Applied Market Mechanics: Tool 02 · Derivatives Panel

Hang two pendulums from the same beam and set one of them swinging. Within a minute the other has started too, because every swing tugs the beam a fraction and the beam passes the tug along. Leave them long enough and the motion trades back and forth between them, one building while the other dies away, then the reverse. If you want to know which of the two is doing the driving at any moment, watch their timing: both are swinging, so size settles nothing, and the pendulum ahead in phase is the one pushing. In a first-year lab you'd find that out with two light gates and a timer, and it's a surprisingly satisfying afternoon.

A spot market and its perpetual future hang from the same beam. The beam is arbitrage: whenever the two prices drift apart, somebody is paid to buy the cheap one and sell the dear one, and the gap closes within minutes. So most of the time they swing together, and on a price chart nothing separates them at all. Every so often one of them leads, though, and which one says a great deal about what is carrying a move. Either people are exchanging the asset itself, or leveraged positions are being opened and closed against each other, and those are very different things to have underneath a price.

This Tool started as a chart I posted on 1 December 2023, in reply to a question about the CME premium: CME's ether future on top, the index underneath, a pane for the gap between them, and the volume of each. Every Long Has a Short took that chart apart and found its premium pane was mostly measuring ether's movement between two closes taken hours apart. What follows is the corrected version, extended by a row and built entirely from symbols a free TradingView account can load. The reading I drew from that chart in the reply was about market phase, and it needs a Tool of its own later in the series.

Like Tool 01, this entry has a second job: it introduces notation later entries lean on, for whether a move opened or closed positions and for which venue moved first. Unlike Tool 01, its thresholds were calibrated before they were written down, on three years of hourly data from three contracts of very different depth. What the calibration hands back is a band, and the choice inside it is yours, with what each end of it costs laid out below.

#The Question

Is this move being carried by spot or by leverage?

#Mechanics underneath

Every derivatives contract has two owners, one long and one short, and that rule from Every Long Has a Short decides what this panel can and can't see. So leverage leaves three kinds of trace, none of them a direction: open interest, the count of contracts that exist; the premium, the price the perpetual trades at relative to the spot it tracks; and aggression, which side crossed the spread to get its trade done. The panel puts those three next to each other, and next to the spot market they're all measured against.

Open interest answers whether a move added leverage or removed it. A fall in price on rising open interest means new pairs of positions were opened while price fell, and every one of them will have to be closed eventually, by choice or by the liquidation engine. That's fuel left in the market. A fall on falling open interest means positions were leaving, and the move was made of leverage being taken off, often forcibly. The same price candle can sit on top of either, and from the candle alone you can't tell which.

The premium answers who was paying to get positioned. Binance's premium index is the perpetual's price measured against its spot index, and it's the in the funding formula Part 5 sets out. When it rises, buyers on the perpetual are paying up relative to spot; when it falls, sellers there are leaning on the price. Funding is built from this same number, averaged over eight hours and clamped, so the panel reads the premium directly and leaves funding at the venue: the input carries everything funding does, an hour at a time, without the blind band in the middle.

Aggression answers who moved first, and it's the part that needs the most care. The Long Shadow of Markets argued that a move starts wherever the size happens to be sitting and the other markets follow because arbitrage drags them. On a chart that shows up as one venue's taker delta, the volume that crossed the spread, turning before the other's. Because the beam is so stiff, the gap between the two is rarely more than an hour or two, and a good share of the time both venues act inside the same quarter-hour. Hourly bars average much of the gap away and daily bars all of it, which is why this part of the Tool is read on 15-minute bars.

Everything on the panel is compared against its own normal. A change in open interest of a thousand ether is enormous on a small contract and a rounding error on ether's, and a funding print of means different things on venues with different clocks. So every question the panel asks is relative: how many times larger is this than usual, for this contract, at this timeframe, over the past thirty days? That's the same instinct Turn the Candles Off applied to volume: a single reading means little, and the comparison carries the information.

#Configuration

The panel lives in two TradingView tabs on the same timeframe. On a free account the volume indicator belongs to a tab's main symbol, even when moved into its own pane, and a tab holds one chart, so the perpetual and spot each get a tab of their own. The perpetual takes the first, since leverage is what the panel is about, and spot is the reference it's read against.

Other settings:

A paid plan's multi-chart layout puts both tabs side by side, which is convenient and changes nothing in the method.

Three things the panel deliberately leaves out. The first is the premium of a dated future such as CME's: TradingView's free plan only allows typed spreads on the daily, and a daily spread between CME and spot subtracts two closes taken hours apart, which is exactly the mistake in my 2023 chart. The second is the long/short ratio, which counts accounts on one venue and says nothing about money. The third is TradingView's volume delta and CVD indicators. They estimate each bar's buying and selling from the direction of smaller candles inside it, which isn't the same thing as aggression, so taker delta comes from the venue's own data instead (Procedure, step 5).

