Benthic Whitepaper

1Introduction

Two curves are diverging. On one side, the demand for AI compute is compounding faster than any resource in modern computing history. On the other, an enormous installed base of GPUs and accelerators sits idle — in gaming rigs, in half-used data centers, in mining farms stranded by the end of proof-of-work. The market that should connect them barely exists, and the token economies that could coordinate it are busy paying for work that produces nothing.

Benthic exists to close that gap. Instead of rewarding miners for guessing hashes (proof of work) or for locking up coins (proof of stake), Benthic rewards them for performing useful computation the network actually needs — AI inference and training jobs — and proves that the work was done correctly before paying for it. We call this Proof of Useful Compute (PoUC).

1.1 Why now, and why a token first

A compute network is only as valuable as the demand flowing through it, and demand follows liquidity and attention. Benthic therefore launches in the opposite order from a typical infrastructure project: the $BENTHIC token launches first, as a fair, community-owned coin at ponsfamily.com, seeding an aligned community and a liquid unit of account. The mining network — the pool, the verifier mesh, the settlement contracts — is then built on top of that token, so that the people who believed early become the miners, verifiers and customers of the network they own.

1.2 Why Robinhood Chain

Rewards must settle somewhere fast, cheap and close to real financial rails. We target Robinhood Chain — an EVM-compatible Layer 2 built for tokenized, finance-native assets — as the settlement layer for mining epochs. Compute proofs are verified off-chain; only the compact, signed result of each epoch touches the chain, where rewards are minted and distributed to the pool. This keeps the network cheap to operate while inheriting the security and reach of an established L2.

1.3 Contributions

• A fully specified Proof-of-Useful-Compute mining mechanism that meters AI work in verified TFLOP-hours and prices it in $BENTHIC.
• A redundant verifier mesh with a challenge game that makes faking work unprofitable in expectation.
• An open mining pool (the Trench) with a transparent, share-based reward split and a halving emission schedule.
• A fair-launch token — no insider allocation, no pre-mine of the mining reward — that bootstraps the network it will later power.

2The Deep Compute Layer

In the ocean, the benthic zone is the floor — the layer where everything that drifts through the water column eventually settles and is turned into something useful by the organisms that live there. Benthic models compute the same way. Idle accelerators are drifting resources; AI jobs are the nutrients; the protocol is the floor where the two settle into value. The system decomposes into three layers.

• The current (demand). Streams of AI jobs — inference requests, fine-tuning and training batches — enter the network and are matched to available compute.
• The dive (work). A miner accepts a job, runs it on its hardware, and returns the output together with a proof. One unit of accepted, verified work is a dive.
• The settle (reward). Verified dives are aggregated per epoch and paid out in $BENTHIC according to each miner's share of useful compute.

2.1 Notation

Let the network contain N miners. In epoch t, miner i completes a set of jobs whose accepted, verified compute is w_i,t, measured in TFLOP-hours and scaled by a quality multiplier q_i,t ∈ (0, 1] reflecting latency, correctness and uptime. The epoch emission is E_t $BENTHIC, the pool fee is φ, and the headline outputs are each miner's reward ρ_i,t and the network's aggregate compute W_t = Σ_i q_i,t w_i,t.

3Proof of Useful Compute

The core problem is old and hard: how do you pay someone for a computation without re-running it yourself? Benthic does not claim a single silver bullet; it combines cheap statistical checks with an economic dispute game so that honest computation is the only strategy that pays. The pipeline from job to reward has four stages.

3.1 Metering work

Raw FLOPs are easy to claim and hard to trust, so Benthic meters accepted output, not self-reported cycles. Each job has a known compute profile (model, input size, precision); a correctly returned result is credited its profiled TFLOP-hours, scaled by the miner's quality multiplier. A miner's effective work in an epoch is

where c_j is the profiled compute of job j and q_i,t folds in correctness rate, response latency and uptime. Quality is not a bonus but a gate: a miner that returns wrong or late outputs sees q decay, and repeated failure removes it from dispatch.

3.2 The verification game

Verification uses redundant sampling: a fraction s of jobs are dispatched to two or more independent miners, and their outputs are compared (exactly for deterministic jobs, within tolerance for stochastic ones). Any accepted result may additionally be challenged inside a fixed window by a verifier who re-runs it and posts a bond. If the challenge succeeds, the cheating miner's stake is slashed and the challenger is rewarded; if it fails, the challenger loses its bond. With slashing set above the expected gain from cheating, honest computation is the unique profitable strategy.

