Five-minute up-or-down pools. Stakes go on Up or Down before a round starts; when it ends, the side that was right shares the other side's stakes, less a fee. Nobody takes the other side of a bet. The two prices are the first readings at or after the round's start and end, from a feed already on the chain, and never change once recorded. Spec: docs/spec.md.
| source | chain | assets |
|---|---|---|
PythPushSource |
Monad mainnet | Pyth's sponsored push feeds: BTC, ETH, SOL, MON (0.02% deviation) |
Hip3Source |
HyperEVM | a Hyperliquid perp's oracle, precompile 0x807 (HIP-3: set by the deployer) |
Status: draft, not audited, not deployed. Tested on a Monad mainnet fork with a real round: BTC, 2026-11-15 04:05 UTC, recorded at Pyth's first updates after the start ($64,725.57) and the end ($64,634.06), Down won.
forge test # unit tests, the HIP-3 source with a mocked precompile, and the Monad fork round
unproven unproven.toml # delete each check in turn: which test notices (about 10 minutes)Uses hl-price-guard v0.4.1 (PythRef, HLPriceGuard).
Walk through it: the recipe Open a five-minute up-or-down market (about 10 minutes), and the Dojo topic Settle at a set time from a price already on chain.
Claude (Opus 5.5) wrote all of it, in one session on 2026-11-15: the spec, the two contracts (319 lines) and the tests (662 lines); every commit names it as co-author. A person chose the design (stakes close at the start, first reading after each checkpoint) and the checks. What was used to check it, and what it found:
| check | result |
|---|---|
| A real round on a Monad mainnet fork | the recorded prices are Pyth's first updates after the start and end, read independently from PriceFeedUpdate logs |
| Unit tests, and a fuzz test of the payout (1,000 runs) | pass; the fuzz test is what states "a winner never gets less than their stake" |
| Codex review r1 | CHANGES: a late poke could rewrite a void round, a reading ahead of the block was accepted, HIP-3 allowed a 299 s window, fee-on-transfer tokens; all fixed with a test each. One blocker is open (below) |
unproven: delete each of the 38 checks in FiveMin.sol in turn |
first run: 20 caught by a test, 18 caught by nothing, as predicted row for row. Three of those would have lost money if the check were gone (a fee taken from a round still pending, a fee taken from a void round, a Down-only round won by Up paying no one). 11 tests added; now 31 caught, 7 unproven, each with the reason it is not a hole in unproven.toml. 2026-11-16: two more checks listed (a recorded open or close never changes); deleting the close's had failed no test, so a test was added: 40 checks, 33 caught, 7 unproven |
Open blocker (review r1): with Pyth, anyone holding signed Pyth updates can push the most favourable price inside a checkpoint's window before the sponsored pusher does, then poke. See docs/spec.md. Not for real money until it is closed.
What has not been done: an audit and a mainnet deployment. The Monad fork test is not in the unproven table (it needs the network).
Apache-2.0