Most companies hiring at the entry level make their decisions on four signals: which university the candidate attended, what their GPA was, how they performed in a behavioural interview, and whether they have a set of certificates a recruiter recognises. Three of those four were not designed to predict job performance. None of the four predict it well. Hiring teams know this — the conversation about skills-based hiring has been running for at least a decade — but the structural fix has been slow, because the alternatives have been imperfect and the cost of changing the funnel has been high.
Verifiable evidence of work has changed what is possible. Not what is theoretically possible — what is operationally feasible at the volume early-career funnels run at. This article is about the four assumptions that are still load-bearing in most companies' early-career hiring, what each one actually predicts, what better signals exist, and a three-step fix that any hiring team can begin this quarter without rewriting their applicant tracking system.
The four broken assumptions
Pedigree
The first assumption is that the prestige of the candidate's university predicts how good they will be at the job. The shape of this assumption shows up in the hiring funnel as a target-school list, a recruiter focus on certain campuses, an unconscious tendency to read certain CVs more closely.
What pedigree actually predicts, in early-career hiring, is who applied to and was admitted to a particular university four years ago — most often when the candidate was eighteen. That admission decision was made on a different combination of signals (test scores, extracurriculars, essays, in some countries family connections) and was filtered through the institution's specific selection priorities at that moment. Those priorities are loosely correlated with what makes someone good at a junior consulting role, a junior engineering role, or a junior product role. Loose correlation is not nothing. It is also not the strongest predictor available.
The structural problem is selection bias. If you only hire from target schools, your data set on who succeeds in your roles only contains people from target schools. The model self-confirms. Companies that have run controlled experiments on hiring outside the target list — most major technology firms in the last fifteen years — have found that the non-target hires perform within the same distribution, often slightly better at the median, when controlled for the other selection signals.
GPA
The second assumption is that GPA — within and across universities — predicts job performance. The shape of this assumption is a 3.5 cutoff on application forms, a recruiter heuristic of skipping CVs with lower GPAs, an interview hour spent discussing transcripts.
What GPA actually predicts is conscientiousness in a structured academic environment. That is a real trait, useful for some roles. The problem is that GPA also captures: the difficulty of the courses the student took, the grading culture of the specific institution, the student's life circumstances during their degree, and the field of study (engineering grades and humanities grades are not interchangeable). The signal-to-noise ratio is low.
The further problem is that GPA does not measure what most early-career roles actually require. Most entry-level work is not solving the kind of problems that show up on a graded examination. It is producing reviewable artefacts under deadline, communicating clearly, navigating ambiguity, and finishing what was started. None of those map onto exam performance reliably. A candidate with a 4.0 in a programme that grades primarily on closed-book exams may struggle with a junior role that demands open-ended written deliverables.
Behavioural interviews
The third assumption is that thirty to sixty minutes of structured behavioural interview reveals whether a candidate will do the job well. The shape: STAR-method questions, "tell me about a time", interviewer scoring grids, panel debriefs.
What behavioural interviews actually predict is who is good at behavioural interviews. There is a genuine skill in producing crisp STAR-format answers under pressure, and that skill correlates with some traits that matter at work — verbal communication, composure under scrutiny, ability to structure thinking quickly. Those traits are useful. But the interview format has well-documented inter-rater reliability problems: different interviewers asking the same questions to the same candidate often produce significantly different scores, especially when the rubric is loose. The signal that survives panel calibration is real but narrow.
The deeper issue is what the interview cannot test for. Behavioural interviews ask candidates to describe past work. They do not show the past work. The candidate's verbal account of how they handled a difficult stakeholder is filtered through their memory, their narrative skill, and their incentive to perform well in the room. The interview replaces the artefact with a story about the artefact. Stories are easier to fake than artefacts.
Completion certificates
The fourth assumption is that a list of certificates on a CV — courses completed, programmes finished, badges earned — tells you something about the candidate's skills. The shape: applicant tracking systems with certification fields, recruiter heuristics for trusting certain providers, structured "do you have X certificate" filters.
What completion certificates actually predict is that the candidate spent some hours on a topic and reached the end of a structured course. They do not reliably predict whether the candidate can do the work the certificate is named after. The mismatch comes from two structural facts about most certificate-issuing programmes: the assessment is often light (multiple-choice, completion-based, short take-home), and the certificate travels separately from any artefact the candidate produced. A recruiter sees "Certified Data Analyst" and has to trust both the issuer's standards and the candidate's claim, with nothing reviewable in between.
The newer alternative — verifiable credentials tied to specific work the candidate produced — addresses both of those problems at once. The credential is checkable cryptographically, and the work it attests to is linked from the credential itself. The recruiter does not have to trust the candidate's claim about the certificate; they can read the work.
What we are actually trying to predict
The hiring assumption underneath all four signals is that we are trying to predict job performance, but that framing is loose enough that it lets weak signals survive. The sharper question is: which specific behaviours, in the first six to eighteen months on the job, do we need this candidate to demonstrate?
For a junior analyst, the answer is concrete: produce structured written analyses against deadlines, communicate findings clearly, take feedback and revise without dropping the work, navigate ambiguity in the brief without spinning out. For a junior engineer, similar but adjacent: read existing code, produce changes that work, communicate intent clearly in commits and design docs, ship in small reviewable chunks. For a junior marketer, again similar shape: scope a brief, produce campaign or content artefacts, measure them, learn from the measurement.
The pattern across roles is that early-career performance is overwhelmingly visible in deliverable-shaped work: scoped artefacts produced under realistic constraints, reviewed against criteria a practitioner would apply. If we want to predict who will be good at that, the strongest signal is who has already done it.
