Measurement & Diagnosis
What Counts as Connected Evidence?
Your facts can be right and still not support your claim. How to check what your evidence establishes, and whether to clarify, find more or narrow the claim.
Your facts can all be right and still not support the claim you’re making.
Take a hypothetical contractor that maintains packaging equipment. A client’s bottling line stops at 2 p.m. on a Tuesday, and the contractor’s work order logs the client’s call at 2:10. Its technicians replace a drive coupling and recalibrate a sensor, and at 6:10 on Thursday morning the line is running again. The contractor’s website later carries a short case study: “When a client’s bottling line went down, we got it running again within 48 hours.”
The records support the timeline and document what the technicians did. The work orders show when the call came in, who was on site, what they replaced and when they signed off, and the client’s production log shows when the line stopped and restarted.
But the sentence says more than either record. “We got it running again” claims that the contractor’s work is the reason the line restarted. The work orders establish who did what. The production log establishes that the line restarted and when. Neither establishes that one caused the other. In the same two days the site’s electricians might have replaced a blown fuse in the control panel, or the equipment manufacturer might have pushed a firmware update. The records document a sequence. The claim describes a cause.
That gap is what this article is about. Evidence is connected to a claim when it establishes the relationship the claim depends on.
The Question to Ask of Any Evidence
Most reviews of evidence ask whether it’s true, current and from a credible source. Every answer can be yes while the claim still fails. A more useful question is this one:
What relationship does this evidence establish, and what would the reader still have to assume?
For the bottling line, the answer comes in parts. That the contractor’s technicians replaced the coupling and recalibrated the sensor is established by the work orders. That the line was running at 6:10 on Thursday is established by the production log. That the work is why it was running is inferred, and whether anything else changed in those two days is unknown. “We got it running again” asks the reader to treat the inference as settled.
Write the Verb
The quickest way to apply the test is to look at the verb, because the verb is where the claim’s relationship lives.
A firm can perform work, contribute to an outcome or cause it. Each verb asks for something different. “Performed” needs a record of the work and of who did it. “Contributed to” needs that, plus something tying the work to the outcome: the fault the work addressed, and some sign that addressing it mattered. “Caused” needs evidence that the work actually produced the outcome, which a sequence alone can’t provide: the repair came before the restart, but so did anything else that changed in those two days. That isn’t the same as showing the work was the only cause. A repair can play a real causal role alongside other factors, and the claim can say so, but only once something beyond the timeline establishes that role.
Sources have verbs too. A trade publication can verify a figure or repeat it, and citing the publication doesn’t tell the reader which it did. A client testimonial can confirm what the client was in a position to see: that the technicians arrived the same afternoon and explained what they’d found. It can’t confirm what caused the restart unless the client knows that too.
This is why a more ambitious verb needs different evidence, not more of the same. Another work order or a photograph of the technicians on site makes “we performed this repair” better documented. It doesn’t move the claim from performed to caused.
Suppose the contractor also has the technician’s fault report. It records that the coupling had sheared, that the line’s controller was logging the sensor out of range, and that the controller fault cleared after the replacement and recalibration. That record adds something specific. It establishes the faults the technicians found, and a change that followed their work: a controller fault present before the repair and cleared after it. It doesn’t establish that those faults stopped the line, that they were the only problems, or that clearing them is why the line was running on Thursday. What it supports is a narrow claim: the technicians found two faults, corrected them, and the controller fault they were addressing cleared after the work. It doesn’t reach the restart, and it doesn’t reach “caused.”
That is often where the strongest claim sits, at a change the record ties closely to the work rather than at the final outcome, which many things can influence. It’s narrower than “we got the line running again,” and much harder to dispute.
So here is the claim the evidence supports:
“Called to a stopped bottling line at 2:10 on a Tuesday afternoon, our technicians found a sheared drive coupling and a sensor reading out of range. They replaced the coupling and recalibrated the sensor, and the controller fault cleared after the work. The client’s production log shows the line back in production at 6:10 on Thursday morning, 40 hours after the call.”
That’s better commercial copy than the original. It tells a buyer what the contractor found, what it did and what changed, with a record behind every clause. The restart is reported in its own sentence, as something the client’s log recorded, without asking the reader to credit the contractor with everything that made it possible. And the clock now runs from a named event. “Within 48 hours” left the reader to guess whether the hours started at the breakdown, the call or the technicians’ arrival.
