Roadmap & Vision

What we built, why it compounds, and where it goes from here.

The Premise

DeFi has not solved credit. It has avoided it. Every lending protocol built in the last five years operates on the same unstated assumption: that all borrowers are identical, trust no one, and overcollateralise the system into functional solvency. This works for processing volume. It works poorly for the actual use case of credit - rewarding participants who have demonstrated verifiable financial behaviour over time.

On-chain data has always contained enough signal to build a genuine credit layer. Repayment records, deposit history, liquidation events, governance participation - all timestamped, all immutable. The obstacle is constructing a scoring engine that can consume that data inside a smart contract, return a defensible result, and do it without introducing an off-chain trust dependency at the critical step.

Polkadot Asset Hub, with its hybrid EVM-plus-Wasm runtime and native XCM, makes this solvable in a way it isn't on any other execution environment. Kredio is the protocol that solves it.

What We Built: The Intelligence Layer

Dual-Mode Credit Scoring

Credit scoring at Kredio is intentionally two-part. KreditAgent is an ink! Wasm contract that computes a deterministic score from six on-chain inputs - repayment history, deposit volume, borrow frequency, liquidation record, account age, and governance participation - and returns a collateral ratio and interest rate via a SCALE-encoded cross-VM staticcall from inside the Solidity market contract. The score, ratio, and rate are all computed and locked into the position atomically, in the same transaction, with no oracle delay and no possibility of front-running.

NeuralScorer is the second layer - a PVM contract that runs independently after every borrow event and emits a confidence delta on-chain. It catches the gap between the deterministic score and the statistical behaviour of the broader borrower pool. A score of 72 may be mechanically correct but statistically unusual for an address of that age; the NeuralScorer surfaces that discrepancy as a permanent event record. Together the two layers produce a score that is simultaneously provably derived and continuously audited.

Forward-Looking Risk Assessment

Most DeFi protocols check position health at liquidation time. RiskAssessor checks it continuously. It estimates the number of blocks before a position would cross the liquidation threshold, given current collateral values, the outstanding debt, market volatility, and the position's tier-specific collateral ratio. The result - estimated_blocks_to_liq - is emitted on-chain after every scan, forming a per-borrower risk timeline that accumulates across the full life of the protocol.

This matters for protocol health. A system that only detects liquidation events learns about risk after the fact. A system that sees estimated blocks-to-liquidation across all open positions has predictive capital coverage information - and that data becomes the training signal for Phase 4.

Autonomous Yield Intelligence

YieldMind determines where idle capital should sit at any given moment. It takes the current pool utilisation rate, the yield differential between custody and deployment, and the minimum liquidity buffer as inputs, and emits a recommended allocation split across internal custody and external yield sources. Each decision comes with a documented reasoning_code- an integer that maps to the specific market condition that drove the output - making every capital allocation decision auditable in retrospect.

The Score in Six Tiers

The credit score maps to six discrete borrowing tiers. Higher scores unlock better loan-to-value ratios and lower interest rates. On mainnet, the tier structure is designed as follows:

Movement between tiers is automatic and continuous. Every completed repayment increments the on-chain counter. Every deposit contributes to the deposit volume signal. Account age accrues passively. There is no application, no approval gate, and no human decision - the score is a function of behaviour, re-computed on every borrow.

The XCM Settlement Engine

KredioXCMSettler lets any Polkadot parachain with XCM Transact capability execute Kredio protocol actions without the user leaving their home chain. A borrower on Hydration or Bifrost sends a single XCM extrinsic with a compact-encoded intent payload. On Asset Hub, the settler decodes the intent and executes the corresponding protocol action in the same block - atomically, with full settlement guarantees.

Every intent produces three on-chain events: received, dispatched, and acknowledged. This three-event lifecycle provides a complete, auditable record of every cross-chain protocol interaction. The intent separation matters - a SWAP_AND_LEND intent and a plainDEPOSIT_LEND intent follow the same settlement path but carry different protocol implications; they are distinct by design rather than handled by a single catch-all route.

FULL_EXIT is the most powerful intent: repay outstanding debt and withdraw collateral in a single XCM extrinsic from a remote chain. A cross-chain borrower who opened a position via XCM can close it entirely without switching chains or interacting with Asset Hub directly.

Mainnet: What Plugs In

Productive Collateral

On testnet, collateral is held in custody while a position is open. On mainnet, collateral will be deployed to yield while locked. The primary integration is DOT staked via Bifrost's liquid staking pool - a depositor posts DOT collateral, which is converted to vDOT earning approximately 15% APR, and the collateral balance is denominated in the live vDOT/USD oracle price. Collateral earns its keep while it secures the loan.

This changes the protocol's capital efficiency fundamentally. A 200% collateralisation requirement against earning collateral is economically different from 200% against idle collateral - the yield offsets a meaningful fraction of the borrowing cost and reduces the net cost of borrowing for lower-tier users.

Real Yield for Depositors

YieldMind will route idle lending pool capital to external yield sources. The two primary targets at mainnet are Bifrost (conservative yield, liquid staking returns) and Hydration Omnipool (higher yield, liquidity provision exposure). YieldMind selects between them based on current protocol utilisation, pool depth, and volatility signals - not a hardcoded split. Depositors earn a blended yield from borrower interest plus the external routing decision, with the allocation logic fully auditable on-chain via the emitted AllocationComputed events.

