Log · build in public
What we thought, learned, built and stopped, making new markets.
Every entry has one measured number or one decision. Stopped ideas stay, with the number that stopped them. Anything written with AI says so.
The question now · since 2026-11-15 · 6 entries
How do you take the price at a set time, without letting anyone choose it?
Next: 2026-11-16 14:33 UTC: record an xyz:NVDA round in US market hours (GitHub Actions runs it)
11-15BuiltDeleted each of the 38 checks in an AI-written five-minute market, one at a time. No test noticed 18 of themfivemin · unproven · Recipe 418 / 3811-15StoppedA Sunday xyz:NVDA round, recorded and replayed, opened and closed at the same oracle price: void. With US stocks closed, the oracle changed only twice in six minutesfivemin · one round recorded live and replayed161.110 = 161.11011-15LearnedHyperEVM's price precompiles return today's value whatever block you name: there is no history to readDojo explainer: settle at a set timeno history11-15ThoughtLet nobody choose when the price is read: take the first reading recorded after the time, and fix it therefivemin spec3 s window11-15LearnedReview: whoever holds signed Pyth updates can pick the price inside the window. Monad waits; HIP-3 goes firstfivemin spec · open blockerHIP-3 first11-15LearnedA price you can read with no key and no contract: Pyth's BTC feed on Monad updates every 17 s at the medianfivemin · measured on Monad mainnet17 s11-15StoppedBringing memecoin speculation to Monad: nad.fun launched 5 tokens to 60 traders in a dayResearch: one day of nad.fun's curve$360/day11-15LearnedAnthropic's pre-IPO perp: OKX moved the share count from 10B to 13.1B, which looked like a 24% crashHL Price Guard · the listing record−23.76%11-15BuiltThe same Uniswap v4 hook on Monad: one swap's gas, measured on both schedulesDojo: Uniswap v4, lesson 9+17%11-13StoppedA stock dealer on Robinhood Chain: an hour after each fill, it had lost to adverse selectionResearch: 17 single-stock/SPY pools, 7 days of swaps−10 to −18 bp
Each entry links to the public commit, page or recipe it came from. If a measurement has no public source, the entry says where it was measured.