Technologies
Proof of Useful Work. A decentralized compute market.
Proof of Useful Work on Tenrai is planned as a decentralized market for computing power, not as a new mining puzzle. Customers pay for real compute jobs, from AI to zero-knowledge proofs; miners with GPU power to spare run them; independent miners check the results; and the chain settles the payment. TenQ keeps producing the blocks, and the validators keep making them final.
Research only: nothing here can be used yet
Proof of Useful Work is research. Its development has not started, it is not part of mainnet, the testnet or any release, and no one can use it today. A few reserved building blocks in the code prepare the way; everything else on this page is a goal or an open question, and any detail can still change.
A market, not a puzzle
Today every TenQ attempt has one purpose: finding the next block. Proof of Useful Work asks whether the same graphics cards could also do work that people actually need. Tenrai's answer is an open market: real jobs, posted and paid for by real customers, from AI providers and apps to companies, with Tenrai's own Trai AI as one customer among many. Every job is useful for a simple reason: someone wanted it done and paid for it.
How it works: a customer posts a job and pays into escrow, a miner computes it, independent miners check the result, and only a successful check releases the payment. Jobs run beside block production, never inside it.
Fast blocks, small chunks
Tenrai's speed is a real advantage for a compute market. With 50 to 100 blocks per second and finality within seconds, a large job does not have to be one large deal: it can be offered in small chunks, and each chunk is checked and paid on its own as soon as it is done. Miners are paid as they go instead of waiting for the whole job, customers see results arrive step by step, and a faulty result costs only one small chunk.
Why not matrix multiplication?
Some projects build Proof of Useful Work around matrix multiplication, because AI runs on it. We decided against it. Math that could be useful in theory does not make mining useful in practice, and the incentives it creates work against the very GPUs it claims to serve.
Made for special chips
Matrix multiplication is the most hardware-optimized operation there is. An FPGA or ASIC can compute exactly the steps the protocol checks without ever running a real AI task, and FPGAs can be bought today.
Math is not demand
A valid matrix proof shows that the math was done, not that anyone needed it. Rewards flow even when no one asked for a result.
Cheapest wins, not most useful
Rewards go to the lowest cost per accepted proof. Hardware built only for the puzzle can beat GPUs doing real AI, and more mining power does not bring more useful work.
Hardware stays hidden
A proof does not show whether a GPU, an FPGA or an ASIC made it, so no one can check who really secures such a network.
Our answer: keep the two jobs apart
TenQ produces the blocks with work built to be hard for special chips, and the validators make them final. Useful work is real work for real customers, paid only after independent checks. Its usefulness is proven by demand, not by the shape of a puzzle.
Room for every machine
TenQ, the proof of work that produces Tenrai's blocks, is built to take away the edge of special hardware, not to lock it out. Proof of Useful Work completes the picture: graphics cards, CPUs, ASICs, FPGAs and, in theory, even quantum computers could all earn with the jobs they do best, checked like any other. Special hardware is welcome when it does real work that someone pays for; what Tenrai avoids is a shortcut to block rewards.
Because block rewards shrink by 0.3 % every day, verified work is meant to become the main way miners earn over time.
One machine, many kinds of work
The goal is not a single AI service but a virtual machine with versioned runtimes, one for each kind of job: the Tenrai Virtual Compute Machine. Its groundwork is already in the code: a fixed registry of runtimes that every node shares and no node can extend on its own, and an early deterministic runtime for AI inference. The other runtimes are reserved.
AI inference
Running trained models with exact integer math, so every honest machine reaches the same answer and every step leaves a trace.
AI training
Training and fine-tuning models, the most demanding kind of job.
Programs in a sandbox
General programs in a WebAssembly sandbox, sealed off from the machine they run on.
Parallel number crunching
Large integer workloads that spread across thousands of GPU cores.
Zero-knowledge proofs
Proof work, the heavy lifting behind private and verifiable computing.
Work for other networks
Mining and other work for other blockchains, booked like any other job.
Checked by independent miners
The check: one miner computes a job, and a second miner, chosen independently, computes it again. If both valid results match, the job is confirmed. If they differ, a third independent checker steps in to find the faulty or forged result.
Selection no one can steer
Checkers must be chosen so that no one can influence the pick, and no miner can quietly hold several checking roles.
Matching is not proof
Two equal answers do not prove a correct one. The rules must define what counts as correct, also for jobs with randomness.
No check, no payment
An unclear or failed check never releases a payment.
Rules, not promises
Every decisive check is enforced by the protocol or by independent miners, never by a miner's own software alone.
TenQ and Traidag stay in charge
Proof of Useful Work would add to the network, not rebuild it. Finality comes from the validators and their stake, not from hashrate, so miners can give part of their power to jobs without weakening the chain: the difficulty simply adjusts. TenQ mining stays, the validators keep the final word, and the chain never depends on the job market.
Mining never stops
No jobs, or the job market down: miners simply keep mining TenQ.
The chain never waits
Long jobs never hold up a cut, and the block rate does not depend on how long a job runs.
Settlement without surprises
Job payments run through escrow and settle by fixed rules: no double payouts, no replayed results, no payment for unfinished work, and an invalid result can never make the ledger invalid.
Block rewards unchanged
Block rewards stay split half to miners and half to stakers. The ledger already reserves a compute area for jobs.
Open questions
The decisions still to make:
- Which kinds of jobs come first, and their exact specification.
- How results of jobs with built-in randomness are checked.
- How checkers are chosen, how many must agree, how disputes end and what cheating costs.
- How a job's payment is split between the miner who computes and the miners who check.
- Whether useful work could one day replace part of TenQ or stays an extra income for miners.
- How the network is protected against fake miners, replayed results and jobs built to overload.
Built on TenQ and Traidag
Miners earn with TenQ, and every payment is final with Traidag.