Skip to content
PROTOCOL RESEARCH

Spark risk

Sky's lending and savings Star, still clean at $4.598B: audit roster re-verified this run (not carried), no exploit in a 612-record hacks database, and the Liquidity Layer's direct Ethena position measured at about $100 against a $1.1B governance target. But this run broke four of the draft's claims. The Ethena wind-down is half the size the draft reported in reverse: the position peaked at $953.65M combined sUSDe plus USDe in August 2025, not $475M, because the draft read only one of the two legs, so governance has demonstrably drawn that line to 87 percent of target within the last twelve months. The $260M "USDT" line in the Liquidity Layer is not USDT: the ALM Proxy's on-chain USDT balance is exactly zero, and the figure matches both the $260M Anchorage allocation in Sky's June financials and the $260M OTC Bitcoin loan book CoinDesk reported, so roughly 6 percent of headline TVL is term credit presented under a stablecoin ticker. The rsETH dodge was routine low-utilization cleanup per Phoenix Labs, not a bridge-risk call, and the draft's "risk process" framing is withdrawn. SPK circulating supply is available after all at 3.038B of 10B, so that line is no longer _refresh_.
SolidResearched Aug 5, 2026

Scorecard

strong: Audits & contractsRe-verified this run, not carried: ChainSecurity on SparkLend Advanced (eight price oracles plus cbBTC/rETH/weETH KillSwitch peg oracles) and ALM Controller v1.0.0-v1.7.0 and v1.10.0; Cantina on every released version v1.0.0-v1.10.0; Certora on v1.8.0 and v1.9.0. Immunefi confirmed at $5M max with a $50k Smart Contract Critical floor, paid in DAI, live since 2023-11-01. One live component, ALM_PROXY_FREEZABLE, comes from a different repo than the audited range; it holds zero balances today.
watch: Admin controlSky governance holds the keys through Spark proxy spells inside Sky executive votes, behind a pause delay that oscillates by spell: 30h in 2024, 48h May 2025, 24h Oct 2025 to 2026-05-01, 48h since mid-May 2026. Not fixed and not a ratchet. Underneath, a RELAYER moves capital only between venues with a governance-set rate limit, and a FREEZER can revoke a relayer or detach a controller from the ALM Proxy.
strong: OracleAave v3 oracle architecture plus a ChainSecurity-audited custom set confirmed this run from the audit scope: fixed-price, capped-price, wstETH, rETH, weETH, rsETH, ezETH and spETH oracles, three KillSwitch peg-ratio depeg oracles, and a MorphoBlue upgradable oracle. The specific feed provider was not re-confirmed: DeFiLlama carries no oracle field for SparkLend, and the July wording of Chainlink is _refresh_, since the same assumption proved wrong for Sky, which runs Chronicle.
watch: Liquidity & exitPool exits need idle liquidity and Savings rides PSM liquidity. Two Liquidity Layer lines that look like cash are credit: the $260M USDT line is not held by the ALM Proxy at all (balanceOf reads zero) and matches the Anchorage OTC Bitcoin book, and $19.65M of the $20M USDC line is the Arkis Spark Prime vault. About $280M of the SLL's $1.584B Ethereum book is term credit that does not redeem on demand.
strong: Yield (real vs emission)Measured live today: Savings USDS 3.52 percent, USDC 3.52 percent, USDT 2.75 percent, ETH 1.75 percent; SparkLend USDS supply 2.04 percent base with a separate reward-driven USDS pool at 3.23 percent. Funded by stability fees, T-bill RWAs and deployment carry. With the Ethena line at dust, basis yield is not currently a component, so the rate leans on plainer sources than in July. SPK emissions sit on top of deposit yield, not inside it.
watch: Holder concentrationCirculating supply is available and no longer _refresh_: 3,038,479,522 SPK of 10B total (30.4 percent), $46.6M market cap at $0.0153, cross-checked on two price sources. Schedule unchanged: 65 percent farmed over ten years, 23 percent ecosystem, 12 percent team. On-chain top holders remain unavailable (Etherscan Pro-gated, re-tested this run). The concentration that matters more for depositors is Sky itself, which supplies much of the deposit base.
strong: Track recordNo Spark entry in DeFiLlama's 612-record hacks database and an empty hacks field on the protocol; an X sweep run this session across 2026-06-01 to 2026-08-05 returned no incidents. Collateral is blue chip: wstETH $2.39B, LBTC $211M, WBTC $201M, cbBTC $178M, WETH $103M, weETH $76M against $1.806B borrowed. Zero losses in the April 2026 Kelp exploit, though the rsETH exit that caused it was low-utilization cleanup rather than a bridge-risk judgement.
🟢 strong🟡 watch / caveat🔴 weak / fund-loss risk
Verdict is a gate (worst flaw wins), not an average. Our read, not financial advice.

