Resumes
Self-reported proximity to expertise.
A third party attests to titles and dates — never to whether the underlying skill is real.
Provenant replaces resumes and staged assessments with verifiable, longitudinal evidence of how candidates actually think, build, and reason over time.
CANDIDATE-CONTROLLED · LOCAL-FIRST · SOURCE-VERIFIED
Resumes, coding tests, and personality assessments all assume the candidate works alone, under artificial conditions, for a fixed window. AI quietly broke that assumption — and the signals that depended on it.
Self-reported proximity to expertise.
A third party attests to titles and dates — never to whether the underlying skill is real.
Now measure prompting fluency, not depth.
With AI in the loop, timed coding snapshots no longer separate understanding from performance.
Weak, gameable, and detached from real work.
Marginally predictive at best, and increasingly challenged on bias and legal compliance.
Snapshots can be rehearsed. Longitudinal reasoning is much harder to fake.
Anyone can generate a polished answer in seconds. Far fewer can steer intelligence toward something true — questioning assumptions, recognising a wrong turn, and recovering from it.
That capability is not visible in a single test. It surfaces across months of real work: in the questions asked, the errors caught, the tradeoffs made under pressure.
↳ candidate questioned model output;
recognised failure mode unprompted.
↳ candidate questioned model output;
recognised failure mode unprompted.
The candidate's device is untrusted by design. Evidence is verified against third parties they do not control, and raw history never leaves their machine.
The candidate's agent surfaces evidence. Provenant independently verifies each artifact at source — not on the candidate's machine.
A neutral model assesses depth of reasoning across the verified history — mapped to the specific role and its requirements.
An adaptive examiner asks the candidate to extend and defend their own past reasoning — in real time.
A concrete walkthrough for a single engineer — the launch wedge. Every line is checked against a third party the candidate does not control, then committed to a tamper-evident root before assessment runs. The same model extends to any profession with externally verifiable evidence.
Commits cross-checked against external contributors and merge history — never just the candidate's self-attestation.
Review threads cross-referenced with reviewer accounts the candidate does not control.
Production URLs pinged at fixed addresses to confirm authorship and continuity.
Publish events bound to candidate identity at the registry level.
Conversation history committed in a tamper-evident merkle structure — cannot be backdated or rewritten.
Candidate is asked to extend and defend their own past reasoning in real time. A curated or generated profile cannot survive it.
A note on scope. Provenant verifies job-relevant demonstrated capability against artifacts the candidate has chosen to disclose. It is not a personality, culture-fit, or psychometric assessment.
We launch where the artifacts are already API-accessible and structured at scale — engineering, in every discipline that produces public evidence. The same verification model extends to any profession whose work the world can independently check.
Software, hardware, security, infrastructure, data — disciplines whose work already lives on the verifiable web.
Regulated professions where filings, signed reports, and registries make evidence externally checkable.
Any profession that produces work the world can independently confirm.
The cryptographic verification primitives don’t change across professions — only the source registries do.
Software, hardware, security, infrastructure, data — wherever work produces externally verifiable artifacts.
We treat the candidate's device as untrusted, and never accept self-report as evidence. The verification layer is a cryptography problem — not a feature an HR-software incumbent can bolt on.
Conversation history is committed in a Merkle-style structure. Records cannot be backdated or rewritten without breaking the chain.
Repos, deployments, packages and filings are checked against third parties the candidate does not control — never via local software.
A curated or AI-generated profile collapses the moment the adaptive examiner asks the candidate to extend their own reasoning.
The full pipeline runs on the candidate's machine. The employer receives a privacy-preserving report, never the raw history.
An evidence-anchored screen, mapped to the specific role. Each claim links back to a verifiable source the candidate chose to disclose.
Strong technical depth in distributed cryptographic systems. Limited evidence of large-team coordination.
Defended 14 of 14 prior reasoning checkpoints under adaptive probing.
Provenant is currently rolling out to a small cohort of engineering teams and design partners. Request access to put verified longitudinal evidence in front of your next hire.
INVITES SENT WEEKLY · NO MARKETING EMAILS · UNSUBSCRIBE ANYTIME