The indistinguishable paragraph
Read two paragraphs about the same company — one from a research process that took four weeks, one written confidently in twenty minutes — and you will usually be unable to tell them apart. Both will name the business, cite a growth rate, gesture at a competitive advantage and conclude. The prose gives nothing away. Confidence is free.
This is the central practical problem for anybody consuming investment research: the output does not reveal the process. So the only reliable approach is to stop evaluating conclusions and start evaluating the machinery that produced them.
You cannot audit a conclusion. You can only audit the process that would have been capable of reaching a different one.
Four structural tests
There are four questions that separate research from opinion, and none of them are about the investment. They are all about the work.
- 1
Was the process fixed before the conclusion?
A defined sequence of filters, applied in the same order to every candidate, decided before anyone knew which names would survive. If the criteria were assembled after the favourite was chosen, they are a justification, not a method.
- 2
Are the rejections recorded and named?
This is the single highest-signal test. A research programme that publishes only what it liked has told you nothing about its discrimination. One that names the popular company it rejected, and the specific reason, has demonstrated that the filter has teeth.
- 3
Is the claim falsifiable?
'This is a high-quality compounder' cannot be wrong. 'Return on capital stays above X while the debt-to-reserves ratio stays below Y, and if it does not by the next two annual reports, the thesis is broken' can be. Only the second is a research claim.
- 4
Can you get to the source?
Every number should terminate in a document you could open yourself — an annual report, an exchange filing, a regulator's disclosure. A number whose provenance is 'the industry expects' is not evidence; it is a rumour with a decimal point.
The evidence hierarchy
Not all inputs are equal, and most analytical failure is a failure of ranking rather than of arithmetic. A working hierarchy, strongest first:
| Tier | Source | What it is worth |
|---|---|---|
| 1 | Audited financial statements and regulatory filings | The highest-quality input available, and still backward-looking. Signed by someone with legal exposure. |
| 2 | Disclosed quantitative data — shareholding patterns, exchange data, published price history | Reliable as fact, but silent about cause. Tells you what happened, never why. |
| 3 | Management commentary and guidance | Genuinely informative about intent, and structurally optimistic. Useful as a hypothesis; never as a premise. |
| 4 | Third-party estimates and sell-side models | Someone else's assumptions, arriving with the authority of a spreadsheet. Worth reading for the assumptions, not the output. |
| 5 | Narrative, theme, sentiment, 'everyone knows' | Occasionally correct, never evidence. This tier is where almost all expensive mistakes originate. |
What a filter with teeth looks like
The abstract version of all this is easy to agree with and hard to apply. It is more useful to look at what an ordered, pre-committed filter actually does to a universe of candidates — including the parts that feel uncomfortable.
Prospera's own published Portfolio Intelligence methodology is explicit about this sequence, and about its casualties. Companies are required to have retrievable data before anything else is considered — which means genuinely good businesses were excluded purely because they had not been listed long enough, and the methodology says so rather than quietly dropping them. Candidates must have beaten a relevant benchmark over the measurement period, on the reasoning that a company unable to beat a cheap index fund has not earned a place in an actively researched portfolio. Accounts are read in detail. Valuation is a standalone veto, and well-known names were rejected on it. Across the programme, 96 companies were examined at that level of depth.
The conflict you should assume is there
Finally, read everything — including this — with the structural question in front of you: who is paid, by whom, for this conclusion existing?
- Research attached to a product has an interest in the product being owned.
- Research attached to a transaction has an interest in the transaction happening.
- Research attached to an audience has an interest in being interesting, which is a subtler and more corrosive bias than either of the above.
- Research attached to nothing at all is rare, and usually also unresourced.
None of these disqualify a piece of work. Conflicts are not disqualifying; undisclosed conflicts are. The reasonable standard is that the incentive is stated, the process is fixed, the rejections are published and the claim is falsifiable. Anything meeting all four is research, whatever its source. Anything meeting none of them is opinion, however institutional the letterhead.
The practical upshot is a habit rather than a checklist. When you next read a compelling investment case, do not ask whether you agree. Ask what the author would have to see to change their mind, and whether they have ever written that down. If the answer is nothing, you have been reading advocacy — which may still be correct, but should be priced accordingly.
Important information
This article is general information and investment education. It is not investment advice, a research recommendation, or an offer or solicitation to buy or sell any security, fund or instrument, and it does not take account of the objectives, financial situation or particular needs of any individual reader. Illustrative figures are used to explain a concept and are not forecasts or performance records. Investments are subject to market risk, including the possible loss of capital. Consider your own circumstances, and where appropriate take professional advice, before acting on anything you read here.