What's the difference between a confirmed, an asserted, and an inferred claim?

Confirmed means a locatable source supports it. Asserted means the candidate stated it themselves, which is legitimate but should be labeled as self-reported. Inferred means the research concluded it from other facts, which may be right but is still a guess.

What actually counts as a usable source in candidate research?

Whether a colleague could locate it in under a minute from what is written. A named source with a link is strongest, a named source without one still qualifies, and a described-but-unnamed source like 'public sources' does not count at all.

What should happen when a piece of candidate evidence is genuinely ambiguous?

It should be flagged, not resolved into a guess or silently discarded. The entry should state what was found, what is unclear, and why — flagging is the only option that leaves the reader better off.

Why is a merged identity considered the most damaging error in candidate research?

Because it produces a candidate with a remarkable but false trajectory that reads as fluent and internally consistent, and it survives any review that is not specifically checking whether every fact in the entry belongs to the same person.

Why can't a researcher reliably catch their own sourcing errors?

Because the person who made an inference is the one least able to see it as one — from the inside it feels like a fact, since they remember the reasoning that produced it. A genuine check needs a separate pass that starts cold.

Sourcing Evidence: What "Confirmed" Actually Means

Q: What does 'confirmed' actually mean in a candidate research claim?

A specific, locatable source supports it, and someone who was not part of the research could open that source and see it themselves — not the researcher's memory of having seen it.

Two sentences in a candidate entry:

Led the integration of the acquired European business.

Held the VP Operations role during the period the European acquisition closed.

The first came from a named press release. The second is an inference from a date overlap. In a finished shortlist they sit in the same font, in the same voice, and a reader has no way to tell which is which.

That is the whole problem. Not that researchers guess — good ones guess well, and informed inference is part of the job. The problem is that prose flattens the difference, and the reader ends up trusting the document rather than the evidence. Sourcing discipline is therefore structural, not a matter of care: something has to force the distinction to survive into the output.

Confirmed, asserted, inferred

Three things get written the same way and are not the same thing.

Confirmed means a specific, locatable source supports the claim, and someone who was not part of the research could open that source and see it. Not the researcher’s memory of having seen it. The source.

Asserted means someone stated it — usually the candidate, in their own profile or CV. Self-reported information is legitimate and useful. It is also written by a person with an interest in how it reads, and it should be labeled as what it is.

Inferred means the research concluded it from other facts. “Was in the role when the plant opened” supports “was probably involved.” It does not support “opened the plant.” The inference may well be right. It is still an inference, and it should say so.

All three can appear in a shortlist. What cannot happen is all three appearing as the same thing.

What counts as a source

The bar is operational: could a colleague locate this in under a minute, from what is written?

That single test does most of the work, because it rules out the phrases that feel like sourcing and are not.

Tier 1 is a named source with a link. A LinkedIn profile at a specific URL. An SEC filing at its EDGAR page. A press release at the newswire URL that carries it. The reader clicks once and is looking at the evidence.

Tier 2 is a named source without a link. “Company X press release, October 2023.” “Crunchbase profile for Jane Smith.” Still verifiable, just slower: the reader knows exactly what to go and find. This is the floor for a confirmed claim, not a lesser class of evidence.

Tier 3 is a described but unnamed source, and it is not a source at all. “Public sources.” “Company website” with no page. “Industry press” with no outlet. “LinkedIn” without a profile. Each is a citation-shaped object nobody can check, and an uncheckable citation is worse than none — it signals verification that did not happen.

The axis is locatability, not prestige: a press release with its URL and the same release without one are both usable, and a “company website” is not, however reputable the company. Worth noting separately from the tier is who wrote it — “per LinkedIn profile” and “per the 2024 annual report” can both be Tier 1, but only one was written by someone with an interest in how it reads.

Why the line is load-bearing here specifically

Executive search runs on a resource that is easy to spend and slow to rebuild: the client’s belief that the document in front of them is accurate. A shortlist gets read closely by someone who knows the market. They will recognize a name and check one claim. If that claim turns out to be an inference wearing the clothes of a fact, the cost is not that one entry. Every other claim now has to be independently verified, which is exactly the work the shortlist existed to do.

It is also why length should follow the evidence: five candidates who hold up beat eight where three rest on inference.

Flag ambiguity; do not resolve it and do not discard it

Titles that match with no scope detail. Dates that overlap oddly. Two sources that disagree about when someone left. This is why a usable rubric needs a third state, not just two.

There are three things you can do with it, and only one of them is right.

Resolve it — pick the reading that seems most likely and write it as fact. Cheapest, and it converts a known unknown into an invisible risk.

Discard it — drop the claim, and often the candidate. Looks rigorous. Quietly removes people whose public record is thin, which says nothing about whether they can do the job.

Flag it — record what was found, what is unclear, and why. The entry says: title confirmed, scope not established, two sources disagree on end date.

Only the third leaves the reader better off, and it costs a sentence. The same logic applies to criteria. Each requirement should be able to come back confirmed with a source, unconfirmed because nothing was found, or flagged with the ambiguity named. Collapsing three states into two is where sourcing discipline quietly dies. Screening against that three-state rubric is what makes the standard operational rather than aspirational.

Mis-attribution is the failure that does the most damage

The worst outcome in candidate research is not a missing candidate. It is a confidently wrong one.

Two executives with the same name in the same sector, and a research pass that merges their records. The result is a candidate with a remarkable trajectory who does not exist. It is fluent, detailed, internally consistent, and it survives every review that is not specifically checking whether the facts belong to the same person.

It is also the failure mode that gets worse, not better, with more automation: a system optimizing for a coherent narrative will prefer the merged record to two partial ones.

Adversarial review is what holds the standard

A sourcing standard that depends on the researcher checking their own work does not survive contact with a busy week.

The reason is not diligence. The person who made an inference is the one least able to see it as an inference: from the inside it feels like a fact, because they remember the reasoning that produced it. Reviewing your own work in the same sitting inherits every assumption that produced the errors.

What works is a separate pass that starts cold and goes looking for specific failures rather than reading for quality:

Claims cited to a source that does not actually support them.
Facts about two people combined into one record.
Inference written in the voice of confirmation.
Sources too vague to locate.
Criteria marked confirmed with no citation attached.

And then the part that matters most over time: log what the review catches. That pattern is the only honest measure of whether the process is improving, and a better roadmap than any amount of intention. A catch rate falling over time means the upstream work got better. One that never moves off zero deserves suspicion — not because a clean search is impossible, but because it reads identically whether the review is working or not running at all.

That is why Cerna runs its QA as a separate adversarial pass rather than a final read-through — and why the shortlist you receive has already been argued with before it reaches you.