Spark Protocol

No bad-debt event appears in the fetched sources, and the closest live test went the other way. When Kelp DAO's bridge was compromised on 18 April 2026 and 116,500 unbacked rsETH were minted, Aave V3 and V4 faced roughly $195M of potential bad debt while SparkLend was left with $37,300 of frozen residual rsETH, worth 15.32 rsETH, because PhoenixLabs had filed an exit proposal on 16 January 2026 and halted new rsETH supply by spell on 29 January 2026 (KuCoin). Spark's own framing is that its risk pillars are designed to stack so that a failure in any single component, oracle, issuer, liquidator or market liquidity, does not cascade into bad debt (Spark docs). Spark co-founder Sam MacPherson nonetheless cautioned on 19 April 2026 that protocols claiming no rsETH exposure may still carry indirect exposure through user collateral (KuCoin).

Not the utilisation curve alone. SparkLend rates respond to supply and demand, but transparent rates act as a fixed rate in the short term and change based on Sky Governance, and the Spark Liquidity Layer automatically moves USDS, USDC and USDT into and out of SparkLend based on target borrow rates, utilisation and available inventory at other venues (Spark docs). In practice Sky Governance sets the target and the SLL enforces it by supplying or withdrawing inventory, which is closer to a central-bank standing facility than to a pure market-clearing curve. Per-market supply and borrow caps are also configured by Sky Governance (Spark docs).

If a position's health factor drops below 1 but stays above 0.95, up to 50% of the position's debt can be covered in a liquidation call; at or below 0.95, up to 100% can be covered, with the collateral transferred depending on debt covered, oracle prices and the applicable liquidation bonus (Spark docs). Governance has a softer tool as well: the borrowing power of any asset can be reduced to 0% without impacting existing borrowers, which lets it phase out a collateral type without forcing liquidations as the first response (Spark docs). Forced liquidation is still used when required, as with the remaining BTC e-mode positions on 8 June 2026 (Spark docs).

Spark and Sky publish an explicit six-layer loss waterfall: internal junior risk capital at the Prime level, then Prime-external junior risk capital, then external senior risk capital via srUSDS (planned, not yet deployed), then the Sky Surplus Buffer, then Sky's Aggregate Surplus Buffer, then a SKY token backstop where Sky mints SKY to recapitalise (Spark docs). Only if all of that is exhausted are residual losses socialised equally across all USDS holders, including Spark Savings vaults (Spark docs). Spark holds over $35M of stablecoin equity capital at Layer 1 and describes the aggregate protection as several hundred million dollars (Spark docs). Credora's model puts the Probability of Significant Loss at 0.39% for the ETH vault and 0.46% to 0.52% for the stablecoin vaults, rising to 1.03% for stUSDS, as of 29 June 2026 (Spark docs).

Materially less than the separate brand suggests. Spark borrows from Sky's stablecoin reserves and operates as a two-sided capital allocator deploying that capital across DeFi, CeFi and RWAs (Spark docs), and its Q1 2026 report calls SKY the parent protocol and describes the Spark treasury as junior capital required by SKY to access the SKY credit line (Spark Q1 2026 report). Spark's own governance process is a process for amending the Spark Agent artifact inside the Sky Atlas, filed in the Spark Prime section of the Sky forum and voted on Snapshot at s:sparkfi.eth, with a Spark Risk Council review that can reject a proposal before it reaches a vote (Spark governance docs). Hindenrank scores this dependency as the number one risk, noting that solvency is backstopped by Sky's $6.5B reserve and that this creates single-entity systemic risk (Hindenrank).

Mostly from the Spark Liquidity Layer, not from SparkLend. In Q1 2026 the SLL ran $1.93B of average deployment at a 5.8% average APY for $27.62M of gross returns and $3.05M net, on a captured spread of 0.64%, while SparkLend contributed only $156k of revenue, down from $215k in Q4 (Spark Q1 2026 report). Distribution rewards added $3.31M, led by sUSDS at $1.66M and stUSDS at $710k (Spark Q1 2026 report). A credit analyst should therefore treat Spark primarily as a leveraged stablecoin carry book with a money market attached, not as a lending protocol whose earnings scale with borrow demand.

SPK is used for signalling and Snapshot voting on the Spark Agent artifact, with eligible proposers being holders of 1% of supply, 100M SPK, or nested contributors such as Phoenix Labs (Spark governance docs). Of the 10 billion genesis supply, 65% sits with Sky for a 10-year farming campaign emitting 1,625M per year in years one and two and tapering to 203.13M per year by year seven, 23% is ecosystem and 12% is team on a 12-month 25% cliff then three-year vest, and Sky retains the ability to mint additional SPK under extreme circumstances per the Sky Atlas (Spark docs). The 17 June 2026 unlock alone was 769,050,000 SPK, about 24.1% of market capitalisation, following a 17 May 2026 unlock that preceded a 25.6% price decline within 12 days (Crypto Daily).