Live facts

auto-sourced now
TVL$3.75B
30d↑2%
Audits2
Last hacknone

DeFiLlama + our exploits feed. Cross-check the dated report against today.

TL;DR

Spark is Sky's lending and savings Star. SparkLend is an Aave v3 fork, Savings passes through the Sky Savings Rate, and the Spark Liquidity Layer routes Sky reserves across venues. Nothing broke this month. There is no Spark entry in DeFiLlama's 612-record hacks database, an X sweep run this session across June to August 2026 returned nothing, and TVL is essentially flat at $4.598B.

The headline this month is a measurement, and it does reduce risk. The Liquidity Layer's direct Ethena position, which our dependency map carries as a target of up to $1.1B under a cap of 20 percent of USDe supply, is not drawn. The ALM Proxy holds 80.912135 sUSDe, about $100, verified independently on-chain rather than taken from an aggregator, and it has held dust since November 2025.

Two things qualify that, and both come from re-doing work the draft carried rather than checked. First, the wind-down is twice as large as the draft said, in the direction that matters: the combined sUSDe and USDe position peaked at $953.65M on 2025-08-02, not the $475M the draft reported, because the position always had two legs and only one was read. Sky governance has drawn this line to 87 percent of its target within the last twelve months. Treat the current dust as a decision that can be reversed, not as a structural limit.

Second, the OTC Bitcoin credit book is not the invisible thing the draft described. It is visible, and it is probably already inside the TVL number at the top of this report, wearing a stablecoin ticker.

Checklist

Audits and contracts. Re-opened this run rather than carried forward, which is the correction the draft most needed. SparkLend inherits the Aave v3 codebase. ChainSecurity audited SparkLend Advanced, whose scope is now confirmed in detail: eight price oracles (fixed-price, capped-price, wstETH, rETH, weETH, rsETH, ezETH, spETH), three KillSwitch peg-ratio depeg oracles for cbBTC, rETH and weETH, a MorphoBlue upgradable oracle, and two interest rate strategies. On the ALM Controller, Cantina audited every released version from v1.0.0 to v1.10.0, ChainSecurity covered v1.0.0 to v1.7.0 and v1.10.0, and Certora covered v1.8.0 and v1.9.0. No named firm is invented. The Immunefi bounty is confirmed at a $5,000,000 maximum with a $50,000 floor for Smart Contract Critical, calculated as 10 percent of funds directly affected, paid in DAI, live since 2023-11-01.

One honest gap. Spark's public ALM Controller releases stop at v1.10.0 (2025-02-17), but the address registry lists a live ALM_PROXY_FREEZABLE sourced from a different repository at its own v1.12.0. That component sits outside the audit range verified above. It holds zero balances today, so nothing is at risk in it, but do not read the audit roster as covering every deployed SLL contract.

Admin control. Unchanged in kind, and still the main structural risk. Parameter changes, listings and upgrades execute through Spark proxy spells inside Sky executive votes, behind the Sky pause delay. That delay is chosen per spell and moves in both directions: 30 hours in 2024, 48 hours in May 2025, 24 hours from October 2025 through 2026-05-01, and 48 hours since mid-May 2026. It is not a fixed parameter and not a one-way ratchet, so do not plan around a specific number.

Below governance sit two operational roles, confirmed from Spark's own documentation this run. A RELAYER moves capital between approved venues, constrained by approved integrations, balances, slippage checks and rate limits; each rate limit carries a maximum amount and a per-second regeneration slope, and a venue is unusable until governance sets a non-zero limit for its key, so an unapproved destination simply reverts. A FREEZER can immediately revoke a compromised relayer and, on the freezable proxy, detach a compromised controller entirely. A captured relayer can shuffle capital inside the whitelist, not out of it, which is the point of the design.

