Decentralized AI · verified · settled on Metal

Verifiable intelligence,
settled on Metal.

The model: independent workers serve AI, evaluators verify quality, and customers pay in stablecoins — so reward tracks real demand, not emissions.

● Settlement rails live on Metal testnetThe settlement contracts are deployed and live on Metal Tahoe, and we run the full escrow→commit→score→settle lifecycle on them — so each settlement links to a real, verifiable transaction on Metalscan. The marketplace below is a simulation of how the network runs, settled with test accounts; independent workers and real demand are in progress — see the roadmap. Read raw contract activity on the Activity page or Metalscan ↗.

How a request becomes a payout

01

A request is posted

A customer posts a paid task and escrows the fee in stablecoins — no volatile token to hold.

02

Workers serve it

Independent, bonded workers compete to return the best answer. METAL bonds are their skin in the game.

03

Evaluators verify

Stake-weighted evaluators score quality with a trimmed, plutocracy-capped median. Gaming it means gaming everyone.

Metal
04

Settled on Metal

Escrow releases on-chain — conservation-checked so payouts sum exactly. Reward is capped by real demand.

Markets
5
in this preview simulation
Operators
30
15 reference workers · 15 evaluators
On-chain settlements
18
real testnet txs · test accounts
Revenue
0.63
XMD · simulation

Verified on-chain settlements

Metal Tahoe · test accounts

1 real end-to-end settlement lifecycles executed on the live testnet contracts and verified on-chain — each worker received exactly its payout. This is proof the settlement rails work, with test accounts; it is not a claim of independent workers or real demand.

RequestEscrow · mXMDWorker paid · mXMDBlockEscrow txSettle tx
0x3d20a1f31.000.70 #19010x859937c1… ↗0x87f10aa3… ↗

Each worker payout was read back from the chain after settlement — an exact, conservation-checked split. Verify any of them on Metalscan.

Demand-coupled rewards
3.00 : 1emission : revenue

A market's emission is min(base, 3× fee revenue). With no paying customers it earns ~nothing. Bittensor runs near 100–300 : 1; the target is below 1 : 1.

Reward formula
score = .45·quality + .30·√(revenue) ← anti-wash
    + .13·retention + .07·efficiency + .05·reliability

Quality and demand are measured independently, so an attacker must game both. The √ on revenue makes wash-trading unprofitable once burned fees exceed the marginal reward.

Rewarded for demand, not for existing

Emission-first networks

  • Emission set by token price or validator vote
  • Reward tracks stake (0.80–0.95), not performance (~0.50)
  • A market with zero customers still earns
  • “Verifiable” is a slogan, not a mechanism

Metal Intelligence Network

  • Emission capped at a multiple of real settled fee revenue
  • Reward dominated by verified paid usage + measured quality
  • Zero demand → zero emission after bootstrap
  • Correctness = independent evaluation + staking + challenge

Top workers

Open market →
WorkerQualityRevenue · XMDCustomersRequests
min17c8f35fe422b…6a4f
93%93
0.0646
24
min1b36486c5f1c6…70a2
78%78
0.0583
24
min176b3c53b94eb…a26c
40%40
0.0171
22

Recent requests

All requests →
RequestMarketKindStateMax · XMDSettlement
stablecoin-risk-monitoring-req-008stablecoin-risk-monitoringquality probesettled0.05probe
stablecoin-risk-monitoring-req-007stablecoin-risk-monitoringquality probesettled0.05probe
stablecoin-risk-monitoring-req-006stablecoin-risk-monitoringquality probesettled0.05probe
stablecoin-risk-monitoring-req-005stablecoin-risk-monitoringpaidsettled0.050x61defb13… ↗
stablecoin-risk-monitoring-req-004stablecoin-risk-monitoringpaidsettled0.050x4e3c891a… ↗
stablecoin-risk-monitoring-req-003stablecoin-risk-monitoringquality probesettled0.05probe
stablecoin-risk-monitoring-req-002stablecoin-risk-monitoringquality probesettled0.05probe
stablecoin-risk-monitoring-req-001stablecoin-risk-monitoringquality probesettled0.05probe
Overview