#Notation

Every quantity on the panel is read as a multiple of its own normal. For any series , its multiple at bar is the value over the median size of the bars before it:

with thirty days of bars ( on the hour, at 15 minutes). The median looks only backwards, so a reading never uses a bar from after the one being read, and a multiple of means three times the usual size, in either direction.

Opening and closing describe open interest across the bars of a move. With the multiple of the change in open interest, , a bar is opening when and closing when , for a threshold chosen from the band in Reading it. Paired with the direction of price, that gives four states: opening ↑, closing ↑, opening ↓, closing ↓. Neither word says which side opened or closed, because every contract has both.

Spot-led, perp-led and together describe which venue's aggression turned first. With , the taker delta from Every Long Has a Short, and its multiple on each venue, the leader is the venue whose first reaches in the direction of the move, on 15-minute bars. The lead is the time between the two venues' first crossings ("perp-led by 60 min"). Together means both crossed in the same 15-minute bar.

Anchored cumulative delta sums a venue's taker delta from a Tool 01 anchor bar , and comes in two forms, because each answers a different question:

where is the venue's average bar volume from to . is in average bars: means sellers crossed the spread by three average bars' worth since the anchor. is the share of all volume since the anchor that was one-sided, and runs from to . Both use only bars up to .

A reading is written leader, then state, with the thresholds given: "perp-led opening ↓ (, )".

#Procedure

1. Choose the venue. Every row from one venue, including spot. Taker delta from one venue says nothing about another's, and open interest summed across venues mixes contract specifications and counting rules.
2. Name the range. Take the Tool 01 range you're reading, write its anchors down, and note its anchor timeframe. Its climax bar is the anchor for cumulative delta.
3. Read the state on the anchor timeframe. Compute for the bar or bars of the move and write opening or closing with the direction of price.
4. Read down to see how it formed. Drop to the hour (Tool 01's rule: a range is read on its own timeframe or lower, never higher) and mark where open interest rose and fell inside the move, with the premium beside it.
5. Find the leader on 15-minute bars. Find the first hour of the move, an hour whose price change is several times its normal. Take each venue's 15-minute taker delta from the venue's own data (Binance publishes taker-buy volume in every candle; TradingView's delta indicators only estimate it), compute , and search the six hours up to the end of that hour for each venue's first bar at or beyond . Write spot-led, perp-led or together, with the lead in minutes.
6. Read the cumulative delta from the anchor, both forms, for each venue.
7. Write the reading with its thresholds, and the anchor it was measured from, before deciding what it means.

#Reading it

The four states, read with the rule that every long has a short:

Which side started a closing isn't in the data: a liquidated long and a short taking profit leave the same footprint. Opening moves leave fuel behind, because whatever was opened will be closed later. Closing moves burn it, and a move made mostly of closing tends to end when there's nothing left to close.

Choosing . I calibrated the threshold on Binance's BTCUSDT, ETHUSDT and KAVAUSDT perpetuals, hourly, from 2022 to 2024, a window kept well away from the worked example. KAVA is there as the thin contract, with a median open interest of million dollars against ether's million. Each hour's change in open interest was put in multiples of its own normal and grouped:

Past about , hours where open interest changed unusually come with larger price moves and above-normal volume on all three contracts, and the relationship keeps rising with no single break in it. So – is the band, and where you draw the line inside it is a choice about errors. A lower counts more moves and catches positioning earlier, at the cost of reading ordinary turnover as positioning. A higher one counts fewer moves, cleaner ones, later, and misses quieter positioning entirely. On ether at , about of hours count; at , about . The largest changes lean towards flushes: on ether, closing ↓ is of the hours counted at and of those counted at .

I tried a second rule too, open interest change as a share of the volume traded in the same hour, which ought to say how much of the trading added or removed positions. It fails hour by hour. A high share usually means a quiet hour, with volume around normal on BTC and ETH, and on BTC the median price move falls from to as the share rises. It's left here struck through, in the lab book's sense, so nobody has to find that out again.

Choosing , and reading the leader. The same calibration took every hour whose price change was at least its normal, the first of any cluster, and asked which venue's taker delta crossed first. On hourly bars, BTC and ETH showed both venues acting in the same hour in – of these moves, so the hour is too coarse to separate them. On 15-minute bars, for February to December 2024, at :

Three things follow. On deep pairs, together is common, about a third of moves, and it's a reading in its own right: the beam is doing its job. Neither venue leads by default. At , BTC and ETH split almost evenly; at the looser and stricter settings the perpetual leads somewhat more often (ETH at : perp-led, spot-led), and on the thin contract spot does. And when there is a leader, the lead is long: a median of to minutes between the two venues' first crossings. The same trade-off applies to as to : lower thresholds name a leader more often, and some of those leaders will be a single noisy bar.

