HyperMultiAssetedADL

Complete Analysis of the October 10, 2025 Hyperliquid Auto-Deleveraging (ADL) Event

Largest known ADL event analysis: $2.1 billion in forced position closures across 162 assets in 12 minutes. Analysis based on blockchain data with complete clearinghouse state reconstruction.

CANONICAL DATA FILE

For researchers: Use ONLY the real-time reconstructed file (cash-only baseline):

adl_detailed_analysis_REALTIME.csv

This file contains:

34,983 ADL events (100% coverage - ADLs occurred during 21:16:04-21:26:57 UTC within the full 12-minute event window)
Real-time account values (cash-only) at exact ADL moment (snapshot unrealized PnL removed)
Real-time leverage calculated with reconstructed account states:
- Median: 0.20x ← Most ADL'd accounts use low leverage
- 95th percentile: 5.10x ← Even high-leverage accounts are well below Hyperliquid's 40x max
- 99th percentile: 122.69x ← Outliers are data artifacts from liquidation delays (see docs/findings/HIGH_LEVERAGE_OUTLIERS_EXPLANATION.md)
Negative equity detection (302 accounts, −$23,191,104.48 aggregate deficit)
Zero shortcuts - 3.2M events processed chronologically

Processing details:

3,239,706 events processed (fills, funding, deposits, withdrawals)
437,723 accounts reconstructed in real-time
Full event window: 21:15:00 - 21:27:00 UTC (12 minutes) - complete analysis timeframe
ADL events occurred: 21:16:04 - 21:26:57 UTC (10.88 minutes) - actual ADL activity period
Liquidations occurred: 21:15:03 - 21:26:57 UTC (11.90 minutes) - liquidation activity period
Method: Chronological event replay from clearinghouse snapshot

Canonical dataset status: Only the cash-only reconstructed CSVs and raw S3 extracts are now present in this repository. Every analysis script, markdown study, and CSV artifact is derived directly from the canonical files in data/canonical/cash-only balances ADL event orderbook 2025-10-10/:

adl_detailed_analysis_REALTIME.csv (34,983 ADL events with fixed position size calculation)
adl_fills_full_12min_raw.csv (raw ADL fills)
liquidations_full_12min.csv (liquidation events)

📁 Repository Structure

The repository is organized into clear directories for easy navigation:

HyperMultiAssetedADL/
├── README.md                    # This file - repository overview
├── .gitignore                   # Git ignore rules
│
├── docs/                        # All documentation
│   ├── methodology/            # Methodology and how-to guides
│   ├── findings/               # Research findings and discoveries
│   ├── analysis/               # Analysis reports
│   └── reports/                # Verification and audit reports
│
├── scripts/                     # All Python scripts
│   ├── analysis/               # Analysis scripts (9 scripts)
│   │   └── outputs/            # Analysis script JSON outputs
│   ├── data/                   # Data extraction scripts
│   ├── reconstruction/         # Account reconstruction scripts
│   └── verification/           # Verification and testing scripts
│
└── data/                        # All data files
    ├── canonical/               # Canonical processed data
    │   └── cash-only balances ADL event orderbook 2025-10-10/
    │       ├── adl_detailed_analysis_REALTIME.csv
    │       ├── adl_by_user_REALTIME.csv
    │       ├── adl_by_coin_REALTIME.csv
    │       ├── adl_fills_full_12min_raw.csv
    │       └── liquidations_full_12min.csv
    └── raw/                     # Raw analysis outputs
        ├── ADL_ORDERS_COMPLETE_LIST.csv
        └── high_leverage_outliers_analysis.csv

Key Benefits:

✅ Clear separation of documentation, code, and data
✅ Easy to find what you need
✅ Organized by purpose (methodology, findings, analysis, reports)
✅ Scripts grouped by function (analysis, data, reconstruction, verification)

Executive Summary

Event: October 10, 2025 Market Crash Full Analysis Window: 21:15:00 - 21:27:00 UTC (12 minutes - complete event timeframe) ADL Activity Period: 21:16:04 - 21:26:57 UTC (10.88 minutes - when ADLs actually occurred) Data Source: Hyperliquid S3 (ADL events from blockchain data)

Note: The full 12-minute window (21:15-21:27) includes the complete cascade including liquidations that started at 21:15:03. ADL events specifically occurred from 21:16:04 onwards, after the first liquidations triggered the ADL mechanism.

Key Findings

Metric	Value
Total Assets ADL'd	162 tickers
Total ADL Events	34,983 events
Total Net Notional	$2.10 BILLION
Total Realized PNL	$834.3 Million
Total Negative Net Equity	−$23,191,104.48 (302 accounts)

Net Equity Definition: Net equity = Cash Balance + Total Unrealized PNL. Negative net equity represents bad debt that must be covered by the insurance fund. See Insurance Fund Impact section for details.

