Where every number on Markov comes from, how it’s computed, how fresh it is, and which ones are models. If a panel can’t say where a number came from, it doesn’t show the number.
Three kinds of number, each with its own look, so you can tell them apart without reading the fine print.
It is read directly from a venue’s public feed, such as a price, a resting order, a fill or a liquidation. It shows as a plain number, drawn with solid lines.
64,180.5Example
It is fixed arithmetic on measured data, with nothing assumed about the market: session VWAP, the regime score, an annualized funding rate. We state the formula and any approximation.
Σ(price × volume) ÷ Σ volume
It is an estimate of something no venue reports, made under stated assumptions, such as dealer gamma levels. It is prefixed ≈ with a dotted underline and drawn dashed, and its basis shows on hover or keyboard focus.
approximately 63,500 (Modeled, example only, not a real level)Example
Every data panel carries one: source, state and freshness. The state is a shape first and a colour second, so it reads in greyscale.
What each surface shows, where it comes from, how it’s computed and which machine fetches it. Most of the free tools fetch straight from the venue in your browser, so those numbers never pass through us.
Chart
Candle snapshots when the chart opens, then Hyperliquid’s live candle stream at 1, 5 or 15 minutes. A live candle that arrives after a gap (a reconnect, a laptop waking from sleep) triggers a backfill of the missed bars before it is drawn.
Chart · Liquidity layer and depth profile
Markov subscribes to the book twice, at 4 significant figures for detail near price and at 3 for the wider range, about 20 price levels a side each. Each 1-minute column of the heatmap is the time-averaged resting size per price row over that minute. The depth profile at the right edge is the latest book.
It shows size that is resting, not intent. Orders can be pulled before price reaches them, and one venue’s book is not the whole market.
Chart · Whales layer and tape
Fills are folded into one print per aggressive order (same time, side and taker wallet), so an order that sweeps many makers counts once, at its volume-weighted price. Prints from $10k are kept. Bubbles start at each coin’s threshold ($100k for BTC and ETH by default, and you can change it).
Side is the taker’s: an aggressive buy or an aggressive sell. Wallets are public Hyperliquid addresses, not identities.
Chart · Liquidations layer, and the Liquidations page
On the chart, Markov reads Binance’s forceOrder stream, Bybit’s allLiquidation stream and OKX’s liquidation-orders channel. These are forced liquidations that actually printed, with longs marked below the bar and shorts above. CEX 1 and CEX 2 report at most one liquidation per second, so totals are a floor. Stored history from Markov’s collector carries the same three venues. Bybit reports each liquidation at the position’s bankruptcy price, a few tenths of a percent past where the market traded, so live Bybit prints are placed at the nearest price Bybit traded in that minute. The collector’s history keeps the price Bybit reported.
These are liquidations that happened. Markov doesn’t draw projected “liquidation levels”, which are a guess about where leverage sits. The Liquidations page reads the same three venues (Binance for every perp, Bybit and OKX for its six coins) and keeps what it has seen on your device, so repeat visits within a day fill its map in. From the US it fills from OKX, whose share of a total is a floor like Binance’s.
Chart · OI pane, and the Derivatives page
On the chart, Markov reads Hyperliquid’s asset context, open interest × mark price in USD, sampled every 30 seconds, with the collector’s stored samples (one per bar) before that when history is available. On the Derivatives page it reads Binance’s openInterest and openInterestHist endpoints. Where Binance refuses your region, the panel falls back to OKX’s open interest, read through Markov’s edge, and labels it. The two are never mixed.
Chart · Funding pane, and the Derivatives page
On the chart, Markov reads Hyperliquid’s hourly funding settlements. On the Derivatives page it reads Binance’s current rates, ranked by distance from its default 0.01% per 8 hours so the unusual ones come first. Where Binance refuses your region, the table falls back to Hyperliquid and labels it. The two are never mixed.
The annualized figure is a straight-line estimate that assumes the current rate holds. It is arithmetic, not a forecast.
Chart · Levels layer
Prior UTC day high, low and close (only when that day is nearly complete in the data), the 00:00 UTC open, and session VWAP: Σ(typical price × volume) ÷ Σ volume over today’s 15-minute candles. That VWAP is a candle approximation of the trade-by-trade figure, and is labelled as one.
They are measured from real candles, so they draw as solid lines. Modeled levels will draw dashed.