ETH Liquidity Without Intermediaries

The ETH bridge at mainnet will route through Snowbridge - the trustless, governance-secured bridge between Ethereum and Polkadot Asset Hub. Snowbridge eliminates the relayer trust assumption present in the testnet implementation; deposits are verified by Ethereum consensus light client proofs running inside Polkadot, with no custodian and no multisig. ETH holders gain access to the Kredio credit system without bridging through a centralised intermediary.

Multi-Asset Collateral at Mainnet Depth

The Connected Protocol

Kredio is not a standalone lending pool. It is a credit layer that sits on top of the Polkadot ecosystem's native yield infrastructure. The XCM Settler accepts intent from parachains that have no lending market of their own. YieldMind routes capital to Hydration and Bifrost. The ETH bridge brings Ethereum capital into the credit system. The identity registry allows KILT credentials to deepen the score for users who choose to link them.

Each integration compresses the effective cost of Kredio's value proposition. Borrowers benefit from productive collateral; lenders benefit from external yield routing; parachains benefit from cross-chain credit access. The protocol is designed so that adding one integration makes every other integration more valuable.

The YieldMind allocation engine was built to make decisions, not to hold a number. On testnet the decisions are real - the logic, the reasoning codes, the conservative/balanced/aggressive split - but the destinations are mocked. On mainnet, every bucket has a named address and a real protocol behind it.

Conservative capital flows to Bifrost. Deposited DOT that is not actively serving collateral gets routed via the XCM settler to Bifrost's liquid staking layer, where it earns Polkadot native staking yield as vDOT. The capital remains recoverable via XCM at any time. There is no lock-up - the settler can recall it the moment utilization demands it.

Aggressive capital flows to Hydration's Omnipool. Surplus USDC that exceeds the protocol's minimum liquidity buffer enters Hydration's single-sided LP position, earning swap fees on every trade that passes through the pool. Hydration is the deepest native liquidity venue on Polkadot - its fee revenue reflects real market activity, not an assigned APY.

Snowbridge makes the third dimension possible. Capital does not have to stay within the Polkadot ecosystem to work for Kredio's depositors. An Ethereum-side yield strategy - USDC deployed into Aave on Ethereum, ETH collateral earning staking rewards via Lido - becomes reachable through the same bridge infrastructure that brings ETH collateral onto Asset Hub. The XCM settler and the Snowbridge inbox already have a defined relationship in the codebase. Extending that relationship to move yield-seeking capital outbound is the same pipe, running in the other direction.

The result is a lending protocol whose idle capital is never truly idle. Depositors earn from borrower interest, from staking yield on DOT collateral, from LP fees on surplus USDC, and in later phases from cross-chain yield positions opened via Snowbridge. Kredio does not offer a fixed yield number - it builds infrastructure that routes capital to wherever yield is currently best, governed by on-chain intelligence that adjusts with every block.

Development Phases

Why This Compounds

Standard DeFi yield compounds trivially - principal grows, interest compounds on a larger base. The compounding in Kredio is structural. Credit reputation accumulates and cannot be instantaneously replicated. A borrower who reaches Elite tier over eighteen months of on-chain behaviour has built something that a new wallet cannot reproduce overnight. That asymmetry is non-trivial: it creates a meaningful incentive to protect reputation rather than exploit the protocol and restart.

The AI layer compounds separately. NeuralScorer confidence deltas accumulate across thousands of borrow events and become the training data for Phase 4 model updates. RiskAssessor's historical estimates versus actual liquidation outcomes become a ground truth dataset that makes future estimates more accurate. YieldMind allocation history becomes a backtestable track record that informs future strategy selection. The evidence base grows with every transaction.

The ecosystem integration compounds further still. When computeScore becomes an XCM endpoint in Phase 2, every parachain that queries it adds a new dimension to the network's understanding of that borrower - and every additional data point deepens the NeuralScorer's cross-validation capability.

On the Build Choices

There is a real engineering question behind every testnet demo: what is the minimum viable version of this feature that makes the underlying architecture visible and correct?

For some features, a mock is the right answer. The mock yield pool inKredioLending serves a clear purpose - it lets the YieldMind allocation logic run, produce events, and demonstrate its decision-making without requiring a live Bifrost endpoint that does not exist on Paseo testnet. The mock is a stand-in for an interface, not a substitute for the design.

For other features, building an elaborate mock would have consumed the hackathon while producing something less useful. Connecting to a fake vDOT contract that accepts pretend DOT and returns simulated yield would have demonstrated the concept at the cost of the real work - which is the credit intelligence, the dual-mode scoring architecture, the XCM acknowledgment engine, and the event schema designed for long-term data accumulation. That work does not compress. It either exists or it does not.

The choice to build the XCM settler as a complete intent-acknowledgment engine - rather than a simple send-and-forget call - was made precisely because the failure modes of real cross-parachain operations are where integration work actually lives. A settler that dispatches and never follows up is not a cross-chain protocol. It is a one-way message system. The three-event lifecycle on every intent is not demonstration overhead. It is the operational standard that real cross-chain capital flows require.

When Bifrost is available as a destination, the settler receives a new address and a new call payload. The acknowledgment loop does not change. The event schema does not change. The YieldMind logic that decides when to deploy capital does not change. The integration work required to connect a real protocol is the work of writing a correct XCM message - not the work of replacing a scaffolded architecture with a real one.

The mocks that exist in this codebase are interface placeholders, present because the destination protocols do not exist on Paseo testnet. The infrastructure that exists is production-quality, designed to receive those protocols exactly as they are on mainnet. That is the distinction this build was optimized for.