The premium. Read it against its own normal, too. Binance's ETHUSDT perpetual sat below its index in of hours from August to early October 2025, at a median of , so a premium of there is ordinary and a premium of zero is the perpetual being bid. A premium that jumps while open interest rises says the opening was done by buyers paying up; one that sinks says sellers did it.

Cumulative delta. Read both forms together. When and both trend, the whole stretch since the anchor has leaned one way. When moves and barely does, the aggression is a burst against a history that was otherwise balanced. That second case is common, because dilutes as the window lengthens, which is exactly why both forms are kept.

#Worked Example

10 October 2025, ether's crash, read on Binance. The example is drawn after the event, from data, and the Tool 01 range it's measured from was drawn for this example by Tool 01's own order: the week of 18 August, the top, traded ether on Binance spot, less than the week before it, so the weekly showed no volume climax and the search dropped to the daily. There, 22 August stands out at ether, more than double the days either side.

• Climax: BINANCE:ETHUSDT · 1D · 2025-08-22 00:00 · close
• Reaction: BINANCE:ETHUSDT · 1D · 2025-09-25 00:00 · close

The 24 August high of was a wick. On these anchors is , and the 6 October close of went back through it, so the range had its edges before the crash.

The state on the anchor timeframe. The day closed at , below , and the perpetual's open interest fell from ether at the end of 9 October to at the end of the 10th, a fall of ether. Against the thirty days before it that's the normal daily change, far past any setting in the band. On the daily, 10 October reads closing ↓.

How it formed. Dropping to the hour shows the day was two moves, and they were opposite in kind.

The first leg ran from 14:00 to 19:00, and across it open interest rose by ether, from at 14:00 to at 18:00. Most of that came in the 15:00 hour, when price fell and open interest's change was normal: opening ↓ at any setting from to . New positions were being opened into the fall, and the perpetual's sellers did most of the crossing: in the 15:00 hour its taker delta was ether against spot's . The premium stayed close to its usual level, between and .

The second leg was the opposite. Open interest's change was normal in the 20:00 hour, in the 21:00 and in the 22:00, and by 22:00 it stood at ether: ether of positions gone in three hours, about of them in the last two, ten times what the first leg had added. The perpetual traded ether in the 21:00 hour against spot's , and printed at its low. The premium index ran from to inside that one hour and closed it at , then at 23:00. This is closing ↓, and it's the footprint Part 5's liquidation engine leaves, though the panel can't show how much of it was forced.

One detail shows why the hourly bars need care. The 21:00 hour closed above its open, after the low, so on its own close-to-close it reads closing ↑. The fall and the rebound both sit inside one bar, and the 15-minute bars split them: heavy selling at 20:45, 21:00 and 21:15, then the perpetual's buyers crossing the spread at and normal at 21:30 and 21:45, while spot's sellers were still crossing at in the 21:30 bar. A state read on a bar can't see inside the bar.

Who moved first. The first move of the day was the 14:00 hour, whose price change was its normal with nothing comparable in the six hours before it. Searching 09:00 to 14:45 on 15-minute bars:

At , the leader is a single spot bar at 12:30 that came to nothing, which is what a low threshold costs. At , the perpetual's sellers turned an hour before spot's. At , only the 14:45 bar qualifies on either venue, and both crossed in it together. I'd read the move at : perp-led opening ↓ (, ). That reading is a choice, and the table shows what the other choices would have said.

Since the anchor. The cumulative delta puts the day in context.

From the 22 August climax to 13:00 on 10 October, the perpetual's sellers had out-crossed its buyers by average bars, a slow, steady lean that ran through the whole range while price went sideways. Spot's had reached . By 23:00 on the 10th the perpetual stood at and spot at . In its own units, spot's selling on the day was the larger change: the perpetual added average bars of selling and spot added , nearly doubling its total. The share of volume tells the slow half of that story: the perpetual sat at before the crash and after, too diluted by seven weeks of trading to register the day. So the perpetual had been leaning for seven weeks, and on the day spot caught up.

Put together, on this panel and at these thresholds, 10 October reads: a range's floor that had been leaned on by the perpetual's sellers since the climax; a first leg that was perp-led opening ↓, leverage added into the fall; and a second leg that was closing ↓, positions leaving at ten times the rate they had been added, with spot's sellers making their largest relative contribution of the whole range. Funding, meanwhile, said almost nothing (Failure Modes).

