Make evidence of real work the thing a second employer can act on.
Fromcredential→Toreusable proof
A credential records that something happened. Evidence of work records what someone did, under what conditions and how well. Today most of that evidence stays with the employer that saw it, so the next employer verifies from scratch. The first economic object of an open human-capital network is proof that stays useful after the transaction that created it. It sits close to real work, carries its context, is verified by someone accountable, and is accepted by a second buyer without starting over.
What a buyer can actually act on
A course certificate shows that someone completed learning. A credential shows that an institution made a claim. A résumé shows what a person says they did, and a reference shows what someone else observed. A buyer deciding whether to hire needs something closer to the work: what the person actually did.
The best-studied evidence points the same way. Work samples, tasks that mirror the job, have strong validity, and applicants tend to see them as fair. Differences between groups are generally small, depending on the competencies being assessed. Work samples are also costly to build and run, which is why they pay off most when competencies must be present from the first day [1].
The largest study of digital skill credentials sharpens the point. It covered 46,791 workers over nine years on one of the largest online labour platforms. Earning a skill-test credential raised earnings by about 9% over the following two weeks, mainly by raising the value of projects won rather than their number [2]. The authors attribute the effect to lower employer uncertainty, not higher productivity. It was smaller for experienced workers, whose verified work history and client feedback already did the signalling. The benefit went only to those who scored near the top, helped only in the credential’s own category of work, and was indistinguishable from zero after about a dozen credentials. Credentials helped little with the hardest step, winning a first project with no history. Tests can show hard skills, but they say little about reliability, communication and trustworthiness, which references convey better [2].
A credential records that something happened. Capability evidence records what the person demonstrated.
Two earlier findings, summarised in the same study, bear directly on this architecture. When a researcher randomly hired inexperienced workers and gave them feedback, those workers went on to earn considerably more, because the feedback informed later employers. And standardised information about past work disproportionately helped applicants from less-developed countries, who otherwise faced more employer uncertainty [2]. Information created in one transaction had value in the next.
Evidence has to carry its context
A capability claim without context quickly becomes another label. Employers asked about digital credentials say they want the competencies behind a credential, a reason to trust it and the level of proficiency it represents, all presented simply. They still lean heavily on the reputation of the provider behind it [3]. In a European employer survey, the factors that most built trust in short credentials were skills defined together with industry, relevance to skills employers need, and recognition or accreditation of the provider. Only 4% of employers picked access to outcome data [4].
Trust flows from who stands behind the evidence and how close it sits to real work standards, more than from the amount of data attached. A useful evidence record answers a handful of questions. What was the task, and what did the person do? Under what conditions, and with how much supervision? Who verified it, against which rubric? When did it happen, and what followed?
Evidence recordillustrative
- Capability claim
- Designs and runs regression tests for a web checkout flow
- Task
- Paid project, 3 weeks, retail client (consortium member)
- Conditions
- Team of 4; own scope: test plan and automation; weekly review
- Output
- Accepted; 2 rework cycles; 11 defects found before release
- Verification
- Rubric QA-2 (family standard), scored by an independent assessor; client attests acceptance and dates
- Level
- Independent on routine flows; supervised on payment edge cases
- Validity
- 24 months; refreshed by later work
- Afterwards
- Client re-engaged for a second release
- Disclosure
- Held by the worker; client named only with the worker’s consent
Open standards for digital badges, verifiable credentials and comprehensive learner records can already carry a record like this between systems [5]. Moving it is the solved part. Agreeing on what it means is not (Shift 9).
Evidence should be created as close as possible to the economic activity in which it is meant to be trusted.
Go deeper
From proof to reusable proof
The economic breakthrough isn’t better evidence. It is evidence that gets reused.
- Employer A needs a task done
- Worker performs it
- Performance is verified
- Worker holds the evidence
- Employer B evaluates it
- Employer B skips part of its own screening
Employers already produce this kind of evidence; they just keep it. On online platforms, some employers screen workers by hiring them into small paid test projects before larger engagements [2]. That evidence informs one hiring decision and then stops. The proposal is to let it inform the next one too. The review locates the potential surplus in fewer repeated technical tests, shorter screening, fewer mismatches and less early attrition. It is equally clear that this value hasn’t yet been shown for a general-purpose capability record.
There is an equity reason to care as well. Referral networks are one of the oldest ways of carrying trusted information about a worker, and studies consistently find that 30–50% of workers found their current job through one. Referred hires tend to stay longer. But networks follow social lines, so they can widen gaps between groups with little contact, and the research recommends building new mechanisms for the information problems that referrals currently solve [6].