The probability that a fabricated result escapes both redundant sampling and challenge is bounded by

where s is the redundancy rate and p_c the probability a result is independently challenged. Both are protocol parameters that rise automatically for high-value jobs, so the network spends verification effort where fraud would pay most.

input : job j with profiled compute c_j, value v_j
1  dispatch j to miner i  (and to a shadow miner w.p. s)
2  i runs the model, returns output o_i and commitment h_i
3  if shadowed and o_i ≠ o_shadow within tol:  reject, decay q_i
4  publish {j, h_i} ; open challenge window Δ
5  during Δ any verifier may re-run j, post bond, dispute
6      if dispute upheld:  slash stake(i), reward challenger
7  on window close (undisputed):  credit W_{i,t} += q_i · c_j
output: verified work credited to miner i

3.3 Reward function

At the close of each epoch the pool distributes that epoch's emission in proportion to verified work. Miner i's reward is

where φ is the pool fee (routed to the treasury and verifier rewards) and the ratio is the miner's share of total useful compute. The split is fully transparent and computed on-chain from the settled epoch report — there is no operator discretion over who gets paid what.

4The Trench · Mining Pool

A single miner's income from a proportional scheme is lumpy — you only get paid when your work lands in a settled epoch. The Trench is Benthic's open mining pool: miners contribute compute continuously and receive smoothed rewards proportional to their sustained contribution, minus the pool fee φ. Multiple independent pools may coexist; the protocol treats them as first-class and reports their share of network compute (see the live Network dashboard).

4.1 Share accounting

Within a pool, each miner accrues shares equal to its verified work. When the pool wins an epoch's emission, it is split by share balance and the balances reset. This is the familiar PPLNS idea, applied to verified TFLOP-hours instead of hashes, which naturally resists pool-hopping: a miner that leaves before settlement forfeits nothing it earned but gains no advantage from timing.

5Robinhood Chain Settlement

Compute happens off-chain; trust must reach the chain without re-execution. Each epoch, the pool operator aggregates verified work into a signed report — a Merkle root over (miner, verified-work, reward) leaves plus the epoch's parameters — and submits it to the settlement contract on Robinhood Chain.

Miners claim their share against the epoch root with a Merkle proof, so the on-chain cost is constant per epoch regardless of miner count. A challenge window on the report itself lets any party dispute a malformed settlement before rewards finalize. The contract is not yet deployed; parameters below describe the intended design.

6Tokenomics

$BENTHIC is the network's unit of account: customers pay for compute in it, miners earn it, and verifiers are bonded in it. Emission is disinflationary — a fixed schedule that halves at regular intervals so that early network growth is well rewarded while long-run supply is capped.

where E₀ is the initial per-epoch emission and H the halving interval in epochs. Total mined supply converges to a fixed cap independent of how fast the network grows.

6.1 Allocation

There is no team allocation carved out of the mining reward. The circulating float originates from the fair launch (Section 9); the mining schedule of Equation (4) emits new supply only to miners and verifiers who do work. A small, transparent pool fee funds the treasury and verifier incentives.

Figures are preview targets, subject to community calibration before mainnet.

7Architecture & Roadmap

The network is four cooperating roles: miners that run AI jobs, a verifier mesh that keeps them honest, pools that aggregate and smooth rewards, and the settlement contract on Robinhood Chain that mints and distributes $BENTHIC.

7.1 Roadmap

• Phase 0 — Fair launch. $BENTHIC launches as a community coin; liquidity and community form.
• Phase 1 — Testnet mining. PoUC client, verifier mesh and the Trench pool run on testnet with preview telemetry (live now on the dashboard).
• Phase 2 — Settlement. BenthicPool.sol deploys to Robinhood Chain; epochs settle and rewards mint on-chain.
• Phase 3 — Open market. External customers submit paid AI jobs; miner rewards shift from pure emission toward real compute demand.

8Security

The dominant threats to a compute-mining network are fake work and Sybil identities. Benthic addresses both economically rather than by trust.

• Fake work is caught by redundant sampling and the challenge game of Section 3.2; slashing exceeds the expected gain from cheating, so E[cheat] < 0.
• Sybil attacks gain nothing because reward tracks verified compute, not identity count — splitting one machine into ten identities produces the same work and the same pay, minus overhead.
• Verifier collusion is bounded by requiring independent re-execution and by making challenge profitable, so a lazy or colluding verifier is undercut by an honest one.
• Settlement safety follows from the report challenge window on Robinhood Chain: a malformed epoch can be disputed before rewards finalize.

9$BENTHIC & Fair Launch

$BENTHIC is live first as an honest, community-owned coin at ponsfamily.com. There is no insider allocation and no promise that today's coin already does everything in this paper — the mining network described here is being built after and on top of the launch. What the launch buys is the one thing infrastructure cannot bootstrap on its own: an aligned community that becomes the first miners, verifiers and customers.

10Disclaimers

This document is a technical and conceptual overview provided for information only. It is not financial, investment or legal advice, and it is not an offer or solicitation to buy any asset. $BENTHIC is a community token; the mining network, settlement contracts, parameters and emission figures described here are design targets that are not yet deployed and may change. Forward-looking statements carry risk and uncertainty. Digital assets are volatile and you may lose everything you put in. Do your own research and never risk more than you can afford to lose.