Oracle. The custom oracle work is confirmed through the ChainSecurity scope above, which is stronger evidence than a docs page. What is not confirmed is the feed provider. DeFiLlama carries no oracle field for SparkLend, and this file previously described Chainlink feeds without checking. That same assumption proved wrong for Sky one day ago, where the primary feed is Chronicle, not Chainlink. The oracle architecture is sound and audited; the name of the price source is marked for refresh rather than repeated.

Liquidity and exit. SparkLend exits need idle pool liquidity, same as any Aave-style market, and Savings exits ride PSM liquidity. The Liquidity Layer is the part worth re-reading each month, and this month it reads differently than the draft claimed.

Measured today it holds $1.584B on Ethereum. The largest lines are Spark's own receipt tokens, meaning capital recycled back into SparkLend: spUSDS $458M, spUSDT $311M, spDAI $186M, spPYUSD $111M, spUSDC $10M. Then two lines that look like cash and are not.

The $260M USDT line is the important one. The ALM Proxy's on-chain USDT balance is exactly zero. Both Curve LP positions the adapter tracks are worth under five dollars, so they cannot explain it either. The figure therefore comes from the one remaining path in the adapter, an external Anchorage allocation folded into mainnet TVL. Its size matches the $260M Anchorage line in Sky's June 2026 agent allocations and the $260M of outstanding OTC Bitcoin loans CoinDesk reported on 2026-08-02, and CoinDesk states those loans are issued through Anchorage to borrowers including Bitcoin miners. The attribution is an inference, since the Anchorage adapter source returned 404, but three independent figures agreeing at $260M is not a coincidence worth ignoring. Read plainly: about 6 percent of Spark's headline TVL is term Bitcoin-backed credit displayed under a stablecoin symbol.

The $20M USDC line is a smaller version of the same thing. The ALM Proxy holds 19,647,500 units of the Arkis sparkPrimeUSDC1 vault, which matches CoinDesk's roughly $20M of Spark Prime brokerage loans outstanding. So about $280M of the SLL's $1.584B Ethereum book is credit rather than liquid stablecoin. By contrast the $227M PYUSD line is genuinely idle: a direct balance read returns $227,077,651, matching the aggregator almost exactly.

Yield: real versus emission. Measured live today: Savings USDS 3.52 percent, Savings USDC 3.52 percent, Savings USDT 2.75 percent, Savings ETH 1.75 percent. On the lending side SparkLend USDS supply pays 2.04 percent base, with a separate USDS pool at 3.23 percent that is reward-driven. The savings rate is real revenue from stability fees, tokenized T-bills and deployment carry, not emissions. One July component is absent now: with the direct Ethena position at dust, basis yield is not on Spark's balance sheet, so the rate rests on plainer sources than it did a month ago. SPK farming emissions sit on top of deposit yield, not inside it.

Holder concentration. The draft marked this unverified for a second month. Half of it is verifiable and now measured: 3,038,479,522 SPK circulating against 10,000,000,000 total, so 30.4 percent of supply is out, at $0.0153 for a $46.6M market cap, cross-checked on two independent price sources. The schedule is unchanged: 10B at genesis with Sky retaining the right to mint more in extreme circumstances under the Atlas, 65 percent farmed over ten years on a decaying schedule, 23 percent ecosystem, 12 percent team on a 12-month cliff releasing 25 percent with the rest over three years. On-chain top holders remain genuinely unavailable, since Etherscan's endpoint is Pro-gated and was re-tested this run. For depositors the concentration that matters more is Sky itself, which supplies much of the deposit base, so Spark risk correlates with Sky and USDS.

Track record. Still clean, stated with the right evidence. There is no Spark entry anywhere in DeFiLlama's 612-record hacks database and the protocol's own hacks field is empty. An X sweep across 2026-06-01 to 2026-08-05, run in this session rather than quoted, found no incidents, depegs, oracle failures, liquidation problems or governance controversy. Collateral measured today is blue chip: wstETH $2.39B, LBTC $211M, WBTC $201M, cbBTC $178M, WETH $103M, weETH $76M, against $1.806B borrowed.

