How every number on this site is made.

Prices come from Swap events of two Uniswap v3-style pools on Robinhood Chain (factory 0x1f7d7550b1b028f7571e69a784071f0205fd2efa, checked on chain 13 Nov 2026):

• WETH / USDG: pool 0x52e65b17fb6e5ba00ed806f37afcd2daa50271ca, fee 0.01%. The busiest WETH/USDG pool on the chain (fee 0.01 %).
• NVDA / USDG: pool 0xd4eb21209c4d6093f80b5b84f5c45cc093ea14a3, fee 0.05%. The NVIDIA stock token against USDG (fee 0.05 %). It trades on chain around the clock; the stock market does not, so quiet hours are common.

Sample k sits at block k × 34,560 (about one hour: the chain made 9.6–9.8 blocks a second in the 70 days to 13 Nov 2026). Its price is the pool price after the last Swap in the 30 blocks up to that block; with no swap there, the window widens to 300 and then 3,000 blocks. With still no swap, the sample is a gap and is left out, never filled in. Each sample links to its swap transaction in Replay.

Limits: one price per hour is not a candle. Moves inside the hour are invisible, so stops and targets are only checked at samples. NVDA trades on chain around the clock while the stock market does not, so its quiet hours show as gaps.

brediq-data-v1|chain=4663|grid=34560|windows=30,300,3000|price=last-swap-sqrtPriceX96

• A signal is read at a sample; the order fills at the next sample. Never at the price that produced the signal.
• Pool fee: the pool's own fee tier on every buy and sell.
• Slippage: a fixed extra (default 5 bps) plus, when on, trade size divided by the pool's in-range depth at that sample (read from the Swap event's liquidity and price). That estimate is low for trades large enough to cross price ranges.
• Gas: 180,000 gas per swap at the live gas price, converted with the Chainlink ETH/USD feed on Robinhood Chain. Paper runs use a fixed 0.01 USDG per swap so their result can be recomputed later.
• Long only, one position at a time, no leverage. A position still open at the end is sold at the last sample, with costs.
• Every table shows the same trade as "hold": buy at the first sample, sell at the last, same size and costs. Beating zero is not enough; beat that.

brediq-engine-v1|fill=next-sample|fee=pool-tier|slip=bps+size/depth|gas=flat|long-only|one-position

The window is split in two: period A (the first 67%) and period B (the rest). Each rule's parameter grid is run on A only, the best net P&L there is picked, and that pick is reported on B. Indicators on B may read A's prices (they are in the past); they never read B's future.

Next to every result: how many parameter sets were tried, where the pick ranks on B and the median of all sets on B. A pick that tops A and sinks on B was fitted to A.

• Trend cross: 9 sets (Buy when SMA3 crosses above SMA24 …)
• RSI pullback: 8 sets (Buy when RSI7 is under 20 …)
• Range breakout: 4 sets (Buy when price breaks the 12-bar high …)
• Dip from high: 6 sets (Buy 2% under the 24-bar high …)
• Momentum: 6 sets (Buy after a 1% rise over 6 bars …)
• Coin flip (control): 5 sets (Buy on a 4% coin flip (seed 1) …)

The coin flip is the control: seeded random entries with the same exits and costs. A rule that cannot beat it on B has shown nothing yet.

BrediqProofBook stores, before a period starts: the strategy version's spec hash (the spec itself travels in the event log as JSON), the data source hash, the engine hash and the exact start and end. After the end the author posts the hash of the full result file. Anyone can rebuild the spec, read the same samples and compare hashes.

• Data source hashes: WETH / USDG 0x53318a76…, NVDA / USDG 0xbcbdf16b…
• Engine hash: 0x746a4623…
• Withdrawals are possible only before the start and stay counted. Results are posted once and never edited. A result posted more than 7 days after the end is marked late.
• Contract: 0x45e0e2d6b59739f39b69e3e36bb8f05af8bcdfc8

The contract holds no funds and has no owner. It cannot tell whether a result is true: it keeps the order of events and records who checked what. Paper Arena

• Backtest a past period, recomputed in your browser from chain samples.
• Paper a forward test committed on chain before its period, result posted after.
• Live real money. None exist: funded bots are a separate, planned stage.

They sit in separate tabs and are never ranked against each other.

• A good backtest or paper run does not predict the next period.
• Hourly samples miss what happens between them; real fills can be worse, especially in thin pools.
• Two to three weeks of data is short. Treat every result as a hypothesis to test again, not a finding.
• Nothing here is financial advice and nothing here trades.