Portable evidence could give people without networks some of what referrals give people with them.
Who stands behind the evidence
Verification can come from an independent assessor, from the employer or client who received the work, from a trainer, or from an intermediary. Each has a cost and a weakness. Employers are the natural witnesses to work, but they have good reasons not to become public verifiers: legal exposure, skills frameworks that don’t line up across companies, no clear return, and reluctance to share information competitors could use [3].
That suggests a practical division. Paid assessors and intermediaries verify against a shared rubric. Employers and clients attest only narrow facts they can safely confirm, such as acceptance, dates and scope, and richer attestation is optional.
The concept goes further and links an assessor’s reputation to how well its assessments predict later performance [CN]. That is worth building toward. But the review found no mature example of pricing assessors by predictive validity, and attribution is hard, because later performance also depends on selection, training, management, task assignment and workplace conditions. Even inside firms, performance is poorly measured [7]. The first version should pay for verification that saves repeated testing, not for prediction.
Evidence decays, and can be gamed
Capability changes, tools change and evidence goes stale, so records need dates and refresh rules. Standardisation cuts both ways: too much strips out the context that makes evidence meaningful, and too little destroys comparability. More assessments can mean credential inflation rather than better information. Self-dealing can manufacture evidence, so evidence from a narrow set of counterparties should count for less, and a sample of verifications should be checked again.
Not our invention
Work samples and simulations are an established assessment method [1]. Online labour platforms combine tests, ratings and work histories inside one marketplace [2]. Industry certifications assessed against performance standards are trusted because they sit close to work [4]. Learning-and-employment records and verifiable credentials move evidence between systems [5]. The contribution here is the combination: evidence of real work, verified against a shared rubric, carried by the worker, and reused by employers who don’t share a platform.
What could go wrong
- No reuse.Employers may value the evidence and still repeat their own tests, leaving a better résumé rather than infrastructure.
- Verification cost.Producing and scoring evidence of work may cost more than the screening it saves [1].
- Context versus privacy.The details that make evidence readable, such as the client’s name, the rate or the team, may be exactly what workers want to keep private.
- Inflation and gaming.More assessments can produce more credentials without more trust.
- The first-job trap.People with no work history have no evidence of work, and test credentials barely help them win a first engagement [2]. Shift 6 returns to this.
How we’d test it
| The test | What we would measure | What would end the idea |
|---|---|---|
| Evidence increment (Experiment 1) | Hiring outcomes and costs when profiles add verified evidence of work, against the employer’s usual workflow | No gain over a conventional work sample once the cost of the evidence is counted |
| Reuse rate | Share of evidence records a second employer accepts without re-testing | Second employers re-test in most cases |
| Predictive validity (Experiment 4) | Out-of-sample prediction of later performance beyond résumé, reference and interview; fairness across groups; calibration | No added predictive power, or validity can’t be separated from selection and context |
How strong is the evidence?
- Established
- Work samples that mirror the job are valid but costly [1]; for experienced workers, verified work history is a stronger signal than skill-test credentials [2]
- Evidence-supported
- Credentials reduce employer uncertainty for early-career workers with some history [2]; feedback from a first job raises later earnings by informing later employers [2]
- Hypothesis
- Evidence verified once can be reused by other employers at lower total cost than repeated screening
- Unknown
- Whether employers will pay for reusable evidence, and whether assessors can be rewarded for predictive validity
Sources
- [1]U.S. Office of Personnel Management, Work Samples and SimulationsChecked
- [2]Kässi and Lehdonvirta, Do Microcredentials Help New Workers Enter the Market? Evidence from an Online Labor Platform, Journal of Human Resources 59(4) (2024)Checked; the earlier studies it summarises (Pallais 2014; Agrawal, Lacetera and Lyons 2016; Corporaal and Lehdonvirta 2017) not re-checked
- [3]Aspen Institute, UpSkill America, report on employer perspectives on digital credentials and skills profiles (2024)Checked
- [4]Cedefop, Microcredentials for labour market education and training: the added value for end users (2023)Checked (executive summary and chapters 1–3)
- [5]National Governors Association, resource page on learning and employment recordsChecked
- [6]Schmutte, How do social networks affect labor markets?, IZA World of Labor (2016)Checked
- [7]Broecke, Artificial Intelligence and Labour Market Matching, OECD Social, Employment and Migration Working Papers (2023)Checked
- [CN]Open Commerce for Human Capital: concept document consolidated from the discussion (internal)Provided
Spotted a source we missed or a rule we misread? Tell us what we got wrong.