The keeper stopped. Usually the reserve can't cover the next run's gas hold (gasLimit × gasPrice), so the scheduled run failed INSUFFICIENT_PAYER_BALANCE and nothing booked the next one. Check canPayNextRun(), fund() the market, then call revive() once the tick is overdue (nextRunAt + 2 × interval). /keeper has both buttons; see The keeper.
The market is close-only. The price is Stale or Deviated: opens, manual liquidations and refuels refuse it, and the keeper skips the sweep. Wait for the feeds; PriceGuard.read() says which feed is old or how far apart they are.
A deposit reverts with LiquidityPool__NoPrice. Deposits and withdrawals pause while positions are open and the price isn't Ok, so shares are never priced at a bad price. Wait for the feeds.
LiquidityPool__HtsFailed with rc 184 (TOKEN_NOT_ASSOCIATED_TO_ACCOUNT). The receiver isn't associated with cUSDC-LP: call associate() on the share token (the Pool page does it for you).
Testnet refuels fail closed. SaucerSwap testnet's USDC pool is ~22x off the oracle, so the keeper is funded by hand on testnet. Self-refuel is proven by tests at sane prices.
The keeper costs real HBAR. About 1.4 ℏ per run, ~65 ℏ a day at a 30 min interval. When the reserve can't cover the next run's gas hold, the chain stops until someone funds and revives it.
Push oracles lag. Chainlink and Supra update on deviation or heartbeat, so the market trades on a price that can be minutes old, and a 50x position can be hit by oracle lag. The 300 s minimum hold and the 0.1% fees blunt it; they don't remove it.
Liquidation keeps the remainder. A liquidated position's leftover equity (up to 1% of notional) goes to the pool, not back to the trader.
The pool's value ignores the per-position gain cap and loss floor. Exact until a position is past +100% or below zero equity; liquidation keeps positions away from the floor.
LPs are locked while prices are bad. Deposits and withdrawals pause while positions are open and the price isn't Ok, so shares are never priced at a bad price.
One position per account, one market per deployment. More markets are more deployments from markets.json; the app shows the default market.
Funding is linear and simple (skew / pool assets); there's no price impact and no borrow fee.
The keeper's HBAR can't be withdrawn. It can only pay for network runs. Redeploying leaves the old reserve behind.
The owner can tune, not take. PerpMarket's owner can only correct the gas-price estimate; PriceGuard's owner can only change staleness/deviation within hard bounds; the pool's owner links the market and creates the share token once each. No one can move user funds.
The live testnet deploy predates the stale-close fix. A review on 2026-10-03 found that, with one feed down, closePrice() used the other feed without checking its age, so a trader could close at an old price that favoured them. The source now settles a close only at a feed younger than maxStaleness (and waits if neither is), and treats a feed timestamp more than 5 minutes in the future as stale (PriceGuard.MAX_FUTURE_SKEW); regression tests replay the attack. The contracts deployed on 2026-10-02 still have the old behaviour. On testnet the exposure needs Chainlink to stop answering while Supra is stale at a favourable price, and is capped by each position's gain cap; the live deploy was kept rather than strand its funded keeper reserve and its network-run proofs. A new deploy from this source has the fix.
Testnet assets. USDC 0.0.5449 is SaucerSwap's testnet token.
Local chain is for trading and the pool only.yarn foundry:chain + yarn foundry:deploy runs everything on mocks with stand-ins for the two system contracts; the keeper books ticks but nothing runs them, and the mock feeds go Stale 2 h after deploy (move them with MockAggregator.set / MockSupra.set from /debug).