Marinade (mSOL)

Rejected venues wait the longest for re-review; a rejection has to earn another look before the scheduled date.

REJECTED at the 2026-08-15 correction, sleeve cut from 15% to zero until the review answers its questions. The approval bought a specific machine: an algorithmic rebalancer that spread stake across 400+ validators and scored decentralisation into every allocation, the reason we picked mSOL over higher-yielding JitoSOL. That machine no longer exists. MIP-3, approved November 2024, retired it and moved all stake allocation to the Stake Auction Marketplace, a deterministic auction that ranks validators by yield (commission plus bid) and fills top-down, with decentralisation demoted from objective to side constraint. Marinade’s own validators API, read 2026-08-14, shows 46 validators holding mSOL stake, the top five at 41.1% and two hosting providers (Amarutu, Allnodes) near 40% combined. The published kill criterion, validator set below 200, has fired. The comparison has inverted too: Jito’s Steward now spreads stake in near-equal targets across the top 400 validators. The numbers weakened alongside the mechanism. mSOL liquid TVL fell from $1.43B in January 2025 to $181M at this review, per DefiLlama, and identifiable AMM depth is under $2M against the entry’s $25M liquidity floor; the deeper routable figure that floor was measured on could not be reproduced, and the review holds it open. An $8M market sale broke the peg about 15% intraday on 2023-12-12 into deeper books than exist today. What still stands, stated with equal weight: five years on mainnet, no exploit, no loss of staked principal from any contract failure; audits by Kudelski, Ackee, Sec3, and Neodyme on a cadence that runs through 2026; a $250k Immunefi bounty. The 2023 depeg was a liquidity event, not a protocol failure, and delayed unstake worked throughout it. The SAM sandwich-validator episode was misbehaviour inside the delegated set, answered by the DAO blocklisting 50+ validators under MIP-9. Growth moved to Marinade Native ($212M) and the institutional Select line; the operation is healthy, the mSOL float is not. This demotion is not a finding of failure. It is the honest state of a file whose deciding argument described a mechanism that had already been replaced. The primary API’s 46 validators are the validators holding active mSOL stake, not the larger eligible or tracked set; the original below-200 diversification trigger therefore fired. Rejection, not indefinite review, follows until measurable reopening conditions are met.

Kill criteria: the events that revoke this verdict
• Reopen only after active mSOL stake is distributed across at least 200 validators for two consecutive quarters, with the top five below 20% and any one hosting provider below 20%
• Reopen only after a proposed-size mSOL exit quotes below 50 basis points and delayed unstake completes within 7 days in three monthly tests
• Reopen only while mSOL liquid TVL exceeds $250M and independently reproducible executable exit depth exceeds the $25M liquidity floor
• Reopen only after the program upgrade authority, pause powers, and Council signers and 4-of-7 threshold are reproduced from current on-chain state
• Any active-validator count below 200, mSOL/SOL discount beyond 2% for 48 hours, delayed unstake beyond 7 days, or unaudited delegation change keeps the protocol rejected

The research file

The fired criterion

The approval said Marinade distributes stake across 400+ validators and set a kill line at the validator set falling below 200. Marinade’s own validators API, validators-api.marinade.finance, read 2026-08-14, shows 46 validators with mSOL stake, out of 697 tracked, holding 2.34M SOL of liquid stake. Solana Compass independently shows 44 staked validators at epoch 1016 with 2.40M SOL. Including Marinade Native, the union reaches 77 validators. Unless the prior review meant something else by validator set, for instance the 697 tracked or eligible bidders, the kill criterion reads as already breached at the time of, or shortly after, the 2026-07-31 review.

Concentration inside the liquid pool is high. The top 5 validators hold 41.1% of mSOL stake. The top two, SOLGirl and Moise, both hosted by Amarutu Technology in the Netherlands, hold 18.4% between them; three of the top six run on Allnodes in Germany, about 22% combined. The 15% per-validator TVL cap is the only hard brake, and 15% of TVL is a wide cap for a fiduciary product.

The mechanism behind the number changed eight months before the prior review. The thesis leaned on an algorithmic rebalancer weighting performance, commission, and decentralisation; that describes the pre-MIP-3 system. Since November 2024 all allocation runs through the SAM auction, yield-ranked and filled top-down, with decentralisation demoted from objective to constraint. It is still rules-based, but the anti-concentration character the approval priced in is gone from the objective function. The claim that Marinade runs the most explicitly anti-concentration delegation strategy among major Solana LSTs is not supportable post-SAM: the auction channelled stake to sandwich-attacking validators at enough scale that the DAO had to blocklist more than 50 of them, and two hosting providers account for roughly 40% of current mSOL stake.

