By source
By update frequency
Most research tools do not call exchanges when you ask a question. They read
history that was collected on a schedule and stored, so asking the same question
twice gives the same answer. A backtest is only worth reading if its inputs held
still while you read it.
Every venue’s data goes through one path. A scheduled job collects it and
stores it. A live feed service then turns it into stored history and into
ready-made views, and cards and research tools read those. A card can watch
Hyperliquid and trade on Binance, because the data and the venue it trades on are
chosen separately. A venue’s retention limit loses nothing, because we keep our
own copy.
The schedules write a database, not files. Each sync reads what its table
holds, fetches forward and writes back; the service reads the same tables. There
is no file in between, on purpose. When there was one, a deploy overwrote a
freshly synced pool history with a copy from somebody’s laptop. Nothing showed the
loss: every page still rendered, and hedges were priced from a day 10 days stale.
Three tools are the exception and each says so: the live
option book, the latest sample of what a perpetual is paying, and forced closes as
they happen. A price somebody is about to trade on cannot come from a daily snapshot.
For how live data reaches a card, see How live data reaches your card.
These are the streams.
Pool history
Daily price, total value locked, fees earned and in-range liquidity for 41 Uniswap v3 pools: 21 on Ethereum, 14 on Arbitrum and 6 on Base. The history reaches back to 2021-05-05. The same 41 pools also have an hourly candle history, from the same start. Fees are the hard part. An indexer values a swap in dollars only if it can price both tokens, and each deployment keeps its own whitelist, so a pool can record thousands of trades in a day and report zero fees. Where the other side of the trade is a dollar, the fee is rebuilt exactly, and any row that was rebuilt is flagged. The tools pass that flag through. Read by surveying ranges, LP strategies and backtesting a position.Options and volatility
Deribit’s volatility index (DVOL) back to 2021-03-24, plus option chains used to price structures at strikes away from the money. Read by option structures and pricing a hedge.Perpetual funding
Daily funding rates from four venues on each of BTC and ETH.
They are kept separate and never averaged. Binance and Bybit correlate
around 0.85 on BTC. Where two venues disagree, that disagreement is what you want
to see, and a mean would hide exactly those days.
The four are also not interchangeable. Over the 1,187 days all four cover, the
two decentralised venues paid about twice what the two centralised ones paid.
Two independently operated decentralised venues agree within 0.7 points. With
two agreeing, the higher rate looks like a property of the venue type rather than
of one venue.
These rates are quoted on notional, not on capital. A delta neutral position
holds the coin and shorts the perpetual, and the short needs margin. At two
times leverage the capital is one and a half times the notional, so a venue
paying 14.5% returns under 10% before costs. Read a funding figure as a rate on
the position, never as a return.
Open interest
From Binance’s daily archive, at five-minute resolution since 2021-01-01: open interest in coins and in dollars, top-trader long/short by position size and by account count, and taker buy/sell volume.Coverage is per symbol, and the archive is walked one day-file at a time, so
a symbol can be partly filled while another is complete. Ask the tool for the
current depth rather than reading a figure here: it reports how many days it
holds, and that number moves.An absent or partial series is reported as such, never as a zero, because
an open interest of zero says nobody holds a position at all, which is never true
of a live market.
Prices
Candles for four assets from Binance’s public archive, reaching back to 2021-01-01: BTC, ETH, PAXG and USDC, all quoted in USDT.
One-minute data is a rolling window. A backtest that needs one-minute bars covers
at most the last three days, and a longer backtest runs on hourly bars or coarser.
Weekly bars are built from daily bars, not downloaded. Binance publishes
weekly files about two months behind its daily ones. Rolling the daily series up
keeps the coarsest view as current as the finest. The roll up is checked against
Binance’s own published weeks, with no differences.
