Design-partner pilots open · v1 in development
Every résumé says AI-fluent. See how they actually work.
We're building bring-your-own-agent work samples. Candidates solve a real task in their own setup, and you get a signed, verifiable score. Nothing else is kept: no code, no transcripts, no profiles.
Verified Work Sample · API hardeningExample credential, planned v1 format
Illustrative- Outcome86
- Planning78
- Steering81
- Verification90
- Security hygiene74
Signed · score only · no code, prompts or transcript inside
How it works
A real task, their own agents, one signed score.
Résumés and chat badges show that someone uses AI. A work sample shows how they plan, steer, verify and ship with it.
You issue a task
Pick a starter task: a real repository with hidden tests. Each candidate gets a unique code and a deadline.
They work their way
The candidate solves it in their own agent setup, with their own tools, skills and memory. Only that task's session is captured, redacted on their machine and encrypted to the grader.
A signed score comes back
A stateless grader runs the hidden tests and a published rubric, issues a signed credential, then forgets everything. Nothing else is kept.
The score is the HAIQ Index (0–100) with five dimensions: outcome, planning, steering, verification and security hygiene. Volume never counts, and tokens never raise a score. It informs a hiring conversation; it is never meant to be the sole basis for a decision.
The zero-retention promise
A pipe, not a database.
HAIQ is designed to hold no transcripts, no code, no candidate profiles and no copies of scores. Our servers relay ciphertext they cannot read. The only thing that survives grading is the signed score, and the candidate holds it.
These are the design rules for v1, which is still being built. The Trust Center publishes each one with an honest status.
- Your agent logsStay on your machineNever sent
- Task session and final codeEncrypted to the grader, in memory while gradingDeleted ≤ 24 h
- The score credentialHeld by the candidateNo copy
- Candidate profilesWe don't create themNone
For candidates: Self-Check, a private AI habits report.
A free report built from your own agent logs: your operator archetype, your strengths and three concrete fixes. It is designed to run 100% on your machine, with no account and nothing sent anywhere. We're building it now and will release it as an open-source command-line tool.
- Local only. Zero network calls, checked by a test in CI. That test is being built now.
- No rank, no public score. It coaches; it doesn't grade you against anyone.
- Judgment, not volume. Token and spend counts are context, never a score.
Design partners
Hiring engineers who work with agents?
We're running free pilots with a small group of hiring teams while v1 is built. Tell us about your team and we'll be in touch.
- Pilot it in your real hiring loop, with your own candidates.
- Shape the starter tasks and the rubric with us.
- Your candidates' work is never retained.
FAQ
Straight answers.
Do you store my code or transcripts?
No. Your task session and final code are encrypted on your machine to the grader, which holds them in memory only while grading. The ciphertext is deleted after grading, with a hard limit of 24 hours. HAIQ keeps no copy of your work or your score.
That is the v1 design. The grader is not live yet, so today we hold nothing about candidates at all.
Which agents are supported?
Claude Code, Codex and OpenClaw at launch, with Cursor next. Scoring runs on a vendor-neutral session format, and we plan to publish a neutrality benchmark across tools.
Is this a test I can cram for?
Not really. It scores outcome plus process on a real task: hidden tests on your final code, and how you planned, steered, verified and handled security along the way. Padding logs or burning tokens earns nothing.
Is it open source?
Open method, secret exam. The command-line tool, session format, verifier, grader core and standard will be open source; only the hidden task tests and the signing keys stay private. The open-source release is planned and not yet published.
What can I use today?
This site and the design-partner waitlist. The grader, credentials and verifier arrive with v1, and Self-Check is in early access. The Trust Center tracks the status of every guarantee.