Open interest is the second-most-quoted number in perp trading after price, and the most consistently misread. You can find an OI chart on every analytics dashboard a Hyperliquid trader has open. You can find half a dozen takes per day on Twitter that lean entirely on its slope. And if you sit with the actual data for an hour, you can find that the people leaning hardest on it are wrong about what the number is doing at least half the time.
This post is a field guide to OI as a signal: what it actually measures, the three structural reasons it lies, and the only way we let it gate a trade inside a strategy file. It is written specifically for Hyperliquid perp traders because the venue has two quirks that change the math — a transparent on-chain vault layer and an unusually clean (and unusually queryable) OI feed.
If you have ever read "OI is rising, the move has legs" and felt unsure why, this is the long version of "well, sometimes."
What open interest actually measures.
Open interest is the total notional of perpetual futures contracts that are currently open — every contract opened and not yet closed, summed across both sides of the book. It is a stock, not a flow. Volume is the flow; OI is what is left on the field after each trade.
The number you usually see — "ETH-PERP OI: $1.42B" — is the dollar notional of the long side (equivalently, the short side, since the two net to zero). It tells you how much capital is sitting in the trade. It does not, by itself, tell you anything about direction. A market with $1.42B of OI is the same as a market with $1.42B of OI whether longs got there by patient accumulation, whether shorts got there by being squeezed, or whether a single delta-neutral vault built both sides over a weekend.
Where OI is genuinely useful is in its change. ΔOI tells you about positioning behavior — net contract creation or destruction over an interval — which combined with price and funding tells you a story you cannot get from price alone. The rest of this post is about that combination, and about the three traps that wait for traders who skip the combination and read OI as a standalone signal.
When OI is genuinely informative.
There are two reads of OI that we trust before any others.
OI up, price up, funding up. Three confirmations. New longs are paying to enter, the funding rate is repricing to reflect their crowding, and the price is being walked higher by their willingness to do exactly that. This is the textbook "fresh bid" pattern and it is real. On Hyperliquid in March, ETH-PERP went from $3,410 to $3,520 over six hours while OI rose from $1.20B to $1.45B and funding flipped from +0.005% to +0.018%. That triple confirmation is the only configuration in which we would take a fresh-longs read as actionable, and even then only as one input among several.
OI down, price up, funding flat or down. Shorts covering. The move is mechanical — short positions are being closed, often by stops and sometimes voluntarily, and the act of buying back to close is what is lifting price. This pattern looks like a rally and is in fact a release of pressure that, once released, has no fuel left. We use this read in the opposite direction from the first: a "rally" with falling OI is the kind we are happy to fade when the next regime gate allows it.
Anything outside those two clean configurations and the signal is muddier than most readers want to admit. OI down + price down + funding negative could be capitulation (good, mean-revert) or could be a measured de-risking (bad, more to come). OI up + price flat could be balanced new positioning or could be a single vault opening both legs. The chart alone will not tell you which.
The three ways OI lies.
We have watched OI mislead traders in three specific ways often enough to give them names.
One: vault-pair inflation. Hyperliquid's transparent vault architecture is one of the venue's best features and also the thing that makes its raw OI number the dirtiest of any major perp venue. A single market-making vault — HLP itself, or any of the dozens of community vaults — can open offsetting long and short positions across separate accounts as part of normal inventory management. That activity adds OI on both sides without representing any directional view. On a quiet day this can be a non-trivial fraction of the change in ETH-PERP OI, and on the day a new vault deploys it can be most of the change. The dashboard reads "OI up $80M"; the reality is "one vault rotated inventory."
Two: basis and hedge-driven shorts. When funding is persistently positive, basis traders short the perp against a long spot position to harvest funding. That short adds to OI. It is not directional and it is not going to cover on the next move up. A persistent + funding regime can produce a slow grind upward in OI driven almost entirely by hedgers, which the chart will quietly label "more shorts" — they are not shorts in the sense most readers mean.
Three: cohort flips during liquidations. A sharp price move triggers a liquidation cascade. Forced closes drop OI; opportunistic openers at the lows add it back. Twenty minutes later the OI number is roughly where it started — but the composition of who owns the open positions has rotated entirely. Anyone reading the chart who did not watch the tape during the cascade walks away thinking "no real change" when in fact the entire book just turned over.
The common thread: the dashboard cares about the level, the level can be quiet while the composition is loud, and the composition is where the next move comes from.
How to use OI without getting embarrassed.
There are three rules we follow inside actual strategy files, and we recommend them whether or not you ever run one of those files through Engine.
Use ΔOI, not OI. Levels are dirty for the structural reasons above. Deltas — change over a fixed window — are cleaner. The window matters: 30 minutes is good for short-horizon reads, 4 hours is good for regime reads. Anything longer than a day is mostly noise on a venue with active vault rotation.
Always require a second signal. OI never trades a position by itself in any strategy file we ship. It is a confirmation rule, always combined with funding sign and slope, sometimes with the top-trader ratio if you have access to one. The Funding Harvester strategy we walked through in a prior post does exactly this. The signal section reads:
## Signals
- funding_rate < -0.012% over 2 epochs
- Confirm with OI delta > +$1M in last 30m
The funding rule fires the candidate. The OI rule decides whether to actually take it. Either rule alone is too noisy; together they catch the configurations we care about and skip the noise. If you flip either one off in a backtest, the win rate barely moves but the average loss balloons — the kind of result that tells you the second signal was buying you risk reduction, not signal strength.
Audit your own signal on the decision log. This is the Engine-specific part of the recommendation. When the agent acts (or skips) on an OI rule, it writes the read into the decision log. You can scroll back to any timestamp and see the OI delta the agent used, the rule that fired, and the trade that followed. If you are using OI as a signal and cannot produce that audit trail, you have no way to know whether the signal is helping you or quietly hurting you. The point of writing the strategy as a file is not aesthetics; the file is what the audit checks against.
A useful number, not a strategy.
OI is a confirming signal, never a leading one. Treated as a leading signal it produces the worst class of error in systematic trading: a read that feels like evidence but is actually structural noise. Treated as a confirmation — paired with funding, used as a delta over a windowed interval, and exposed in the decision log so you can grade it after the fact — it earns a place in the rule stack.
The traders who get embarrassed by OI on Hyperliquid are the ones who treat the dashboard as oracular. The traders who use it well are the ones who treat it as one of four numbers they cross-check before acting, and who write down — in a file the agent reads — exactly which combination is allowed to fire. That is the whole technique. There is no clever read of OI hidden under the obvious one. There is just the discipline of using it where it actually adds information and ignoring it where it does not.