary training, a lasting service burden, or both. One later result may not distinguish these mechanisms. Resist turning an ambiguous observation into a decisive test after seeing it. Specify beforehand what a more persuasive sequence would look like, such as several comparable periods of customer growth accompanied by declining support cost per customer under unchanged definitions.

Even that proposed test has limits. Customer complexity can change, cost allocation can move, and averages can conceal expensive subgroups. Record these alternative explanations next to the test. This does not make research pointless; it identifies what the observation can and cannot establish. The aim is a narrower uncertainty, not an impossible standard of complete knowledge.

If both interpretations survive, write two conditional conclusions rather than one diluted narrative. Under a temporary cost explanation, the model requires expenses to flatten as revenue grows. Under a structural cost explanation, higher revenue may bring proportionate support spending. Explain the consequence in each case and leave likelihood unassigned when evidence is insufficient. The reader then sees why credible analysts can continue to disagree after checking the same arithmetic. That is a more useful outcome than choosing whichever author sounds most confident or whichever conclusion happens to match an existing preference.

A common trap is collecting more opinions after the factual disagreement has already been resolved. Once you know that 15% reported growth and 8% comparable growth use different bases, another ten commentaries are unlikely to improve the arithmetic. The next useful step is to examine the business question those numbers raise.

Another trap is false balance. If one source uses a demonstrably wrong denominator, it does not deserve equal weight merely because it sounds credible. Explain the error and use the supported calculation. Respectful treatment of authors does not require equal treatment of incompatible evidence.

The tradeoff is thoroughness against attention. Continue when an unresolved issue could materially change your understanding and there is a realistic path to evidence. Stop when the remaining uncertainty is inherently future-dependent or undisclosed. A precise research note can contain disagreement. What matters is that the reader can see exactly where the facts end and the assumptions begin.