I want to start with a necessary clarification: Marinade only reports realized APY — meaning actual returns earned by users, not projections or theoretical estimates based on validator bids.
Thanks to everyone who’s contributed to this thread. I want to clarify a few things and correct some misleading claims that have gained traction.
The 37,000 SOL figure circulated refers to estimated missed rewards, not actual loss. No user funds were ever at risk, and no SOL was “siphoned” from stakers. Marinade reports only the realized performance. The APY shown on the app and website reflects actual mSOL appreciation — not theoretical validator bids or ideal conditions.
A small number of validators indeed reduced their bids after winning stake in the auction. That wasn’t the behavior Marinade intended to reward. A fix for that is already live: the bid reduction penalty now punishes this exact pattern by slashing rewards when validators try to undercut their original bid. So far, more than 500 SOL has been redistributed to stakers through this mechanism.
Documentation is available here: Delegation Strategy v2 | Marinade.Finance
Some context around the scale: during the same 126-epoch window, Marinade stakers earned roughly:
- 93,000 SOL from validator bids
- 420,000 SOL in inflation rewards
- 80,000 SOL in MEV
- 2,000 SOL in PSR penalties
- ~595,000 SOL total
The inefficiency identified accounts for around 6 percent of that—not zero, but not what the headlines make it out to be either.
There’s also been discussion around how unstake priority was implemented in the auction logic. Based on internal analysis and community input, including GitHub issue #24, it appeared that the system may unintentionally deprioritize unstaking from low-paying validators.
While the GitHub issue proposed a potential fix, Marinade’s engineering team evaluated it and determined it would not fully resolve it. Instead, the team has decided to address the problem holistically as part of a broader update to the priority system. This is under active development.
The auction was never designed to be perfect from day one. Like any living mechanism, it evolves with new data and behavior patterns. What matters is that the protocol can adapt, which has happened here. This wasn’t about negligence or allowing abuse. It was about identifying where a small set of actors were pushing past the boundaries of what the system anticipated, and tightening those boundaries accordingly.
There’s also a broader point getting lost in the narrative: MEV has existed on Solana long before SAM. Validators have been participating in MEV across the network for years. Marinade didn’t introduce that — it introduced a way to redirect some of that value to stakers. Users could share in MEV for the first time through open bidding, validator competition, and transparent rewards.
That’s not an exploit. That’s alignment.
Marinade also introduced mechanisms like Protected Staking Rewards, which require validators to post a bond that can be used to compensate users in the event of reward underperformance. This is not a free market where validators can disappear after getting stake. There are penalties, obligations, and incentives structured around protecting users.
There are ways SAM can improve — and it will. But the takeaway here isn’t that the system failed. It’s working as intended. Abusive behavior was identified, quantified, and addressed through a protocol-level change. No staker was harmed. All APY was based on actual returns. And the validator auction is already stronger than it was a month ago.
Looking ahead, Marinade is focused on improving the stability of the validator set and reducing unnecessary stake rebalances. This will help increase APY for stakers and ensure more stake flows for honest validators.
Specifically, the team is:
- Designing a validator reputation system to reward consistently reliable operators
- Exploring auction improvements to allow honest validators to earn more
- Continuing to evaluate priority logic, following the recent fix via the BidTooLow penalty
Each of these changes is approached systematically, with a deep understanding of how they impact the broader rebalancing process. That’s the only way to design a complex system that serves users, maintains decentralization, and adapts to validator behavior over time.
Marinade will continue publishing data, evolving the system, and enforcing fairness through code. That’s the promise — and it’s being delivered.