A Verified Skills Protocol for the Post-AI Economy

There is a quiet emergency unfolding across the global labor market, and almost no one is talking about it correctly.

The conversation about AI and jobs has centered on displacement: which roles will be automated, how many jobs will be lost, what new jobs will emerge. That framing misses the deeper crisis, which is that the system we use to connect people to jobs is collapsing underneath the displacement question itself.

Consider what happens when a worker loses a role to automation today. They possess skills: real, hard-won capabilities acquired over years of practice. When they enter the job market, they discover that those skills are illegible to the people who could hire for them. Their resume describes job titles rather than competencies, and their certifications, if they have any, are snapshots from a moment in time that may no longer reflect their current ability. Their most valuable skills, the ones learned on the job through experimentation and through solving problems no training program anticipated, often have no documentation at all.

Everybody has hidden skills their friends, family, and coworkers see every day. Yet the current hiring system cannot surface them.

The problem is getting worse as AI compresses the cycle time between “skill acquired” and “skill obsolete.” A certification earned eighteen months ago may already be outdated, and a degree completed five years ago says almost nothing about a person's current capability. The institutions that issue credentials cannot move fast enough to keep pace with the skills that actually matter in the market.

On the demand side, employers face a mirror image of the same crisis. They need specific, verified capabilities, not proxies for capability. A hiring manager looking for someone who can manage an AI-augmented sales pipeline does not need to see a bachelor's degree in business. They need proof that this person can do this specific thing, assessed against a standard they can trust.

The market is screaming for a new trust layer, and the existing solutions cannot provide one.

The first generation of digital credentialing platforms (Credly, Accredible, Badgr, and others) represented a genuine attempt to modernize how skills are documented. They introduced digital badges, machine-readable metadata, and the Open Badge Standard. They were right about the problem they were trying to solve, and wrong about the architecture they used to solve it.

These platforms suffer from three structural flaws that prevent them from becoming the trust layer the market needs.

The portability problem. Digital badges issued on Credly live on Credly. They can be shared as images or links, but they cannot be meaningfully composed, compared, or queried outside the platform that issued them. A worker with badges from three different platforms has three isolated data points, not a coherent skill profile. The credentials are technically portable but practically stranded.

The coverage gap. Credentialing platforms are only as comprehensive as their issuer network. If an employer, training provider, or university has not signed up and actively issues badges through the platform, those skills do not exist in the system. This creates an inverted reality: the most institutionally credentialed workers are the most visible, while the workers who most need visibility (the self-taught, the career changers, the gig workers) remain invisible.

The comparability problem. This is the deepest flaw of the three. A “Sales Skills” badge from Company A and a “Sales Proficiency” certification from Organization B are, in the current system, incommensurable: there is no mechanism to assess whether they represent equivalent capability, overlapping capability, or entirely different skill sets. Without cross-ranking, without the ability to normalize credentials from heterogeneous sources into a unified assessment, digital badges become digital stickers that look official without proving much of anything.

What is needed is an intelligence layer that sits above all credentials, ingests evidence from any source, and produces a unified, verified, dynamically updating assessment of what a person can do. That intelligence layer rests on a common spine: an open specification for what skills exist, what evidence verifies them, and how proficiency is measured.

Humanproof is built on three components: an ingestion layer, an assessment engine, and a publication layer. They share a common spine: the Humanproof Skill Standard (HSS), an open specification for what skills exist, what evidence verifies them, and how proficiency is measured. Together they transform fragmented, static, platform-locked credentials into a unified, dynamic, portable proof of human capability.

3.1 The Humanproof Skill Standard

The Humanproof Skill Standard is the open specification at the center of the protocol, a structured ontology of atomic skills. Each skill record declares the evidence types that verify it (with authority weights from 0.0 to 1.0), the decay half-life that determines how recency affects confidence, the dependencies and related skills that connect the graph, and the proficiency rubric that defines what novice, intermediate, advanced, and expert mean for that specific skill.

Every assessment engine reads these records directly into its scoring computation. The same evidence under the same engine version produces the same score, anywhere in the world, deterministically.

The standard organizes skills into five top-level domains:

• Technical. Engineering, computer science, data, finance, hard sciences, quantitative disciplines.
• Creative. Design, brand, copy, music, film, visual arts, performing arts.
• Interpersonal. Leadership, mentoring, negotiation, communication, facilitation.
• Domain. Profession-specific bodies of knowledge: medicine, law, education, manufacturing, hospitality.
• Foundational. Literacy, numeracy, critical thinking, research methodology.

The number of domains is deliberate, reflecting how human capability actually divides at the highest level of abstraction. Existing taxonomies tend to be skewed toward the institutions that built them: ESCO is European-labor-heavy, O*NET is US-occupation-heavy, and LinkedIn is technical- and recruiter-heavy. The HSS corrects that imbalance and treats brand identity systems with the same structural depth as React, and a senior nurse's clinical judgment with the same depth as either.

The standard is published under the Creative Commons Attribution-ShareAlike 4.0 license. Anyone can use it, build commercial products on top of it, or implement their own conformant engine. The reference implementation is the Humanproof assessment engine, open-sourced under Apache 2.0. Engine binaries are reproducibly built and identified by content hash, so any third party can rebuild from source and verify they hold the same engine that produced any given score.

