.

Intelligence evolves.
Trust must endure.

Two AI research roles explore migration failures, propose repairs, and build a memory of what survives verification.

Two AI roles. Fixed verification. Lasting evidence.
MODULE-LATTICE (NIST FIPS 204) DIMENSION: R_q^(4×4)
ORACLE ENGINE ACTIVE NIST FIPS 204 · ML-DSA-44
ALGORITHM TRANSITION ECDSA-P256 ➔ ML-DSA-44
TARGET PRIMITIVE MODULE-LATTICE (CRYSTALS)
VERIFIER ORACLE OPENSSL 3.5.7 LOCKSTEP
RETIREMENT POLICY ZERO FALLBACK ENFORCED
SHA256: 7d8f...391a · DETERMINISTIC PASS ENTER LAB
AI-assisted post-quantum transition researchExplore the mission

When the keys change,
who keeps control?

The world will change how it proves identity.
Trust needs a way through.

A signature can still be genuine after its key loses power. A failed upgrade can reopen an old door. QAVEN studies these moments of transition.

Our mission: help digital identity, authority, and ownership survive the shift to post-quantum cryptography.

Why this matters now
LEGACY KEYProof of the past
POST-QUANTUM KEYAuthority in the present

Two minds.
One evolving defense.

One searches for the failure.
One works on the repair.
Every advance must earn its place.

Allowedeventssign → hold → migrate → delivermigrate → bad proof → fallbackdeliver → commit → replaySCHEMATIC / BOUNDED TEST GENERATION

SEEKER

Find the blind spot.

Reorders events, probes forbidden fallback, and turns a complex failure into a case we can reproduce.

candidate_handler(request)require valid_signaturerequire active_keyrequire current_epochrequire unused_nonceFixed checksSCHEMATIC / CANDIDATE, THEN ACCEPTANCE

KEEPER

Make the lesson last.

Proposes a bounded repair, keeps legitimate access working, and carries accepted methods into the next round.

A fixed verification kernel decides. Algorithms, permission rules, and acceptance criteria stay outside the agents’ control.

How we build it

The Observation Room.
An auditable window into autonomous research.

Follow two model roles from a public challenge to the cryptographic checks, state transitions, and version decisions they produce.

DUAL-AGENT EXPLORATION

SEEKER & KEEPER

SEEKER systematically searches for subtle post-quantum boundary failures; KEEPER synthesizes bounded invariant repairs to prevent unauthorized legacy fallback.

Active Chapter: The Retired Key (FIPS 204)
DETERMINISTIC REPLAY

Deterministic State Verifier

All transitions execute within an immutable state machine backed by OpenSSL subprocess cryptography. Every trace, failure proof, and candidate change is reproducible offline.

Kernel Engine: Sub-millisecond Execution
INTERACTIVE GENEALOGY

Policy Lineage & Evolution DAG

Explore the full genetic tree of defense policies. Inspect how vulnerability counterexamples drive verified invariant patches across version generations.

78 Suites · Zero Regression Guaranteed
Interactive Research Laboratory

Chronological event playback, failure analysis, and verifiable cryptographic evidence packs.

It doesn’t just run.
It builds a memory.

Evolution means inherited methods
with inspectable evidence.
Progress is never assumed.

Every verified failure
changes what comes next.

What actually evolves Illustrative method lineage
THE STARTING POINT

A method with a blind spot.

The current handler trusts the signature. It misses the fact that the signer’s authority has ended.

THE CHALLENGE

A failure we can replay.

A delayed request exposes the gap. Its sequence and origin become part of a traceable research record.

THE INHERITANCE

A lesson future versions must face.

An accepted repair gains a parent, a test, and a history. The challenge becomes a regression the next applicable version must still pass.

Actual recorded experiment / PROBE–03
Checking record

Fixed local cryptographic checks.
No AI-generated repairs in this record.

Open evidence

The idea is bold.
The method is open.

What we are. Why the transition matters. How two AI roles turn a challenge into a testable result.

Read the QAVEN docs

Give evolution
a reason to happen.

A contribution is a challenge that changes the research: a new failure, a sharper reproduction, or an independently checked result.

How contributions are recognized

Start with a challenge.

NOT CONNECTED

Select a published experiment above to submit its boundary sequence.