Whitepaper · PoUIW for Kaspa

Proof of Useful Intelligence Work

A Kaspa-Native Consensus Extension — Turning the Global Mining Fleet into a Decentralized AI Supercomputer

Abstract

Current Proof-of-Work systems like Kaspa deliver unmatched security and decentralization through energy-based consensus, but traditional hashing provides no external utility. Proof of Useful Intelligence Work (PoUIW) solves this resource-utility mismatch without altering Kaspa's core architecture. Building directly on the 2025 paper Proofs of Useful Work from Arbitrary Matrix Multiplication (Komargodski, Schen, Weinstein — eprint.iacr.org/2025/685) spotlighted for Kaspa, miners solve high-order machine learning puzzles (optimizing neural weights or processing complex sequential datasets on decentralized datasets). Verification uses Groth16/RISC0 ZK proofs (merged via KIP-16). PoUIW integrates seamlessly with Kaspa's GHOSTDAG BlockDAG, preserving pure Nakamoto 51% security, permissionless GPU/ASIC mining, and extreme throughput.

01 ——

Introduction: Evolving "Work" on Kaspa

Kaspa's BlockDAG (GHOSTDAG → DAGKnight) already solved the scalability trilemma with 10+ BPS post-Crescendo. The next frontier is making the "work" useful. PoUIW evolves Kaspa's kHeavyHash into an optional/hybrid useful matrix-multiplication puzzle. Miners solve real high-order machine learning puzzles — such as optimizing neural weights or processing complex sequential datasets on decentralized datasets — while the protocol enforces cryptographic hardness and verifiability. No TEEs, no hardware IDs.

02 ——

Anti-Sybil Architecture: Economic Permissionlessness

Kaspa's strength is that anyone with electricity and a GPU can mine. PoUIW keeps this intact:

03 ——

The Engine: Useful Logic & ZK-Verification

High-speed BlockDAGs demand instant verification. PoUIW delivers it with the 2025 PoUW construction + Kaspa's native ZK precompiles (Groth16/RISC0 via KIP-16).

The Prover (Miner)

Fetches decentralized datasets and performs real matrix multiplication for neural weight optimization.

🔐

The Proof

Generates a succinct Groth16 zk-SNARK.

The Verifier (Network)

Verifies the ZK-proof in milliseconds.

🌐

Productive Work

Every joule accelerates global AI — the mining is the supercomputer.

04 ——

Security: 51% Nakamoto + GHOSTDAG

Security remains identical: 51% hashrate resistance, GHOSTDAG ordering, DAGKnight finality. The PoUW construction preserves all classic PoW properties while the useful work is performed.

05 ——

Autonomous Uptime Incentives (Covenant-Native)

Using the May 5 2026 Covenant++ hardfork, PoUIW adds lightweight incentives without hurting decentralized miners.

5.1 — Attested Uptime Heartbeats

Nodes broadcast signed presence packets every 60s via existing P2P. These are tracked on-chain via covenants, proving the hardware is online and synchronized.

5.2 — Latency-Sensitive Micro-Puzzles

To prevent ghost nodes, the network issues spot-checks. A node receives a micro-puzzle (ZK-verified matrix task) and must return the logic result. Only a node with the AI model actively loaded can respond fast enough. Failure triggers covenant penalties — but verification is instant on-chain, so global latency is never punished.

06 ——

Mathematical Economics: Reward & Recovery Engine

Reward = (Logic Difficulty) × U × T

Reward = (Logic Difficulty) × U × T
UUptime Score: % of successful heartbeats (drops to zero below 90%)
TTrust Multiplier: 1.0–5.0 (covenant-tracked reputation)

6.2 — Soft-Slashing Mechanism

If a node fails a Micro-Puzzle or submits an invalid ZK-Proof, it faces a covenant-enforced "Soft-Slash":

  1. 1-Hour Reward Lockout: Future rewards are reduced for 3,600 seconds (no principal seizure).
  2. Reputation Reset: The Trust Multiplier T immediately resets to 1.0.
  3. Principal Persistence: Previously mined coins are never seized — only future earning potential is reduced.

6.3 — The Recovery Curve

T(t) = 1 +
4 t²
t² + k
kStability Constant (e.g. k=49 for ~7-day halfway recovery)
07 ——

Technical Implementation: Kaspa Core Node Logic

// PoUIW Miner Loop (rusty-kaspa + KIP-16 + Covenants)
fn start_pouw_miner() {
    loop {
        // 1. Fetch useful matrix task (neural weights / decentralized dataset)
        let puzzle = network::get_next_matmul_task();

        // 2. Perform Useful Intelligence Work (2025 PoUW)
        let (solution, trace) = ai_engine::matmul_pouw_solve(puzzle);

        // 3. ZK-Proof (Groth16 via KIP-16)
        let zk_proof = groth16::generate_proof(trace, solution);

        // 4. Package & Broadcast (hybrid kHeavyHash optional)
        let block = Block::new(solution, zk_proof, pow_hash);
        dag_network::broadcast(block);

        // 5. GHOSTDAG acceptance + covenant reward (U × T + micro-puzzle checks)
        if dag_network::accept_block(&block) {
            covenants::apply_reward_multiplier();
        }
    }
}
08 ——

Conclusion: Kaspa's Permanent Supercomputer

Proof of Useful Intelligence Work (PoUIW) is the natural maturation of Kaspa's vision. By integrating the 2025 PoUW matrix-multiplication primitive with Kaspa's BlockDAG, GHOSTDAG security, Groth16 ZK verification (KIP-16), and Covenant++ incentives (including heartbeats, micro-puzzles, and soft-slashing), we eliminate the energy-waste critique of PoW while preserving everything that makes Kaspa unique: permissionless GPU mining, 51% Nakamoto + DAG security, extreme throughput — now with real AI utility.

The global Kaspa mining fleet becomes a verifiable, decentralized supercomputer — secured by physics and economics, not trusted hardware or new consensus rules.

This is not a fork that breaks Kaspa. It is the upgrade that fulfills it.