#Failure Modes

Funding read as a signal. Binance's settled funding prints for ETHUSDT on 10 and 11 October were at 16:00, at 00:00 and at 08:00. A day that took of the contract's open interest off moved funding by one to two hundredths of a percent between prints, because funding averages the premium over eight hours and clamps it, and the premium's extremes on the day sat inside single hours. The default itself moves, too: in September 2025 only of ETHUSDT's prints were exactly , against across 2021 to 2023, because the premium had settled below its index. A funding print says where the premium averaged; read the premium.

Bars too long for the question. The lead lives inside the hour. On hourly bars, to of BTC and ETH moves from 2022 to 2024 showed both venues acting in the same hour, and at in 2024, to showed neither venue acting unusually at all. Re-read on 15-minute bars, those "neither" readings fell to to : the aggression was there, and the hour averaged it away, which is the derivatives preamble's warning about daily bars repeated one timeframe down. States on the anchor timeframe, leaders on 15 minutes.

A thin contract. On KAVA, with about a two-hundredth of ether's open interest, the calibration curve is there but flat. Its hours with the largest open interest changes moved price times as much as its quietest hours, against times for BTC and times for ETH, because a thin contract's ordinary hours are already noisy. Its unusual open interest still came with unusual volume ( normal in the top group), so the panel still sees activity. It says much less about price. And at on hourly bars, of KAVA's moves had neither venue acting unusually: on a thin enough contract a few accounts are the market, and a panel built to read crowds has no crowd to read.

The wrong venue. The panel reads one venue. If the leverage behind a move sits somewhere else, on another exchange or a decentralised perpetual like Hyperliquid, this venue's open interest won't show it, and its premium will follow by arbitrage without leading. The same goes for spot that never prints on the chart, such as ETF creations. Use the venue where the size is, and treat a quiet panel during a loud move as a sign the move is happening elsewhere.

Estimated delta read as aggression. TradingView's volume delta and CVD indicators classify each smaller candle's volume as up or down by its direction. That's an estimate of pressure, and it can disagree with the exchange's own record of which side crossed the spread. The Procedure takes taker delta from the venue for that reason.

Hedges read as views. A rise in open interest with the perpetual's sellers in front can be a fund selling the perpetual against spot it already holds, collecting the premium or funding with no view on price at all. The panel shows positions and aggression; why anyone holds them is inference, and the preamble's Who's on Each Side found CME's leveraged funds net short in most weeks of 2021 to 2023, a pattern that fits that trade.

Bad data read as a collapse. Binance's open interest archive publishes zero through its own outages: for ten hours from 15:00 on 7 March 2022 on all three contracts I fetched, and for an hour or two at a time since. A zero followed by a normal reading is a change of thousands of times normal, and a panel that believes it reports a flush that never happened. An open interest of zero on a live contract is missing data; treat it as missing.

Dated futures on a free account. The free plan allows typed spreads on the daily only, and a daily spread between CME and spot subtracts CME's 15:00 Chicago settlement from spot's midnight close, hours apart. That was the error in my own 2023 chart. The CME premium belongs on hourly data aligned to the settlement, which the preamble shows how to do, and stays out of this free configuration.

#Reference Card

• Asks: is this move being carried by spot or by leverage?
• Panel: two TradingView tabs, same venue and timeframe. Derivatives tab: VENUE:SYMBOL.P with volume, .P_PREMIUM (line on close) and .P_OI in their own panes. Spot tab: the spot symbol with volume. Funding off the panel.
• Normal: every series in multiples of the median size of its previous thirty days of bars; no bar sees a later one.
• Opening / closing: open interest change at or beyond with the direction of price; four states, opening ↑, closing ↑, opening ↓, closing ↓. Read on the anchor timeframe, then down.
• : –; lower catches more and earlier, with more noise; higher catches fewer, cleaner, later.
• Spot-led / perp-led / together: first venue whose 15-minute taker delta reaches in the direction of the move, searched over the six hours to the end of the move's first hour; together if the same bar. Taker delta from the venue's own data.
• Cumulative delta: from the Tool 01 climax bar, as (average bars) and (share of volume); read together.
• Premium: against its own normal; up with opening means buyers paid up, down means sellers leaned.
• Write: leader, state and thresholds, with the anchor, e.g. "perp-led opening ↓ (, )".
• Distrust: funding prints, hourly or longer bars for the lead, thin contracts, one venue during a move elsewhere, estimated CVD, zero open interest.

---

Applied Market Mechanics is a record of how I read markets, published purely as education. Nothing in it is financial advice or a recommendation to buy or sell anything, and I'm not a financial adviser. Every example has already resolved. I don't disclose positions, so nothing in an entry should be read as a statement of what I hold. Markets take money from people who follow a method exactly as often as from people who don't; size accordingly.