This is the part the four broken assumptions miss. Pedigree, GPA, behavioural interviews, and completion certificates all sit upstream of the work. Verified deliverables sit at the work itself. The signal-to-noise advantage of the latter is not subtle — it is the difference between asking a candidate to describe their thinking and reading their thinking on the page.
What signals actually predict early-career performance
Three signals stack up cleanly. None of them require burning down your existing funnel; they are additions, not replacements.
Reviewed work product. A portfolio of artefacts — feasibility studies, design documents, code repositories, marketing analyses, research write-ups — produced under realistic constraints and reviewed by someone qualified to apply the criteria. The strongest version of this signal includes the rubric the work was assessed against, so you can see not just what the candidate produced but how it scored against published criteria. The closer the work resembles the work in the role, the stronger the signal.
Structured technical or skills assessment. A controlled exercise designed by your hiring team, applied uniformly across candidates, scored by people calibrated to the same bar. Coding tests, take-home cases, structured work-sample tasks. The signal is mid-funnel, not top-of-funnel — the cost of running the assessment is significant, so you reserve it for candidates who have already passed the work-product filter.
Demonstrated trajectory. What did the candidate ship in the last twelve months? What did they ship the twelve months before that? Trajectory tells you whether you are looking at someone whose curve is flat (a few impressive line items, mostly four years ago) or whose curve is steep (consistent shipping, recent enough to indicate current capability). Trajectory becomes legible only when there is reviewable work spread across time, not just a CV summary.
The thing to notice about all three signals is that they are artefact-grounded. Each one anchors the hiring decision in something a hiring manager can actually read, rather than something they have to take on faith. The four broken assumptions are faith-based. The three working signals are evidence-based.
How verified work product changes the equation
The structural shift over the last few years is that work product can now be verified at scale. Verifiable credentials issued in the W3C Verifiable Credentials Data Model 2.0 format with Open Badges 3.0 metadata are checkable cryptographically — a recruiter clicks a credential URL and gets a cryptographic confirmation that the artefact attached to it is genuine, was issued by the named issuer, and has not been altered.
The practical consequence: the question of whether to trust the candidate's claim collapses. The credential is either valid or it is not. The work behind the credential is either reviewable at the URL or it is not. There is nothing left for the candidate to inflate, and nothing left for the hiring team to take on faith.
The cost structure changes too. Reviewing the work in a verifiable credential takes seconds — open the URL, read the brief, read the deliverable, read the rubric assessment. A recruiter can review a candidate's full set of credentials in five minutes, where the same depth of evidence-gathering through traditional means would take a panel of interviewers an afternoon.
What this enables, finally, is a hiring funnel where the top-of-funnel filter is grounded in demonstrated work rather than in proxies for it. Pedigree, GPA, behavioural performance, and certificates can stay where they are useful — but they no longer have to carry the entire weight of the early-stage filter, because there is something better to filter on.
A three-step fix that any hiring team can start this quarter
The fix does not require restructuring the hiring process. It requires three additions, each manageable inside a single quarter.
Step one: add a verifiable-credential field to the application form. Optional. Free-text or structured. Any candidate who holds verifiable credentials in their target field can attach them at application time. The change costs nothing operationally and produces immediate signal — the response rate from candidates with verified work tells you, within ninety days, whether your candidate population has caught up to the architecture.
Step two: rewrite the top-of-funnel screen to weight reviewable work alongside pedigree and GPA. Not in place of them, alongside them. A candidate with a strong portfolio of verified deliverables in your target domain advances regardless of their school. A candidate with a 3.4 GPA and three rubric-graded artefacts in your domain advances. The screen gets wider where the work is strong and narrower where the work is absent — exactly the inversion the broken assumptions block.
Step three: sponsor a challenge in your target domain. Through Ewance or any peer platform that issues verifiable credentials. The challenge runs once or twice a year, scoped to the kind of work your entry-level role actually requires. Candidates ship deliverables, the rubric assessments are visible to your team, and the candidates whose work stands out become known names months before they apply for a formal opening. The funnel becomes evidence-fed at the source.
None of these three steps requires changing your applicant tracking system, retiring a specific interviewer, or breaking your existing relationships with target campuses. The existing funnel keeps running. The fix is additive — three small adjustments that introduce evidence-based signal where the funnel currently has only proxies for evidence.
The honest case for moving
The case for fixing early-career hiring is not that your current process is failing every candidate. It is not. The case is that the failure rate is higher than it needs to be, the missed candidates are more numerous than the funnel sees, and the available alternatives have matured to the point where the upgrade is operationally feasible.
Hiring is a closed-loop system: you only hear about the candidates who got through. The candidates filtered out at the top of the funnel never become data in your performance system. The mistakes are invisible by construction. Verified work changes that — the candidates who would have been screened out on pedigree but who shipped strong deliverables become legible to the hiring team in a way they were not before.
The teams that have already moved are not making a moral case. They are making a pragmatic one: better signal at lower cost. The companies running skills-based hiring programmes report wider funnels, stronger first-year retention, and faster time-to-productive-contribution. The numbers vary by industry and methodology, but the direction is consistent.
If you want to see what the candidate-side of this looks like — students working through real challenges and earning the kind of credentials that change the funnel — start at the industry overview, walk through the verifier at /tools/credential-verifier/, or book a demo to talk through how to integrate verifiable credentials into your specific funnel without rewriting it.
The four broken assumptions are not going to fix themselves. The fix is available now. The teams that move first benefit from the candidates whose work is already verified — before the rest of the market catches up.
