Crypto Liquidation Cascades: What Open Interest and Liquidation Feeds Can Prove
What Matters
Open interest does not reveal every trader's liquidation price, and a liquidation feed does not automatically measure cash losses or market-wide forced selling. DN's evidence and recovery register separates feed coverage, contract units, mark-price rules and order-book recovery. It provides a reproducible observation plan without inventing liquidation clusters, an agent positioning signal or a cascade risk score.
DN Evidence Block
Reviewed: 12 November 2026. Primary evidence: official exchange API and liquidation documentation. Instrument: DN Liquidation Evidence and Recovery Register. Boundary: no live feed collection, historical liquidation total, attributed agent-wallet cohort or measured recovery series is supplied. The worked depth observations and downloadable record are illustrative, with incomplete recovery evidence explicitly marked.
A Cascade Has Several Observable Parts
A liquidation cascade is a mechanism in which forced reductions and market conditions can reinforce each other. Studying it requires more than one large number. Analysts need to know which positions became eligible, how the liquidation engine handled them, what trading reached the market, how available depth changed and which observations are missing.
Those stages are related but not interchangeable. A position's liquidation threshold is not necessarily its execution price. A reported notional is not the trader's equity loss. A decrease in aggregate open interest can include voluntary closures. A decline in market capitalisation is a valuation change, not the amount forcibly sold. A chart that merges these quantities can look precise while answering no well-defined question.
Claim Versus What Public Data Can Establish
| Popular claim | Evidence actually needed | What aggregate data leaves unknown |
|---|---|---|
| Open interest shows a liquidation cluster at a given price | Position entry, margin, collateral, mode, funding and the venue's rules | The distribution of account-specific thresholds |
| Funding reveals agent positioning | Verified agent attribution and matched positions over time | Who holds the exposure and why |
| The feed reports every forced closure | Documented stream coverage, sampling and reconnect behaviour | Events outside the feed or lost during collection |
| Liquidation notional equals cash lost | Contract-unit conversion, equity, prices, fees and recoveries | Realised account-level loss |
| Fast book recovery proves bots stabilise the market | Attributed activity, baseline and a causal comparison | Whether bots, other participants or changing conditions caused recovery |
A heatmap can be a model, but the model must disclose its assumptions. Distributing aggregate open interest across guessed entry prices and leverage levels does not reveal actual positions. DN has removed that presentation from this article rather than publish a simulated distribution as an observed liquidation map.
Read the Feed Contract Before Adding the Numbers
Bybit's all-liquidation stream documents a 500-millisecond push frequency for its supported contract categories. Its field p is a bankruptcy price, not a promise of a public order-book execution at that price; v is executed size. Its side convention also needs reading: a Buy update identifies a liquidated long position. These details matter before an analyst labels a chart or converts a feed row into a value.
Binance's documented liquidation-order stream behaviour includes a latest-order snapshot per symbol within a 1,000-millisecond window. A collector must retain the exact market and stream version it uses. A sampled stream cannot be treated as an exhaustive historical tape simply because its connection stayed open.
Record event time and receipt time separately, alongside sequence or deduplication fields where the service provides them. A reconnect can lose events, replay events or require a snapshot resynchronisation. Duplicate records from two aggregators should not be summed unless they represent distinct underlying events. A successful websocket connection is evidence of connectivity, not completeness.
Contract Units Are Part of the Measurement
Before multiplying size by price, identify whether size means base asset, quote currency, contract count or a fixed contract value. Inverse and linear instruments need their own conversion rules. Preserve the raw fields and the instrument specification so the conversion can be audited. A single generic “quantity × price” formula is not valid for every feed.
Even a correct notional conversion remains a notional measure. To estimate an account's net loss, an analyst would need the relevant equity and settlement information, which public streams may not provide. The absence of that data should remain unknown. It should not be filled by assuming that every liquidated position lost its entire notional.
The same discipline applies to daily totals: state which venues, products, event definitions, time zone and missing windows are included. A number covering three exchanges is a three-exchange observation, not a global total. Later backfills should produce a revised dated result instead of silently changing a previously published figure.
The Liquidation Engine Changes the Interpretation
Hyperliquid's documentation describes mark-price-based eligibility, order-book liquidation attempts, partial-liquidation rules and a backstop mechanism. It therefore does not support the blanket assertion that every liquidation instantly becomes the same kind of full market order. Cross-margin and isolated positions also have different account implications.
dYdX's liquidation documentation describes protocol rules for liquidation orders, including fillable-price and governance parameters. Its contract-loss explanation separately covers insurance-fund and deleveraging mechanisms. Those mechanisms should not be collapsed into one universal formula or confused with an insurance policy issued to a trader.
Centralised venues likewise publish product-specific rules. OKX's August 2026 explanation distinguishes maintenance-margin conditions, partial reduction and ADL, with applicability caveats. BloFin's July 2026 FAQ distinguishes mark, liquidation and bankruptcy prices. Read the applicable product and account mode rather than importing a rule from another venue.