The Jito contrast has inverted on delegation breadth. Jito’s Stakenet Steward docs say it evaluates 1,000+ validators every 10 epochs and delegates to the top 400, num_delegation_validators equals 400, with roughly equal target allocations. On raw validator count and per-validator weight, JitoSOL’s distribution is now wider and flatter than mSOL’s. The surviving argument against Jito is the client-software monoculture, the Jito client running on more than 95% of network stake, not delegation distribution.

The mechanism

mSOL is an exchange-rate token, not a rebasing one. Docs say mSOL increases in value every epoch relative to SOL; the rate is read from the contract, and Marinade quotes APY as a 30-day simple moving average of the 14-day rate change. Deposits mint mSOL at the current rate; rewards raise the rate rather than the balance.

Delegation is no longer the scored algorithm the approval described. Marinade’s original strategy scored validators on performance, commission, and decentralisation, a fork of the Solana Foundation’s stake-o-matic with code public in the delegation-strategy-2 repository. Docs say that after MIP-3, approved November 2024, all stake allocation moved to the Stake Auction Marketplace: validators bid a share of their revenue, Marinade ranks all eligible validators by max_yield, commission plus bid, and allocates stake top-down until all available stake is distributed, rebalancing once per epoch by bot. It is still algorithmic in the narrow sense, a deterministic ranking with no human picking winners, but the objective function changed from decentralisation-weighted scoring to a yield-first auction with decentralisation as a side constraint. The constraints, per docs as of epoch 946: a per-validator cap of 15% of Marinade’s TVL, no allocation that pushes a hosting provider above 30% of Solana network stake, no allocation that pushes a country above 40% of network stake, plus eligibility gates and a DAO blocklist.

Since February 2024, Protected Staking Rewards has validators post SOL bonds; docs say the bond compensates stakers if a validator underperforms or raises commission mid-epoch. Neodyme audited the Validator Bond program in 2024.

Marinade Native, launched 2023, is separate from mSOL: no token is minted, the user’s stake accounts are delegated across the same SAM validator set, and docs describe it as non-custodial. Marinade Select is a third, institutional line. DefiLlama tracks the liquid product and Native as separate protocols.

The unstake paths, per the docs fees-and-pricing page read 2026-08-14: delayed unstake of mSOL deactivates at the next epoch boundary and is claimable after about one epoch, 2 to 3 days, at a 0.2% fee. Instant unstake of mSOL is now described as a swap to SOL at market price with no protocol fee and market price impact; an older docs page described a protocol-owned mSOL/SOL pool with a 0.1% to 9% formula fee, and the current page does not, so which mechanism serves a large exit today is not clear from docs. Native unstakes free on the delayed path and instantly via a liquidity-provider marketplace at a dynamic 0.10% to 0.40% fee. Deposits carry no fee on any product; docs say Marinade takes no cut of staking rewards and earns from the validator side of SAM instead, though a 2026 Neodyme audit covers deposit fee functionality, which sits oddly against the no-deposit-fee page.

Who controls it

Docs and the Marinade blog say the main liquid-staking contract’s upgrade authority was moved from a 13-party community multisig requiring 6 signatures to MNDE holders governing through Realms. Third parties, Messari, repeat this. The authority was not independently read on-chain for this file.

Governance is two-tier per docs: MNDE holders vote by locking MNDE on Realms with a 30-day unlock, and a Council, a 4-of-7 multisig, holds Realms configuration and admin authorities over various programs. Day-to-day admin power is a small multisig even though the headline upgrade authority is token governance. Signer identities could not be verified from docs.

Pause powers could not be verified at all. No docs page was found that states who can pause deposits, unstaking, or delegation, or whether a pause exists. For mSOL the protocol’s program-derived address controls the pooled stake accounts; for Native and Select, docs say the design is non-custodial and users retain control of their SOL, but the exact split of staker versus withdraw authority is not spelled out in the pages reached.

Marinade Select, launched 2025-05-21 by Marinade Labs, is native staking across a curated, KYC-verified validator set that bans front-running and sandwiching, with SOC 2 Type 2 compliance and custodian integrations with BitGo, Zodia, and Copper. Canary Capital’s Solana ETF, SOLC, launched 2025-11-18, stakes 100% of assets through Select with BitGo custody. Messari reports Select TVL grew 205.5% quarter over quarter to 2.7M SOL in Q4 2025 and is the protocol’s primary growth engine. The relevance: the flagship is drifting institutional and native, and mSOL is no longer where Marinade’s growth is.

The record

Marinade has run on mainnet since August 2021, the first Solana LST. The MNDE token launched November 2021 as a fair launch with no ICO, and the on-chain DAO dates from April 2022.