The rsETH episode needs correcting, because the draft turned a good outcome into a claim about process. The facts hold: Spark halted new rsETH supply on January 29, 2026, the same day Aave launched rsETH eMode at 93 percent LTV, and took zero losses when Kelp's bridge was exploited on April 18, 2026 for 116,500 unbacked rsETH worth about $293M, an event that left Aave with an estimated $195M to $230M of bad debt. But the reason Spark exited is documented, and it was not a bridge-risk assessment. A Phoenix Labs governance post on January 16, 2026 flagged low rsETH utilization concentrated in a single wallet, and the delisting was routine cleanup. The outcome was excellent. The foresight claimed for it was not, and this report will not bank a safety argument on it.

One limit on the clean record. A hacks database tracks exploits, not bad debt. "No bad debt" here rests on the absence of any report of it across the sources checked, not on a positive solvency measurement, and no one should read it as an audited statement of zero.

Worst case

Three paths, in order of how much they should worry you.

A Sky governance failure remains the largest. A captured or malicious executive spell can change SparkLend parameters or upgrade contracts once the pause delay expires, and that delay is a governance-set number that has been as low as 24 hours within the past year. Because deposits, governance and much of the liquidity all trace back to Sky, Spark cannot be safer than Sky.

Counterparty contagion through the Liquidity Layer is second, and it is smaller today than our dependency map implies, but less reassuring than the draft made it sound. The Ethena line is at dust now. It also stood at $953.65M twelve months ago, which is 87 percent of the $1.1B target, and governance can restore it without a new vote on the target. The cap headroom of roughly $777M against the $3.8828B USDe supply our 2026-08-05 Ethena refresh measured is a live authorisation, not a theoretical one. The dependency map edge from usds to usde should be labelled as cap, not exposure, and should not be deleted.

The OTC credit book is third and the least inspectable. If Bitcoin gaps and a borrower fails, the loss lands on Sky reserves. The size is disclosed and cross-checks three ways, and the counterparty is a chartered custodian rather than an anonymous desk, which is better than the draft implied. What no depositor can check is the collateral, the loan-to-value, the margining terms or the individual borrowers. At $260M against $4.598B that is survivable. At the stated $1B year-end target, against a protocol whose revenue has fallen from about $80M to about $20M, it stops being a footnote and becomes the thing that decides this rating.

Bottom line

Solid, and the rating is stated with its gate reasoning rather than assumed. No dimension on this scorecard is red. The unverifiable surface is bounded: roughly $280M of term credit, about 6 percent of TVL, with a disclosed size that reconciles across three independent sources and a named regulated counterparty. Against that sits an audited Aave v3 base re-verified this run, real yield near 3.5 percent, blue-chip over-collateralised lending, a public and timelocked admin path, and no exploit in the protocol's history.

What changed this month is mostly that we now know more, not that Spark got safer. Two claims the draft treated as good news are weaker than presented. The Ethena de-risking is real but reversible, and it undid a position far larger than the draft recorded. The rsETH dodge was luck dressed as process.

Three things would flip this to caution: the OTC book growing toward its $1B target without disclosure growing with it, the Ethena line being redrawn toward its cap, or a first bad-debt event in SparkLend. Size Spark as correlated Sky exposure with a counterparty layer attached, not as an independent protocol, and do not read the $260M USDT line on any dashboard as cash.

