The Problem

01 — the problem

The Portfolio Trust Crisis Is Real.

When anyone can prompt an LLM to clone an app in seconds, a perfect GitHub repository no longer guarantees an interview.

  • 01

    The LLM Clone Flood

    Recruiters are overwhelmed by thousands of identical, tutorial-copied repositories.

  • 02

    The Pedigree Trap

    Lacking validation metrics, companies default to filtering candidates solely by college tier.

  • 03

    The Silent Rejection

    Talented engineers with genuine foundational skills remain buried under unread applications.

recruiter@inbox ~ ./scan_pipeline.sh

$ ls ./applicants/ | grep clone

Candidate_482/netflix-clone⚑ skepticism flag
Candidate_483/netflix-clone⚑ skepticism flag
Candidate_484/spotify-clone⚑ skepticism flag
Candidate_485/netflix-clone⚑ skepticism flag
Candidate_486/instagram-clone⚑ skepticism flag
Candidate_487/netflix-clone⚑ skepticism flag

→ 2,847 repos · 96% structural duplicates · 0 verified depth

$

02 — the solution

Shifting the Metric from Code Ownership to Code Understanding.

ProofWork AI doesn't check if you own the code; it evaluates whether you have the engineering depth to maintain, scale, and debug it.

// stress.test()

Dynamic Valuation

We replace passive project viewing with active, contextual repository stress-testing.

// signal.verify()

Transparent Signal

Candidates gain an un-gamable, verified layer of technical credibility to attach to applications.

// bias = null

Fair Sourcing

Recruiters evaluate applicants purely based on execution evidence and engineering grit.

raw_repo.git
✦ gemini.prism()
verified_artifact.badge ✓

Frequently asked

Why are GitHub stars and commit graphs no longer reliable signals?
Stars can be bought, forks can be automated, and commit graphs can be back-filled with scripted commits. None of these prove the author actually understands the architectural decisions in the repo — they measure activity, not comprehension.
How does AI-generated code break traditional portfolios?
An engineer can scaffold a polished-looking repo in minutes with an LLM and never engage with the trade-offs in it. Reviewers can no longer tell, from the code alone, whether the author can defend a single design decision.
Why don't take-home tests solve this?
Take-homes are slow, easy to outsource, and bias against candidates with day jobs. They also test a fresh greenfield problem rather than verifying ownership of the candidate's existing public work.
What signal actually proves engineering ability today?
A verifiable reasoning trace: the candidate explaining, under adversarial questioning, why their code looks the way it does. ProofWork AI generates exactly that artifact from any public repo.