The TVL history, per the DefiLlama API in USD on 2026-08-14: the liquid product peaked at $1.88B on 2021-11-08, was crushed to $64M at end-2022 in the FTX winter, rebuilt to $1.43B in January 2025, and stands at $181M today. Marinade Native peaked at $1.21B in October 2025 and holds $212M today. JitoSOL, for scale, holds $757M today. Part of the 2025-2026 fall is SOL price, but mSOL has also lost share to JitoSOL and to Marinade’s own Native and Select lines. The mSOL float clients would exit into is a fraction of what it was at review-adjacent peaks.

No protocol exploit was found. No source consulted records a loss of staked principal from a Marinade contract failure in five years of operation.

Three incidents did happen. On 2023-12-12 a holder market-sold roughly $8M of mSOL into thin DEX liquidity; third parties report an intraday dislocation of about 15% against SOL, with some coverage saying nearly 20%. The peg restored quickly, but the wick liquidated leveraged mSOL borrowers on marginfi and Kamino; Solend was unaffected because its oracle ignored the LST market price. The root cause was secondary-market depth, not the contract. The 2%-for-48-hours kill criterion is calibrated to duration, so this event, repeated, would not trip it; it would still hurt any client forced to exit during the wick. Second, the SAM sandwiching episode of 2024-2025: after SAM made stake purchasable, validator operators complained publicly that yield-maximising auctions rewarded validators running sandwich attacks and sybils, with critics claiming about 2.7M SOL had flowed to questionable operators. The DAO responded with MIP-9, blocklisting first 73 proposed and then more than 50 confirmed malicious validators, using Ghostlogs and on-chain analysis. That is validator misbehaviour inside Marinade’s delegated set, mitigated by governance blocklist after the fact. Third, in October 2023 Marinade blocked UK front-end access over FCA promotion rules, with withdrawals still possible via SDK; a regulatory posture note, not a loss event.

The audit cadence, per the docs audit page: Kudelski Security 2021, Ackee Blockchain 2021, a Neodyme code review 2021, Neodyme and Sec3 in 2023 on a contract upgrade, Neodyme 2024 on the Validator Bond and Protected Staking Rewards programs, and Neodyme 2026 on canonical stake accounts and deposit fees, finding no issues at any severity level. The recurring Neodyme cadence through 2026 is a genuine positive.

The Immunefi bounty has been live since 2021-12-01: a $250k maximum for critical, $50k minimum at 10% of affected funds, high capped at $15k, paid in mSOL and MNDE, no KYC. Against $181M of liquid TVL the cap is about 0.14% of TVL, thin by the standard applied to StakeWise at $200k against $700M.

The chain

The Solana chain verdict is approved-with-limits with two watch items, and the current readings on both have worsened. Third parties report about 795 active validators in late January 2026, down 68% from the 2,560 peak of March 2023, with roughly 900 by another mid-2026 count; the decline is attributed to node economics squeezing small operators. The Nakamoto coefficient read 20 in late January 2026 against 31 in March 2023, with some mid-2026 coverage citing 19.

Marinade’s interaction with those watch items is mixed, and worse than the prior memo assumed. The SAM network-level constraints, hosting provider under 30% and country under 40% of network stake, are real anti-concentration brakes at the chain level, and the old scored strategy’s decentralisation bonus is gone. In practice mSOL’s 2.34M SOL now sits with 46 validators, about 0.5% of network stake, so its power to move the chain-level Nakamoto coefficient either way is small. Direction of travel matters more: a yield-first auction channels stake to the highest bidders, which favours sophisticated, often co-hosted operators; two of the top three mSOL validators share one Netherlands hosting provider. Marinade’s marketing still claims decentralisation as a goal, but the mechanism no longer optimises for it, it only bounds the damage. Net: Marinade neither meaningfully improves nor materially worsens the chain’s watch items today, and the claim that holding mSOL is a vote against concentration is now mostly narrative.

The exit

Delayed unstake is the reliable path: deactivation at the next epoch boundary plus a claim, 2 to 3 days, 20 basis points, with capacity limited only by the stake pool itself. That is well inside the 7-day kill criterion under normal operation.

Secondary depth is thin and much thinner than at the prior review. Per DefiLlama yields data on 2026-08-14, the largest identifiable mSOL DEX pool is Raydium WSOL-MSOL at about $674k, then Kamino SOL-MSOL at $610k, with Orca pairs under $200k each; identifiable AMM pairs total under $2M. Most mSOL in DeFi sits in lending markets, Save at $17.9M and Kamino Lend at $17M, which is loop collateral, not exit depth. Aggregators, Jupiter and Sanctum’s router and reserve, source deeper effective liquidity than raw pool TVL suggests, and the entry’s $25M minLiquidityUsd was presumably measured through such a lens, but a $25M figure could not be reproduced from pool data today. A seven-figure market sell is what broke the peg 15% in December 2023, and visible pool depth today is smaller than it was then.