Data appendix

  • TVL: $4,598,280,620 (DeFiLlama "spark" parent, chain sum). Down about 0.7 percent from ~$4.63B on 2026-07-03, essentially flat. Children before parent dedup: SparkLend $3.676B (the draft's $3.643B is stale), Spark Liquidity Layer $1.596B, Spark Savings $1.135B. Borrowed $1.806B. Protocol hacks field empty. 90d and ATH series: refresh.
  • Audits, re-verified this run: ChainSecurity, scope confirmed for SparkLend Advanced (fixed-price, capped-price, wstETH, rETH, weETH, rsETH, ezETH, spETH oracles; cbBTC, rETH and weETH KillSwitch peg-ratio oracles; MorphoBlue upgradable oracle; two interest rate strategies) plus ALM Controller v1.0.0-v1.7.0 and v1.10.0 and the Spark Liquidity Layer report of 2024-10-22. Cantina: every released ALM Controller version v1.0.0-v1.10.0, plus the Spark Liquidity Layer report of 2024-10-23. Certora: v1.8.0 and v1.9.0. Immunefi: $5,000,000 maximum, $50,000 Smart Contract Critical floor, 10 percent of funds directly affected, paid in DAI, launched 2023-11-01; full band SC Critical $50k-$5M, SC High $10k-$100k, Web/App Critical $5k-$50k.
  • Audit coverage gap: public spark-alm-controller releases end at v1.10.0 (2025-02-17), while the address registry lists ALM_PROXY_FREEZABLE 0xe5c6318456a7Cb6f74f93B4eee4616dB5fcef699 from a separate repository at v1.12.0. Balances checked and zero.
  • Admin and governance: Spark proxy spells execute inside Sky executive votes behind the Sky pause delay, which is set per spell: 30h (2024), 48h (May 2025), 24h (Oct 2025 to 2026-05-01), 48h since mid-May 2026, per our 2026-08-05 Sky refresh. SLL operations run through a rate-limited RELAYER with a FREEZER able to call removeRelayer and removeController (docs.spark.fi, read 2026-08-05). Addresses from the Spark address registry, confirmed on both the marsfoundation and sparkdotfi mirrors: ALM Proxy 0x1601843c5E9bC251A3272907010AFa41Fa18347E, ALM Controller 0x5c46Fc65855c0C7465a1EA85EEA0B24B601502D3 (v1.10.0), Spark Proxy 0x3300f198988e4C9C63F75dF86De36421f06af8c4, ALM Freezer multisig 0x90D8c80C028B4C09C0d8dcAab9bbB057F0513431, ALM Relayer multisig 0x8a25A24EDE9482C4Fc0738F99611BE58F1c839AB.
  • Ethena line, measured and corrected: ALM Proxy holds 80.912135 sUSDe (Etherscan raw balance 80912135195743362751), about $100. USDe is present at dust rather than absent: 0.0000749 USDe (raw 74896275897803), correcting the draft's claim that no USDe line appears. This matters because the adapter folds pending sUSDe cooldown amounts into the USDe line, so the dust reading also rules out a pending unstake. The second proxy the adapter does not read, ALM_PROXY_FREEZABLE, was checked directly and holds zero of both. History, corrected: the position has two legs and the draft read one. Combined peak $953,651,466 on 2025-08-02 (sUSDe $473.3M plus USDe $475.8M), not "about $475M"; 2025-10-01 was $350.5M (sUSDe $175.35M plus USDe $175.15M), not "about $175M"; dust from 2025-11-01 onward. The $1.1B target (The Block, 2025-01-14) under a cap of 20 percent of USDe supply, worth about $777M against the $3.8828B supply our 2026-08-05 Ethena refresh measured, is live headroom that has previously been drawn to 87 percent.
  • Liquidity Layer composition (measured): Ethereum $1,584,242,592, led by spUSDS $458.2M, spUSDT $311.4M, USDT $260.0M, PYUSD $227.1M, spDAI $185.5M, spPYUSD $111.0M, USDC $20.2M, spUSDC $10.1M, Spark Blue Chip USDC Morpho vault $772k, sUSDe $100. Off-Ethereum: Arbitrum $10.0M, Base $1.0M, Optimism $1.0M, Unichain $40k.
  • The $260M USDT line is not USDT held by Spark (verdict-relevant): proven, the ALM Proxy's on-chain USDT balance is 0, and both Curve LP positions the adapter tracks are worth under $5 (0.797 and 3.878 LP tokens), so neither idle balance nor Curve explains it. Inferred, not proven: the only remaining mainnet path in the adapter is an external Anchorage allocation added via getExports('spark-anchorage'), whose source returned 404 on every path tried. Its size matches the $260M Anchorage line in Sky's June 2026 agent allocations and the $260M of outstanding OTC Bitcoin loans CoinDesk reported on 2026-08-02, which it says are issued through Anchorage.