The Next Buyer Is Asking Something Else
Precise evidence about a past job answers a question about that job. The next buyer usually has a different one.
Imagine a plant manager whose filling line stops for a few minutes several times a shift, with no pattern anyone has found. She doesn’t need someone to replace a part she already knows is broken. She needs someone who can find out what’s wrong. The bottling-line record shows the contractor’s technicians identifying two faults and repairing them. It may say very little about whether its people can diagnose an intermittent problem whose cause isn’t yet known, which is a different skill, demonstrated by different work.
The fault report helps a little, because it shows the technicians identifying more than one fault. But this buyer would want to see the contractor working through an ambiguous fault: what it ruled out and how it isolated the cause. If the contractor has done that work, that’s the evidence to put in front of her. If it hasn’t, the bottling-line story shouldn’t be stretched to cover it.
This is where evidence meets what the business can actually do and what it’s offering this particular buyer. Capability, Proposition and Evidence support each other, but they aren’t interchangeable. Evidence first has to support the claim. Then a buyer has to be able to evaluate it, and it has to bear on the decision she’s actually making. A record of past work is evidence of capability only for the kind of problem it shows the firm solving.
When Three Sources Are One Source
The second way evidence fails to connect is corroboration that isn’t.
Consider a hypothetical adhesives supplier that launches a product rated to hold at 180°C, made at a site that has just gained ISO 9001 certification. Three trade publications cover the launch. Two run the press release with light editing. The third adds a line confirming that the certification appears in the certification body’s public register, but reports the 180°C rating as the supplier’s claim.
A sales deck might cite all three as independent coverage. Read for what each establishes, they support different things. The register check supports the certification: the third publication was in a position to know it, and says how. None of the three tested the temperature rating. All of them rely on the supplier for it, so together they show the figure was widely reported, and nothing more.
Independence has to be judged one assertion at a time. The same article can check one fact, repeat another and quote the supplier’s engineer on a third. A quote shows that the supplier was a source. It doesn’t show that the journalist tested what was said.
That doesn’t make the supplier’s own data weak. A test report from its lab, with the method and conditions stated, may be better evidence for the rating, under the conditions tested, than any of the articles. The useful question isn’t whether a source is first-party or third-party. It’s what the source was in a position to know, and how it knows it.
A second question is easy to skip. A private record can substantiate a claim inside the business without giving an outside buyer anything to evaluate. If the lab report sits in a file, the buyer reading the product page still has only the supplier’s word for the rating.
Two sources can also be useful together without corroborating each other. The register establishes the certification, and the lab report supports the rating under the conditions it tested. Neither confirms the other’s part, and presenting them as if they did overstates both.
What to Change
The test works best on one claim at a time. Pick a sentence from your website or a sales deck and work through it.
Write the claim as a complete sentence. Name the business, the verb and the scope: which work, for whom, under what conditions. Then note what a reasonable buyer would take it to mean. “We reduced energy costs for a regional food producer” will be read as a claim of cause.
Read each source for what it actually establishes. Separate documented activity, a recorded outcome and any inference about contribution or cause. Mark each relationship the claim needs as established, inferred, contradicted or unknown, as with the bottling line. The inferred and unknown ones are what the reader is still being asked to assume.
Choose the repair that fits the gap.
- Make the connection explicit when the evidence already supports the relationship but the page doesn’t show it. The fault report exists, but the case study never mentions it. Put the actor, the role, the timeframe or the relevant finding next to the claim it supports. That doesn’t mean publishing every underlying document. It means stating what the record shows, identifying where it comes from and saying what it establishes. Where the claim carries enough weight, add a checkable extract, an external record or another way for a buyer to verify it. Naming a private record doesn’t make it independently verifiable.
- Find evidence for the missing relationship when it matters and could be established. Be specific about the kind of record that would help. If the gap is who contributed to an outcome, another source confirming the outcome won’t close it, and neither will another testimonial or press mention that doesn’t speak to the contribution. A record of the fault, the client’s account of what changed, or a before-and-after measurement tied to the work might.
- Narrow the claim when the evidence supports less than the wording implies. Treat it as a decision about what the business is prepared to stand behind, and change the claim itself rather than adding a caveat beneath it. As the bottling line shows, the narrower claim is often the more persuasive one, because a buyer can see what it rests on.
