What counts as evidence that a candidate can do the job?

The strongest evidence that someone can do a job is watching them do a piece of it, or hearing a specific, checkable account of having done it before. The weakest is how they describe themselves. Between those sit job titles, responsibilities and credentials, which tell you what someone was exposed to, not how well they did it.

Most hiring mistakes come from treating all of these as equal: counting “detail-oriented” or “five years in accounting” as if it were the same kind of information as a reconciliation they walked you through. This article gives you a simple way to tell them apart.

A ladder of evidence

This ladder is our framework, built on the research below. It runs from strongest to weakest:

Level	Kind of evidence	Example for a Senior Accountant	Where you get it
1	Doing the work now	Reconciles a sample account in a 45-minute exercise	Work sample, job knowledge test
2	Specific, checkable accomplishment	“Ran the close for a 60-person distributor; cut it from nine days to five”	Application, structured interview
3	Responsibility held	“Responsible for month-end close”	Resume
4	Proxy	“Seven years of accounting experience”; a degree; a well-known employer	Resume
5	Self-description	“Detail-oriented, strong communicator, team player”	Resume, cover letter

Each level up costs more to collect. That’s why the first screen relies on levels 2 and 3, and why the later stages should move up to level 1.

What the research says

A 2022 re-analysis of selection research by Sackett, Zhang, Berry and Lievens estimated how well different methods predict job performance, where 0 means no relationship:

Method	Estimated validity	Ladder level
Structured interview	.42	1–2
Job knowledge test	.40	1
Work sample test	.33	1
Unstructured interview	.19	Mixed
Years of experience	.07	4

The pattern matches the ladder. Methods that observe the work, or ask everyone about specific past behaviour in the same way, predict better than counting years. Two cautions: these are averages across many jobs and studies, and even the best method leaves a great deal unexplained. No evidence makes a hire certain. Stronger evidence makes it less of a guess.

Newer methods, such as asynchronous video interviews and tests scored by AI, are still being studied. A 2026 review of the field by Sackett, Lievens and Landers covers them alongside the established methods; until the evidence on a new method is as strong, treat it with more caution than the ones in the table.

Self-descriptions sit at the bottom for a simple reason: everyone writes them, and writing them costs nothing. That’s even more true now that many applications are drafted with AI, which makes polished, confident wording universal. Our article on AI-written applications covers how to screen when every resume reads well.

How to read an application for evidence

You can’t run a work sample on 200 applicants, so the first read has to find level 2 and 3 evidence quickly. Look for:

• Ownership. “Ran”, “prepared”, “signed off” and “built” describe someone doing the work. “Assisted with”, “exposure to” and “involved in” usually describe someone near it. Neither is disqualifying, but they aren’t the same.
• Scale and context. “Ran payroll” for 8 people and for 800 are different jobs. Look for the size of the team, the budget, the client or the volume.
• A result. A before-and-after, a deadline met, a problem solved. Specific results are easier to check and harder to invent.
• Recency. Something done last year counts for more than something done a decade ago, unless the skill hasn’t changed.
• Consistency. Dates, titles and claims that agree across the resume, the application answers and later conversations. Inconsistency isn’t proof of anything, but it’s worth a question.

A worked example

Larkspur Partners, a fictional accounting firm used throughout our guides, has a must-have: has run a month-end close independently. Here are four invented lines from four applications, and how we’d read them:

What the application says	Level	Reading
“Owned month-end close for three entities; reduced close from 9 to 5 days by automating accruals”	2	Meets it, with strong evidence
“Responsible for month-end close and reporting”	3	Probably meets it; confirm they ran it rather than contributed
“Assisted senior accountant with month-end close”	3	Doesn’t show it yet; worth a question if otherwise strong
“8 years in accounting; highly organised and detail-oriented”	4–5	Unclear: nothing here addresses the requirement

The fourth applicant may well have run a close many times. The application just doesn’t say so, which makes it unclear, not a fail. A one-line question settles it, and is far cheaper than losing a good accountant to a badly written resume.

“What does this candidate bring?”

Requirements tell you whether someone can do the job as it is. Sometimes you also want to know what they’d add: a skill the team doesn’t have, a different way of working, or the ability to grow into a bigger role. That’s a fair thing to look for, as long as you treat it as evidence too.

• Name it in advance. “Has set up a process from scratch” or “has trained others” is assessable. “Brings something special” isn’t, and an undefined criterion fills itself with personal preference. Our guide to hiring for your team shows how to work out what your team is missing.
• Look for evidence of learning, not claims of it. “Moved from audit to industry and was running the close within four months” is level 2 evidence of learning quickly. “Fast learner” is level 5.
• Be careful with “potential”. Cognitive ability tests predict performance reasonably well (an estimated .31 in the 2022 re-analysis), but the same paper reports a large average Black–White difference in scores on them, which raises fairness and legal questions. If you assess potential, use the same structured questions or exercise for everyone, and don’t let it override the must-haves.

When the evidence is unclear

Unclear is an answer. It means you don’t know yet. The worst thing to do with it is decide silently, in either direction. Instead:

1. Ask one specific question before shortlisting: “Did you run the close yourself, or support someone who did?”
2. Test it later, if the applicant is otherwise strong: a work sample or a structured interview question aimed at that requirement.
3. Write down what you found, so the next person doesn’t have to ask again.

Where SortTheStack fits

SortTheStack reads each application against your requirements and marks every one met, not met or unclear, quoting the lines it relied on, so you can check the evidence yourself. It doesn’t produce a score. Inconsistencies, such as overlapping full-time jobs, are flagged under Profile consistency for you to ask about. See how the screening works.

Sources

• Sackett, P. R., Zhang, C., Berry, C. M., & Lievens, F. (2022). Revisiting meta-analytic estimates of validity in personnel selection. Journal of Applied Psychology, 107(11), 2040–2068. doi.org/10.1037/apl0000994
• Sackett, P. R., Lievens, F., & Landers, R. N. (2026). Hiring people in organizations: The state and future of the science. Annual Review of Organizational Psychology and Organizational Behavior, 13, 49–75. doi.org/10.1146/annurev-orgpsych-020924-072127