  • Spark Prime (measured): ALM Proxy holds 19,647,500 units of the Arkis sparkPrimeUSDC1 ERC-4626 vault 0x38464507E02c983F20428a6E8566693fE9e422a9, assuming 6 decimals, which matches CoinDesk's roughly $20M of outstanding Spark Prime loans and accounts for most of the $20.2M USDC line.
  • PYUSD line confirmed genuinely idle: direct balance read returns $227,077,651 against the aggregator's $227,077,653.
  • Morpho and RWA lines, correcting the draft: Morpho vault tokens do appear. Spark Blue Chip USDC shows $772,042 and Spark Blue Chip USDT $6.75. syrupUSDC appears at $0. Aave receipt tokens are near zero (aEthUSDT $5.94, aEthUSDC and aEthUSDS effectively $0). BUIDL-I, USTB, Superstate USCC, Janus JTSRY and syrupUSDT are tracked by the adapter but report no balance.
  • Oracle: Aave v3 architecture with the ChainSecurity-audited custom oracle set above. Feed provider not confirmed: DeFiLlama carries no oracle field for SparkLend, and the prior Chainlink wording is refresh.
  • Holder concentration: SPK 0xc20059e0317DE91738d13af027DfC4a50781b066, 3,038,479,522 circulating of 10,000,000,000 total (30.4 percent), price $0.01535 and market cap $46.6M on CoinGecko, cross-checked at $0.015335 on DeFiLlama and a $46.86M market cap on the protocol record. Schedule: 65 percent ten-year farming, 23 percent ecosystem, 12 percent team (12-month cliff at 25 percent, remainder over three years), with further minting possible in extremis per the Sky Atlas. On-chain top holders: refresh, Etherscan endpoint Pro-gated and re-tested this run.
  • Yields (measured 2026-08-05, DeFiLlama): Spark Savings USDT 2.75 percent ($387.9M), USDS 3.52 percent (Arbitrum $361.4M, Base $12.0M), USDC 3.52 percent (Ethereum $268.3M, Avalanche $12.0M), ETH 1.75 percent ($64.7M), USDG 3.5 percent on Robinhood Chain ($12.8M). SparkLend USDS supply 2.035 percent base ($281.6M) plus a separate reward-driven USDS pool at 3.229 percent ($543.3M). Note there is no mainnet Spark Savings USDS pool in the dataset; the 3.52 percent USDS figure is measured on Arbitrum and Base and matches Sky's sUSDS rate.
  • Incident scan: no Spark record in DeFiLlama's 612-entry hacks database and an empty hacks field on the protocol. Kelp appears at 2026-04-18 for $293,000,000, classified as an infrastructure LayerZero OFT bridge exploit. X sweep run this session across 2026-06-01 to 2026-08-05: no incidents found. No bad debt found, with the caveat that a hacks database does not measure bad debt.
  • rsETH sequence, with the honest cause: Phoenix Labs governance post 2026-01-16 flagged low rsETH utilization concentrated in one wallet; Spark halted new rsETH supply on 2026-01-29, the same day Aave launched rsETH eMode at 93 percent LTV; the Kelp bridge exploit of 2026-04-18 minted 116,500 unbacked rsETH; SparkLend reported zero losses while Aave faced an estimated $195M (Lookonchain) to $230M (CoinDesk) of bad debt. The exit was routine low-utilization cleanup, not a bridge-risk assessment.
  • Recent news: CoinDesk, 2026-08-02: Spark shelved its consumer app for a business-to-business backend model; revenue fell from about $80M in the bull market to roughly $20M; OTC Bitcoin-backed loans at about $260M outstanding against roughly $400M originated with a $1B year-end target, issued through Anchorage to borrowers including Bitcoin miners; Spark Prime about $20M outstanding; Robinhood Earn drew more than $200M in 24 days through a Morpho vault allocating across Ethena USDe, Maple syrupUSDG and Spark spUSDG; about $150M migrated into Uniswap v4 USDS pools against USDT and PYUSD, routing roughly $1.5B in their first 30 days.
  • Credora, corrected: the report at reports.credora.io/spark/latest.pdf is REP-SPARK-20260803-V1 dated 2026-08-03, not the REP-SPARK-20260518-V1 of 2026-05-18 the draft cited. The file is image-only, so its rating remains unread: refresh.

Maintained monthly. Methodology: DeFi Research Instruction v2.

← all protocols