Where AI Comes In
None of this began with AI. A careful buyer could always ask what a case study proves. AI-assisted research sharpens a different question: what can be established before anyone talks to the firm?
Go back to the bottling line. The client’s production log is separate from the contractor’s account, and it documents the restart. But it’s the client’s private record, and even read in full it wouldn’t show who caused the restart. The contractor’s revised case study can report what its technicians did accurately and still be, for any outside reader, the contractor’s own account, with nothing independent to check it against.
An outside source can add something, but only within its own reach. Suppose the equipment manufacturer publishes a directory of authorized service providers, and the contractor is listed for that manufacturer’s filling equipment. That confirms the contractor is authorized to service the equipment. It doesn’t verify the repair or the restart, and it says nothing about whether the contractor can diagnose the plant manager’s intermittent problem, even if her line runs the same equipment.
Timing matters here. Once the plant manager contacts the contractor, she can ask to speak to the bottling-line client, or to see a record the client is willing to share. A firm with good records can do well at that stage. Before contact, she, or an assistant answering her question, works from what’s accessible. An assistant can’t rely on a fault report or a production log it can’t see, however strong those records are. The contractor doesn’t have to publish its client’s repair records. It can make available what it’s permitted to share, and be clear about what a buyer will only be able to verify during due diligence. An accessible credential, like the directory listing, can help a buyer check suitability without verifying the historical job. But the evidence that would settle the later conversation may be missing from the earlier comparison.
So a firm can have enough evidence to substantiate its work during due diligence and too little accessible evidence to be evaluated confidently during shortlisting. For any claim that matters, ask what a buyer could establish now, from which source, and what would stay unverified until a conversation.
What is accessible should at least state its relationships plainly. If the case study doesn’t say who did the work, which outcome is being claimed or which record supports it, a reader or a system assembling an answer has to infer those relationships or leave them out. Stating them explicitly removes the need for that inference. It doesn’t guarantee retrieval, citation, correct attribution or recommendation, and none of this assumes that an assistant needs independent corroboration before it will use a claim. Whether those things happen is something to observe, not assume.
Structured data follows the same rule. Markup can describe a relationship, such as an organization providing a service, but it can’t establish a relationship the underlying evidence doesn’t support. What markup can and can’t show is a separate question. This one comes first: whether the evidence supports the claim at all.
Check One Claim
Choose the claim you lean on most to win business. It might be the first line of your capability statement or the case study your sales team sends first. Write down its verb.
Open the records behind that claim. What relationship do they actually establish?
Now look at it from the buyer’s side. What could a buyer establish before contacting you, and from which source? What would they still need to ask you to verify?
If the records support the claim, show buyers what they can check and make clear what requires a conversation. If the records don’t support the claim, find the missing evidence or narrow the wording.
Questions
Frequently asked questions
What makes evidence connected to a claim?
Evidence is connected when it establishes the relationship the claim depends on. A record that the work happened and a record that the result followed are both useful, but a claim that the work produced the result needs something that ties the two together. The practical test is to ask what relationship the evidence establishes and what a reader would still have to assume.
Is a third-party source always stronger than a company's own records?
No. What matters is what the source was in a position to know and how it knows it. A contractor's own fault report can say more about what was wrong with a machine than a trade article paraphrasing a press release. Independent sources add most when they checked something themselves.
If several publications report the same figure, does that corroborate it?
Not by itself. An outlet that independently checked that specific figure adds corroboration for it. Outlets repeating the same original source add none, even when one of them did its own checking on something else in the story. Count the independent checks of the assertion, not the number of articles.
Won't narrowing a claim make a business sound less capable?
Usually the reverse. A claim that names the problem found, the work done and the result recorded gives a buyer something to check, which a broad claim of credit doesn't. Narrowing is the right repair when the broader claim depends on a relationship the evidence can't support.
Does making evidence relationships explicit help AI systems recommend a business?
It can make an accurate answer easier to assemble, because a system doesn't have to infer who did the work or which record supports a claim. It doesn't guarantee retrieval, citation, correct attribution or recommendation. The useful check afterwards is whether answers to your buyers' questions describe the work, and who did it, correctly. Check too whether they distinguish what the company reports from what an accessible source actually verifies, since an assistant can repeat a first-party account without corroborating it.