DN Liquidation Evidence and Recovery Register
The register has two related records. The event record captures the feed's identity, product, raw units, price field meaning, timestamps, coverage limitations and deduplication method. The recovery record captures the order-book definition, reference price, depth band, baseline, observation frequency and rule used to declare recovery. Neither record assigns an agent identity automatically.
| Recovery design choice | What to specify before collecting |
|---|---|
| Reference market | Venue, instrument, quote unit and order-book source |
| Depth band | Distance from the chosen mid or reference price, such as an explicitly illustrative ±25 basis points |
| Baseline | Pre-event window and aggregation method for each side |
| Threshold | Required fraction of baseline bid and ask depth, tested separately |
| Persistence | Required duration or consecutive valid samples above threshold |
| Gaps | Maximum acceptable interval and whether missing data resets confirmation |
| Outcome | First observed threshold crossing, sustained recovery, not recovered or unknown |
Testing bid and ask depth separately avoids a useful-looking total that hides one depleted side. The reference price and band convention matter during a sharp move: a band following the mid measures something different from fixed absolute price levels. Neither is universally correct; choose the definition that matches the question and disclose it.
A Hypothetical Recovery Observation
Suppose pre-event bid depth is 50,000 reference units and ask depth is 60,000 within the chosen band. The research plan requires each side to reach at least 90% of its own baseline for five seconds. The thresholds are therefore 45,000 and 54,000 units.
At a sample ten seconds after the event, bid depth is 48,000 and ask depth is 55,000. Both clear the thresholds at that observation. Without the intervening samples and the required persistence window, the correct result is observed threshold pass; sustained recovery undetermined. It is not a measured ten-second recovery time. The downloadable example preserves that distinction.
A genuine study would also state how it chooses event start and end, handles overlapping events and compares calm and stressed markets. An agent-versus-other-liquidity analysis requires verified participant or strategy attribution. Fast replenishment after a liquidation burst alone cannot identify who supplied it or whether their activity caused the improvement.
Why Funding Cannot Supply the Missing Agent Cohort
Funding, open interest and subsequent wallet movements can be observed together without establishing a leading agent-positioning indicator. A valid study needs a documented agent definition, entity resolution, exposure direction, observation times and out-of-sample evaluation. A market-maker wallet, exchange omnibus wallet or automated keeper is not necessarily an autonomous agent portfolio.
Confounders include spot hedges, cross-venue positions, changing margin and price-driven notional changes. Selecting only wallets that later moved in the expected direction would contaminate the test. DN therefore retains the proposed agent-positioning measurement as an evidence hold. This article explains the required collection design; it does not assign synthetic identities or publish a correlation that has not been measured.
How Readers Can Use the Evidence
For research, keep raw provider records and versioned transformations, then publish coverage alongside conclusions. For an individual position, use the venue's current account-specific risk information rather than treating an aggregate heatmap as a safe stop location. For infrastructure, monitor data gaps and changing feed semantics as operational failures in their own right.
DN retains existing commercial links to Bybit, BloFin and OKX as optional reference routes for eligible readers reviewing those products. DN may receive compensation. These links do not establish suitability, availability in a particular jurisdiction, lower liquidation risk or guaranteed execution. The official documentation, account terms and live position state remain the relevant inputs.
Methodology and Limitations
This review compares documented mechanisms and data semantics. It does not independently audit exchange matching engines, classify every liquidation or reproduce a historical episode. The depth values are fictional arithmetic; there is no inferred Q4 2026 market condition or score.
Refresh when a stream version, contract specification, margin system, liquidation mechanism or collection method changes. A source review date does not make an absent dataset current. No automatic collection schedule or public API is activated by this record. For borrowing mechanics, see the DeFi liquidation buffer guide; for market structure, the existing order-book and liquidation discussion provides related context.
Frequently asked questions
Can open interest locate actual liquidation prices?
Not by itself. Account-specific margin, entry, collateral, funding and venue rules are missing from the aggregate.
Is bankruptcy price the same as a market execution price?
No. Read the provider's field definition. A bankruptcy reference can differ from the price of a public trade or later position disposal.
Does a liquidation feed measure traders' cash losses?
Not automatically. Feed notional, account equity, fees and realised loss are different quantities, and contract units must first be interpreted correctly.
Does the register prove that liquidation bots stabilise markets?
No. It defines evidence for studying depth recovery. Participant attribution and causal comparison remain separate requirements.
Corrections and Changelog
12 November 2026: Removed invented price-cluster distributions, unvalidated cascade scores and unsupported historical totals. Replaced them with source-specific feed interpretation and an explicit recovery observation protocol. The canonical URL and relevant internal and commercial links are retained.
Corrections should provide the exact feed, document version, field and observation period. The article offers research methodology, not a forecast of the next cascade, an agent trading signal or a safe leverage recommendation.
Machine-readable review record
The record below is a bounded illustrative example. It is not a wallet scan, current-chain dataset or assessment of a deployed service. The schema identifier is an identifier, not a promised API endpoint. Downloads are local snapshots; no schedule or automatic refresh is active.
Primary sources
- Bybit all-liquidation stream — 500ms push; p is bankruptcy price; v size; side convention.
- Binance derivatives websocket streams — Latest liquidation-order snapshot per symbol within1000ms; preserve exact stream scope/version.
- Hyperliquid liquidation documentation — Mark price, order-book attempts, partial and backstop mechanics.
- dYdX liquidations — Liquidation orders and protocol/governance parameters.
- dYdX contract-loss mechanisms — Insurance-fund and deleveraging mechanisms distinct from a trader insurance policy.
- OKX liquidation explanation — Current product-specific margin, partial reduction and ADL documentation; applicability caveats.
- BloFin liquidation FAQ — Current mark/liquidation/bankruptcy price distinction.