How an LST is supposed to behave
An LST is a receipt: you deposit SOL into a stake pool, you get back X units of the LST. Behind the scenes, the pool stakes your SOL to validators and accrues staking rewards (plus, for jitoSOL, MEV rewards). As rewards accumulate, the SOL backing each LST share grows — so the redemption price of 1 LST in terms of SOL slowly rises over time. Sanctum publishes this value assol_value. For jitoSOL
right now, 1 jitoSOL ≈ 1.18 SOL — meaning each jitoSOL you redeem returns 1.18 SOL of
underlying stake. That ratio is what Pegana calls the intrinsic value: the
on-chain truth of what the asset is worth in its denominator (SOL).
How the market actually prices it
Most users don’t redeem through the stake pool. They swap on Jupiter, Orca, Raydium, or one of the LST-specialized AMMs. The price they get is the market value: 1 jitoSOL might trade for 1.179 SOL, or 1.176 SOL, or — during stress — 1.16 SOL. The gap between intrinsic and market is the discount. A small negative discount (market trades 0.05–0.30% below intrinsic) is normal — arbitrageurs need a small spread to redeem-and-sell profitably. A large negative discount is news.What causes LST peg drift
- Network congestion. If Solana slows down, arbitrageurs can’t close the gap quickly. Drift widens until the network catches up.
- Stake pool program upgrades. Any LST whose pool is mid-upgrade may pause redemptions; secondary market discounts to compensate.
- Validator slashing (rare today, but landing in 2026 per Anza’s roadmap). A slashed validator means the pool loses some underlying SOL. The LST’s intrinsic falls; if the market doesn’t immediately reprice, the gap inverts.
- MEV regime change. jitoSOL’s intrinsic includes captured MEV rewards. If MEV economics shift, the implied yield drops and the market can repress the price.
- Large redemption queue. Marinade’s mSOL has a multi-day unstake queue. Holders who need SOL now sell into the secondary market at a discount.
- Whale exit into thin venues. The classic mSOL flash-crash in June 2022 was one whale selling 88k mSOL into a thin Mango pool, dragging a Pyth feed and triggering $21M of cascade liquidations on otherwise-healthy positions.
What it isn’t
LST drift is not stablecoin depeg. There’s no central issuer to credibly defend a $1 number; there’s no Treasury reserve. The peg is a market-clearing mechanism, not a contractual promise. “Depeg” in the LST context means the market refuses to honor the intrinsic SOL value — usually temporarily.How Pegana detects it
Pegana polls Sanctum every 15 seconds for each LST’s intrinsic SOL value and converts it to USD via Pyth’s SOL/USD feed. In parallel, it polls Jupiter for the best swap rate (also normalized to USD). Discount =1 − market / intrinsic, smoothed with a
30% EWMA, and bucketed into per-asset thresholds (drift / depeg / critical) calibrated
to each LST’s 24h p99.
Calibrated thresholds matter. INF runs hot — its p99 drift is ~1.12%, four times
jitoSOL’s ~0.37%. A 1.00% alert on INF is noise; on jitoSOL it’s a real signal. The
methodology page documents every threshold.
Why this matters for builders
If you’re running a lending protocol that accepts LST collateral, your liquidation engine reads a price oracle. If the oracle gets one bad print from a thin venue (the mSOL/Mango case), you can liquidate positions that have nothing wrong with them. Cross-checking against Pegana’s intrinsic feed lets you reject divergent oracle prints before they hit your liquidation queue. If you’re underwriting LST risk — insurance, prediction market, structured product — you need a credible third-party event source for what counts as a “depeg.” Pegana’s class-aware state machine emitsAlertEvent with a clear from→to transition,
plus the intrinsic and market values at the time. The methodology is open; you can
reproduce every alert from raw inputs.
Read next
Live LST states
See the current state of every tracked LST.
How Pegana computes
Sources, formula, thresholds, FSM.
mSOL flash-crash post-mortem
The canonical “market lied” case study.
REST API
Read live state programmatically.