Instant unstake capacity is unverified. Docs now route mSOL instant exits to market swaps, so instant capacity equals that secondary depth plus whatever the liquidity-provider marketplace absorbs; no stated capacity figure was found anywhere. The practical consequence for any future sleeve: exits at size should default to delayed unstake, instant exit is a convenience for small amounts, and the 2%-for-48-hours discount criterion should be monitored on aggregator quotes at trade size, not on pool mid-prices.

The watch list

The observable readings the re-review should track, restated against what is measurable today. Validator count: the validators API count of validators with marinade_stake above zero, reading 46 against a kill line of 200. Rebalancer discretion: the old kill line about the rebalancer becoming discretionary needs a SAM-era definition; the observable events are eligibility or blocklist changes outside DAO process, the 15% TVL cap raised, the hosting-provider or country constraints loosened, or a Select-style curated set replacing the open auction for mSOL. Secondary discount: mSOL/SOL under 0.98 for 48 hours on executable aggregator quotes at size, around 10k SOL, and wicks like 2023-12-12 logged even when shorter than 48 hours. Unstake queue: over 7 days, or delayed-unstake claims failing an epoch.

Depth: identifiable exit depth, an aggregator quote at size or DEX pair TVL, against the entry’s $25M floor; today’s pool-level reading is far below it. TVL: mSOL liquid TVL is $181M and falling, and a slide through about $100M puts the $250k bounty and audit economics against a still-material client exposure while depth shrinks further. Concentration: top-5 validator share of mSOL stake, 41.1% now, and single-hosting-provider share, Amarutu 18.4% and Allnodes about 22%; a single validator reaching the 15% TVL cap is an observable event. Governance: Council multisig composition changes, any upgrade executed with less than DAO process, or a pause event, once the review learns whether pause powers exist. Product drift: Marinade deprioritising mSOL in favour of Native and Select is already visible in the growth numbers, and an announcement of mSOL sunset, migration, or fee changes is a hard review trigger.

Open questions

Pause powers: no documentation was found on who can halt deposits, unstaking, or delegation, or whether a pause mechanism exists at all. The on-chain upgrade authority: the claim that the DAO controls upgrades rests on docs plus third parties; the program’s upgrade authority was not read on-chain this pass. The Council multisig is 4-of-7 per docs, but signer identities are unpublished as far as was found. The stake authority layout for Native and Select, who holds staker versus withdraw authority, is not spelled out.

The instant unstake mechanism today is unresolved: older docs describe a protocol-owned pool with a 0.1% to 9% formula fee, current docs describe a plain market swap, and which is live, with what capacity, is unverified. The entry’s $25M minLiquidityUsd could not be reproduced from visible DEX pool data, which shows under $2M in identifiable AMM pairs; either it was measured on aggregate routable depth, or depth has collapsed since 2026-07-31, and the entry needs a measurement method pinned to it. The 2026 Neodyme audit scope mentions deposit fee functionality while the fees page says no deposit fees on any product, possibly a new canonical-stake-account fee path; unresolved. Exact dates for the MIP-9 blocklist votes and the 50-plus validator blacklist span 2024-2025 in coverage; the forum holds the record.

Sources

The claims above trace to these. Where a number could not be independently verified, the thesis says so.

• Marinade Stake Auction Marketplace docs · primary · accessed 2026-08-15
  Supports: 100% SAM allocation, yield-ranked delegation, eligibility and concentration constraints
• Marinade validators API · primary · accessed 2026-08-14
  Supports: active mSOL validator count, validator stake allocation, hosting concentration
• DefiLlama Marinade liquid staking TVL · secondary · accessed 2026-08-14
  Supports: mSOL liquid TVL, historical TVL
• Jito Steward program docs · primary · accessed 2026-08-15
  Supports: peer delegation method, validator target comparison
• Marinade MIP-9 validator blocklist · primary · accessed 2026-08-15
  Supports: malicious validator blocklist, governance response, SAM incident
• SolanaFloor mSOL depeg report · secondary · accessed 2026-08-15
  Supports: December 2023 sale, mSOL discount, liquidations
• Marinade audits page · primary · accessed 2026-08-15
  Supports: audit firms, review dates, 2026 audit scope

Inherited controls

The verdict above grades the protocol layer. Every position also inherits the asset it holds and the chain it settles on. The least safe layer sets the position’s grade, and the position table names which one that is.