Abstract
Move PSR downtime coverage to 100% validator-covered, removing the DAO’s remaining share. Validators cover the full charge across the 0%-99% uptime band for stake delegated through the SAM auction. The 1% grace period is unchanged and not in scope.
Background
PSR backstops Marinade’s commitment to deliver at least stake-weighted cluster-average yield to stakers every epoch. When a validator underperforms, the shortfall is paid from that validator’s bond first, and from the DAO reserve beyond that.
Where the line sits between the two has moved before. Under the original design the DAO covered the 0-80% uptime band and validators covered 80-99%. MIP.2: Adjustment of PSR Coverage moved it to a DAO 0-50% / validator 50-99% split, effective epoch 696, on three grounds: the DAO had paid more than 50 SOL across 10 epochs as delegations grew and validator turnover increased, an even split distributes downtime responsibility more fairly, and a larger validator share sharpens the incentive to run reliable infrastructure.
This proposal extends that same reasoning to its conclusion.
Proposed Change
Validators cover the full downtime charge, 0-99% uptime. The DAO reserve stops absorbing any part of a downtime shortfall. The grace period stays at 1%.
Marinade Labs implements the change and announces the effective epoch, which must be no sooner than 3 epochs after the announcement so validators can review their monitoring and bond levels first. Marinade Select’s yield floor is a separate mechanism, already fully bond-funded, and is out of scope.
Rationale
- Uptime is the validator’s responsibility, and only the validator’s. Hardware, hosting, monitoring, client version, on-call coverage: Marinade controls none of it and cannot shorten an outage on an operator’s behalf. The DAO has already applied that principle once: [MIP-19]( MIP-19: Improving SAM - Auction Stake Priority, Bond Risk Reduction Mechanism, Higher Validator Caps ) charges validators the entire cost of forced undelegation when their bond falls short. Downtime is the last case where the DAO still co-funds a validator’s shortfall.
- Worth restating what PSR actually is. Solana has no slashing. PSR is a bond-funded compensation system, not a penalty. The bond stays delegated to the validator, validators keep the full upside of the SOL they post; it is drawn on only when their own node fails to deliver what their stakers were counting on.
- The reasons behind MIP-2 don’t stop applying at 50%. MIP-2 was reacting to growing per-validator exposure, when the cap was 2% of TVL. It has since gone to 4% under MIP-10 and to 15% under MIP-19. The exposure MIP-2 was worried about is now several times larger, while the coverage line has not moved.
- Under today’s split, a long outage costs the validator no more than a medium one. Validator exposure is capped at half an epoch of rewards. Past that point each further percentage point of loss costs the operator nothing, because the DAO absorbs it. An operator deciding whether to fix a broken node late in an epoch faces no extra PSR cost for waiting.
- The DAO’s half only ever pays out on the worst incidents. A charge reaches the DAO reserve only after a validator has missed more than half an epoch, meaning a node left down for the better part of a day or abandoned outright. The reserve was set up as a backstop for the protocol.
- PSR coverage is something Marinade publishes to stakers. The coverage percentage is on the site, and it is one of the reasons stakers pick Marinade over staking directly. A guarantee that is partly funded by the DAO rather than by the operator responsible is weaker. It makes protection a cost the protocol carries rather than a standard validators are held to.
Impact
Validators with good uptime see no change; the charge only lands when credits fall short. Only incidents severe enough to cross the 50% line cost more, and only for the portion below it; a validator that misses an entire epoch pays close to double, which is the intended effect. Stakers are unaffected, since coverage from their side is 100% either way.
- Bond sizing already supports this. The auction’s bond floor reserves one full epoch of expected staker rewards on top of bid coverage, and that reserve is exactly the maximum a 100% downtime charge can claim. No change.
- DAO exposure removed is modest and highly concentrated. Since MIP-2 took effect, the DAO has funded 186.9 SOL across 131 downtime settlements from 72 validators, about 12% of all downtime compensation in that period. In the last 100 epochs it was 73.5 SOL across 11 settlements from just 6 validators. The aggregate is small; what matters is that it buys nothing, since every SOL of it went to an incident severe enough to clear the 50% line.
Call to Action
Leave feedback in this, particularly from validators on bond affordability and on whether 3 epochs is enough notice.