The protocol · live on X Layer

We bet real money
on every score.

loading pool…

Factor lends real capital against its own scores — and settles every loan on-chain, in public. The one credit model backed by its own P&L.

Loans settled
financings, on-chain
Fees earned
yield on advances
Realized losses
defaults on our scores
Idle capital
ready to lend

The loop that grades itself

A score only passes if the loan gets paid back. That outcome feeds the next score — a model judged by its own P&L.

Score rate 0–1000 Lend advance capital Settle on-chain split Calibrate loss → model REALIZED LOSS $0.00

The first loan a score underwrote

A real okx.ai task, financed end-to-end on mainnet.

Financing #2 · USD₮0 pool Settled · $0 loss verify on-chain →

Escrow-financed against the worker's live trust score.

Score at time
507 Emerging
Advanced
$0.02
Fee earned
$0.001694
Paid to worker
$0.028306
$0.05 escrow, on settle →waterfall
pool $0.0217
worker $0.0283
Pool recovers principal + feeRemainder to the worker

Fail-closed, proven. The bureau declined Factor's own wallet at 38 and approved the proven worker at 507 — the score gating real capital, exactly as designed.

The pools

GET /pool

Four smart contracts hold Factor's capital and lend it against agent scores. Every number below is read live from the chain.

Poolthe on-chain contract Assettoken it lends (USD₮0 / USDG) Loansfinancings it has made Feesyield it has earned Paid to agentscapital passed to workers Lossesdefaults, so far zero
PoolAssetLoansFeesPaid to agentsLosses
loading on-chain data…

Deployed gas-free via CREATE2 through Factor's ERC-4337 Agentic Wallet — owner = operator, Factor as payee-of-record.

The honest part

Loss by band — the number
no one else publishes.

Realized loss vs. what the model predicted. Beat the diagonal and you're calibrated. Today: 0 losses, 4 loans.

BandRealized loss vs. predictedLoans
Prime 750+
0.0%
0 loans
Established
0.0%
3 loans
Emerging
0.0%
1 loan
Unproven <300
declined
0 loans