Heatmap page
Crypto shows Hyperliquid’s 60 perpetuals with the most open interest, sized by it, and their 24-hour change, through Markov’s edge cache. Forex shows ECB reference rates from Frankfurter, through Markov’s edge cache. US stocks, indices, commodities and futures come from Finnhub, through the same cache, and a board whose source isn’t answering says so.
Tile size is open interest on crypto and a fixed market-cap weight on stocks. Forex, indices, commodities and futures are equal tiles, and colour is the recent move. Futures, indices and commodities are ETF proxies (SPY for ES, GLD for gold, USO for crude…), not CME futures prices or index levels, and the futures tab names them that way.
Regime Compass, on TradingView
It blends three reads of price, each rescaled to −100…+100, in a 40/35/25 mix: volatility percentile (ATR as a share of price, ranked against its own history), Kaufman’s efficiency ratio (net travel over total path) and persistence (a variance ratio of n-bar against 1-bar returns). Above the threshold it reads Expansion, and below the negative threshold it reads Compression.
It is derived from price only. It is a proxy for expansion or compression, never dealer gamma. It evaluates on confirmed bars, so it doesn’t repaint.
Chart · Options layer (Elite, not offered in the US)
It is gamma exposure per strike from Derive’s options: each option’s gamma × open interest × spot² × 1%, summed by strike with calls and puts kept apart, over strikes within ±30% of spot and expiries up to 35 days out. Who is long or short isn’t published, so dealer positioning is signed by a heuristic (calls +, puts −) that can be wrong. It is one venue, not the whole market. The call wall, put wall and gamma flip draw dashed, labelled modeled, with the venue and the time named.
Positioning is regime and risk context. The backtest evidence we’ve seen suggests its independent edge largely disappears once the volatility regime is known, so we never frame it as a signal.
These are modeled, and nothing is shown until the data behind them is approved or licensed.
Nowhere yet
Options positioning mapped onto ES and NQ with a live index-to-futures basis. Not available until the options data is licensed. Whether the first version uses options on the futures themselves or index options mapped onto them isn’t decided.
Rolling out
A browser tab only sees what happens while it’s open. The collector is Markov’s own recorder: it keeps listening when nobody has the chart open, so the chart can open on recent history instead of an empty canvas.
Stored history draws to the left of a dashed boundary and this tab’s live data to the right. Book history is the time-averaged resting size per price level for each bar. A response carries at most the 500 largest whale prints and the 1,000 largest liquidations in its window. When there were more, the chart says from when, and above what size, its lists are complete.
Tell you what to trade.
There are no buy or sell calls, no “take this trade” and no signals. It is context, never instructions.
Publish win rates or profit figures.
There are no win rates or profit figures, on the site or in the scripts. The NY open brief shows how often its range held on past sessions it wasn’t fitted to. That is a hypothetical backtest, not a trading record.
Pass a model off as a measurement.
Modeled values carry ≈, dashed lines and their assumptions. Measured values are plain.
Fill gaps with invented data.
Time nobody recorded is drawn “not recorded” and left out of every count. Nothing is interpolated.
Show a number without its source.
Every panel names its source, its state and how fresh it is. Without a chip, there is no number.
Place orders or hold funds.
Markov reads public market data. It doesn’t execute trades, take custody or copy trades between people.
Names are trademarks of their owners. Markov isn’t affiliated with or endorsed by them.
Coming to ES/NQ · in development
Price is pushed and pinned by how options dealers hedge. The biggest strikes become walls. The line where dealer gamma flips sign sets the regime of the session.
That structure is computable from the options chain, but it lives in markets you aren’t watching. Markov is being built to draw it where it matters: on the ES and NQ chart you trade.
Open interest is end-of-day. Anything intraday is modeled, and we show the assumptions rather than hide them. Positioning is regime and risk context, not a signal.
Above the flip
Dealers sell strength and buy weakness. Volatility tends to compress and price to pin toward the big strikes.
Below the flip
Dealers hedge with the move. Trends can extend and breaks can run, and risk of ruin climbs.
This is the measurable side of the same idea: where size rests, where leverage breaks and who is paying to hold the crowded side. Open chart
Gamma walls and the flip, modeled from options open interest and drawn on your futures. It waits on options-data licensing. How we’ll model it
Each layer names its source and freshness right where you read it. A free account opens it, with no card.