Wrapped assets are absent on purpose. WBTC and XAUt appear in five pools
between them, and both trade thinly on centralised exchanges. WBTC’s archive
starts 2024-01 and XAUt is missing 63 months. A price series built from those
would describe the wrapper’s own illiquidity. BTC is the reference for WBTC, and
the wrapper’s price against the asset is an on chain question that the pool
history already answers.
Prices as they happen
The swap router records its own price for each token pair every time the rate changes, on Ethereum, Base and Arbitrum. Those records become closed bars at one minute, one hour, four hours and one day, priced in USDC. A token with no USDC market gets a derived price, crossed through a bridge token. Quiet markets close their bars on a once-a-minute heartbeat, so a market with no change still gets a bar. Cards read these prices. They are stored as they happen, so a card that was restarted can rebuild its bars from the record.Pool volume by the hour
For the router’s curated USDC pools on Base, the feed service reads each swap and stores the hourly volume of both tokens and the swap count. It also builds ready-made views of that volume. A card that triggers on pool volume reads these views.Lending rates and positions
Four kinds of lending data are read from the network and stored.- Reserve limits and prices on Base. Each lending reserve’s loan-to-value limit and liquidation threshold, and the lender’s own USD price of the asset.
- The same on Ethereum for the newer lending market, per reserve.
- Your account’s lending position on Base. Supplied, borrowed and health factor, read for your operating address only. Nobody else can read it.
- Lighter deposits and withdrawals on Ethereum. USDC deposits and pending withdrawals into Lighter, per hour, for the accounts that need it.
Lighter
Lighter’s market data is stored from late October 2026: mark price, index price and funding for each market, and one-minute candles for the 20 largest perpetuals (the biggest crypto perps and gold). Candles are final about a minute after they close. One-minute candles are kept for three days.Hyperliquid and Binance candles
Hyperliquid’s one-minute candles for BTC and ETH are stored the same way, kept for three days, with an hourly history from 2026-03-31. Binance’s candles are described under Prices. See Hyperliquid for what the venue keeps and what it does not.Liquidations
Two streams, kept apart because they are different things. Centralised venues, hourly, from 2022-06-02. Binance BTC and ETH perpetuals: 3.9 million forced closes across 66,426 hourly buckets, each with a count, a price range and a quantity. No exchange serves this itself. Binance removed its endpoint, Bybit has none, and AsterDEX answers that its endpoint is out of maintenance. This stream comes from a paid provider, and it is the only route to the four years. Hyperliquid, per event, from 2025-10-10. Every forced close with the address it happened to. See Hyperliquid.The two are never merged. The centralised stream is an hourly aggregate and
cannot name a trader. The Hyperliquid stream is one row per event and can.
Merging them would drop the detail on one side and invent it on the other.
Hyperliquid
A third venue, and it differs from the other two in ways that do not average out. Funding settles hourly: 24 intervals in a complete day, not 3, and its history begins 2023-05-12, the venue’s own beginning. Its liquidations are richer than anything a centralised venue publishes: the address that was liquidated, the mark price at the time, whether it was taken at market or backstopped, and which vault absorbed it. Its liquidations are backfilled in full. The fills endpoint caps at 2,000 rows and the vaults trade constantly, so reading them took 278,311 fills across the six vaults that perform liquidations. The result is 3,302 forced closes reaching back to 2025-10-10, which is when those vaults began to liquidate at all. Its open interest cannot be backfilled. The venue reports open interest only as it is now, so that part of the stream records forward only and begins when the watcher began. See Hyperliquid.Macro, on-chain and stablecoins
Twenty-five Federal Reserve series from 2010-01-01, bitcoin valuation from 2010-07-18, ether from 2015-08-08, and stablecoin supply per issuer from 2017-11-29. Prediction-market odds are snapshotted daily beside them. These are the inputs to the slow signal and to what the market expects. See macro, on-chain and stablecoins for what each holds and why every series carries its own publication lag.Where every stream stops
A tool that cannot cover the window you asked for says so. None of them fills a
gap with a zero.