Methodology · data

How the data is collected

A derivatives data platform is only worth as much as its collection network, so this page documents ours in full: where each series comes from, how often it updates, how far back we actually hold it, and where it stops being reliable. The row counts below are queried from the live database every time this page renders. If a series is thin, you will see that it is thin.

What we hold, right now

Read this as inventory, not as a claim. A recent start date means we started recording recently, and depth accumulates from here because we keep what we record.

SeriesWhat it isEarliest recordRows
Coinbase premiumDaily US versus global spot gap, plus carry basis2017-08-176,560
CFTC positioningWeekly Commitments of Traders for Bitcoin futures2018-04-101,232
DVOLDeribit 30 day implied volatility index, daily2021-03-243,932
ETF flowsUS spot BTC and ETH ETF net flows by issuer, daily2025-05-20612
Signal recordEvery signal the engine has ever emitted2026-02-20337,286
Intraday open interest5 minute open interest per symbol2026-07-0894,581
Perp listingsNew and delisted perpetuals across six exchanges2026-07-233,356
Daily market snapshotDominance, market cap, sentiment, end of day2026-07-2319
Real liquidation printsEvents broadcast by Binance, Bybit and OKX2026-07-30435,693
Liquidation rollupPermanent daily totals by symbol and side2026-07-309,232
Order book depthBook snapshots behind the slippage ladder2026-07-311,266,598
Options tapeDeribit trades, recorded every 15 minutes2026-08-0526,513
Options open interestPer-strike open interest, daily2026-08-067,670
Slippage ladderHourly execution cost by order size2026-08-06608

Counts are live from the database at page render. Series we started in late July and early August 2026 are shallow on purpose: they are new, and the archive grows one interval at a time.

Two different sets of six venues

This distinction matters more than any other on this page, and it is the one most easily blurred. The liquidation model aggregates six venues: Binance, Bybit, OKX, Gate, HTX and Hyperliquid. The funding arbitrage screener also covers six, but not the same six: Binance, OKX, Bybit, Gate.io, HTX and BingX. Neither of those is a description of the site, because the terminal additionally pulls Deribit for options, Coinbase for the US spot premium, the CFTC for positioning and SoSoValue for ETF flows.

So when you read a venue count anywhere on this site, it is attached to a specific product. If a sentence ever says the site covers six venues without saying which product, that sentence is wrong and we want to know about it.

Estimated liquidation levels versus real liquidation prints

The liquidation heatmap is a model. It estimates where leveraged positions would be forced out, from open interest, price history and leverage assumptions, across the six model venues. No exchange publishes its liquidation levels, so every liquidation map anywhere, including ours, is an estimate. Anyone presenting one as measured data is misleading you.

Real liquidation prints are a different thing entirely: they are events an exchange actually broadcast. We ingest those from three venues over websockets, Binance, Bybit and OKX, and overlay them on the model so you can see where the estimate and reality agreed. Binance's stream is throttled by the exchange to roughly one message per second per symbol, which means the count of events is a floor, not a census. We treat it as a floor everywhere it is used.

Cadence, and what stale means here

Funding, open interest and pressure refresh every 15 to 20 minutes. Liquidation events stream continuously. The liquidation map recomputes every 2 minutes. Indices refresh every 30 minutes. ETF flows and CFTC positioning arrive on their publishers' schedules, daily and weekly respectively, so they are stale by construction between releases and we print the date they carry rather than the time you loaded the page.

Every page prints the timestamp of the data it is showing. If the timestamp is old, the data is old, and we would rather show you an honest old number than a fresh-looking wrong one.

What we do not collect

We do not have order-by-order exchange data, exchange-side liquidation levels, wallet-level onchain attribution, or any private or licensed feed. We do not buy data. Everything above comes from public exchange APIs and websockets, public regulator filings, and our own computation on top of them.

We also do not backfill what we did not record. Where a series starts in July or August 2026, that is when we started collecting it, and no earlier history exists on our side. Where a series reaches back years, it is because the source publishes history and we imported it.

Corrections

If a number here contradicts what a page shows, the page is authoritative and this document is wrong, and we will fix it. Send the URL and the number to support@bykaranteli.com.