3.2 The Ingestion Layer

Humanproof begins by accepting evidence from everywhere, by deliberate architectural choice. Rather than building another gated ecosystem that depends on issuers opting in, the protocol treats the entire landscape of human skill documentation as input.

Sources include, but are not limited to:

• Digital badges from any platform (Credly, Accredible, Badgr, custom issuers)
• Formal credentials (degrees, transcripts, professional certifications)
• Resumes and CVs in any format
• Professional portfolio artifacts (code repositories, published work, project documentation)
• Performance reviews and employer assessments
• Peer and manager reviews submitted with verified working relationships
• Training completion records
• Self-reported skill evidence with supporting documentation

The ingestion layer normalizes this heterogeneous input into a structured format. Every piece of evidence is tagged with its source, its recency, its issuer where applicable, and its evidentiary weight: a measure of how much trust the system places in it based on source reliability and specificity. The HSS defines these weights canonically per skill, so every conformant engine treats the same evidence the same way.

The system does not require issuers to participate, because Humanproof can index publicly available credential data at scale and build a comprehensive skills graph without depending on institutional cooperation. The network grows from the edges in rather than from the center out.

3.3 The Assessment Engine

The assessment engine is the reference implementation of the Humanproof Skill Standard and the primary innovation this paper describes.

Existing credential systems are attestation-based: they record that an authority said a person has a skill. Humanproof is evidence-based: it evaluates the totality of available evidence and produces an independent assessment of proficiency.

The engine operates through four stages:

Extraction. Using large language models fine-tuned for skill identification, the engine analyzes raw evidence and extracts discrete skills. A resume that describes “managed a team of 12 engineers building a real-time data pipeline” yields specific skill extractions: team leadership, engineering management, real-time systems, data engineering, pipeline architecture.

Normalization. Extracted skills are mapped to atomic HSS records, whose structured ontology covers technical, creative, interpersonal, domain, and foundational capability with cross-domain balance from version 1.0 onward.

Proficiency Assessment. For each normalized skill, the engine evaluates proficiency across four levels (Novice, Intermediate, Advanced, Expert) based on multiple weighted signals:

• Credential authority weight: A certification from a recognized institution carries more weight than a self-reported claim, but less than a portfolio of demonstrated work. The HSS pins these weights canonically per skill.
• Evidence depth: Multiple independent sources confirming the same skill increase confidence.
• Recency: Skills decay. Recent evidence is weighted more heavily. The decay curve varies by skill category and is encoded in the HSS, not the engine.
• Specificity: A credential that names a precise skill carries more signal than one that names a broad category.
• Cross-validation: When evidence from independent sources corroborates the same skill, confidence compounds. The HSS specifies the compounding function so that all conformant engines produce the same result on the same inputs.

Each skill assessment produces a confidence score between 0 and 100, representing the engine's certainty in its proficiency determination. The score is transparent: users see exactly which evidence contributed to it, how each source was weighted, and why.

3.4 The Publication Layer

The output of the Humanproof protocol is a Humanproof Skill Profile (HSP), a portable, machine-readable document that represents the most comprehensive and current assessment of an individual's capabilities.

The profile is designed to be consumed by three audiences simultaneously:

Humans. The profile renders as a visual, shareable credential, a digital certificate that communicates verification status, top skills, proficiency levels, and the overall Humanproof Score at a glance.

Machines. The profile is encoded in JSON-LD, a structured data format that AI agents and automated hiring systems can parse, query, and reason about. The full evidence trace is selectively disclosable: workers choose which evidence accompanies the profile when they share it.

Verifiers. Every HSP is a signed W3C Verifiable Credential, with Humanproof publishing the signing key, the engine version that produced the score, and the engine binary's content hash. The engine itself is open-sourced under Apache 2.0 and reproducibly buildable from a published source tag, with engine version hashes anchored to Bitcoin via OpenTimestamps. A verifier with no relationship to Humanproof can audit any score by checking the signature, rebuilding the engine from source, and re-deriving the result, so verifiability survives even if Humanproof does not.

The profile belongs to the individual, and it is not locked to a platform, gated by an employer, or dependent on an institution continuing to exist. It is a sovereign record of what a person can do.

We are entering a period of labor market transformation that will be the most disruptive in a century. The last transformation of this magnitude, the shift from agricultural to industrial economies, took generations and was mediated by institutions that had time to adapt. This time the transformation is happening in years rather than generations, and the institutions have not adapted.

The human cost of this institutional failure is already measurable. Workers with real skills are invisible to the market. Career changers are trapped by credentials that do not transfer. Self-taught practitioners are dismissed by systems that only recognize institutional attestation. The people most vulnerable to AI displacement, those in mid-career, without elite educational backgrounds, without access to the networks that enable informal credential transfer, are the people most failed by the current system.

Humanproof exists because this is unacceptable. Every person who has learned something valuable, whether in a classroom, on a job, in a side project, or through sheer determined self-instruction, deserves a way to prove it. They deserve verified, portable, machine-readable proof that they can do what they say they can do, independent of any platform's gatekeeping or any connection's endorsement.

What has been missing is the protocol that connects these capabilities into a coherent system: one that serves the individual first, earns trust through transparency, and scales through network effects.

That protocol is Humanproof.