Top 5 ADL'd Assets

Rank	Ticker	Net Notional	% of Total	# Events
1	BTC	$620.9M	29.5%	2,443
2	ETH	$458.0M	21.8%	1,498
3	SOL	$276.2M	13.1%	3,031
4	HYPE	$189.9M	9.0%	6,229
5	XPL	$65.8M	3.1%	2,984

Top 3 (BTC, ETH, SOL): 64.4% of total ADL volume

Analysis Scripts (Canonical Replay)

Study	Python Script
`docs/findings/PER_ASSET_ISOLATION.md`	`scripts/analysis/per_asset_isolation.py`
`docs/analysis/CASCADE_TIMING_ANALYSIS.md`	`scripts/analysis/cascade_timing_analysis.py`
`docs/analysis/BATCH_PROCESSING_DISCOVERY.md`	`scripts/analysis/batch_processing_analysis.py`
`docs/findings/ADL_MECHANISM_RESEARCH.md`	`scripts/analysis/adl_mechanism_analysis.py`
`docs/findings/ADL_PRIORITIZATION_VERIFIED.md`	`scripts/analysis/adl_prioritization_analysis.py`
`docs/findings/ADL_PRIORITIZATION_ANALYSIS_LOCAL.md`	`scripts/analysis/adl_prioritization_local.py`
`docs/findings/INSURANCE_FUND_IMPACT.md`	`scripts/analysis/insurance_fund_impact.py`
`data/canonical/cash-only balances ADL event orderbook 2025-10-10/ADL_NET_VOLUME_FULL_12MIN.md`	`scripts/analysis/adl_net_volume.py` (generated in canonical directory)
`docs/analysis/TOTAL_IMPACT_ANALYSIS.md`	`scripts/analysis/total_impact_analysis.py`

Each script loads the canonical CSVs in this repository and emits the metrics cited in the corresponding study (plus a JSON snapshot in scripts/analysis/). Run them from the repo root:

python3 scripts/analysis/<script_name>.py

Canonical Results Snapshot (Nov 13, 2025)

Study	Key Output
Per-Asset Isolation	100 shared timestamps, 0 cross-asset cases, Jaccard overlap 96.74%
Cascade Timing	First ADL at 61.7s after first liquidation; largest burst 11,279 liq + 11,279 ADL in second 61
Batch Processing	224 timestamps total, first 61s liquidation-only, all shared timestamps run `liquidation → ADL` sequentially
Counterparty Mechanism	100% ADL events carry `liquidated_user`; highlighted $174.18M ETH ADL matched by 265 ETH liquidations
ADL Prioritization (global)	99.4% profitable ADL targets, median leverage 0.20x, p95 5.10x, p99 122.69x (outliers - see `docs/findings/HIGH_LEVERAGE_OUTLIERS_EXPLANATION.md`)
ADL Prioritization (local)	Spearman ρ (PNL vs notional −0.2207), (PNL vs leverage −0.4781); repeated winners table in JSON
Insurance Fund Impact	302 negative-equity accounts (0.86% of ADL), aggregate deficit −$23,191,104.48
ADL Net Volume	Total ADL notional $2,103,111,431, 34,983 events across 162 tickers
Total Impact	Liquidations $5,511,042,925 + ADL $2,103,111,431 = $7,614,154,356 across 98,620 events
Comprehensive Verification	`python3 scripts/verification/verify_all_findings.py` passes all suites (prioritization, isolation, counterparty, timing, insurance, integrity)

COMPLETE METHODOLOGY: For Researchers

** docs/methodology/COMPLETE_METHODOLOGY.md** - Comprehensive guide to reproduce our entire analysis

What's Inside:

All data sources (S3 event data + clearinghouse snapshots)
Step-by-step acquisition (how to download and decompress)
Complete processing pipeline (from raw data to insights)
Data reconciliation (how we merged multiple data sources)
Reproducibility guide (reproduce all 34,983 ADL event analyses)
Common pitfalls & solutions (save hours of debugging)

Perfect for:

Researchers wanting to reproduce our findings
Teams building on this analysis
Anyone needing to understand the complete data flow

Clearinghouse Data Access

November 12, 2025 - Complete clearinghouse data now available

Previously Unavailable -> Now Available

Data Point	Previous Status	Current Status
Entry Prices	NULL for 88% of positions	Calculated from fills
Leverage Ratios	Requires clearinghouse state	REAL-TIME for 34,983 ADL events (100%)
Unrealized PNL	Can't calculate without entry	Real-time for all positions
Account Values	Not available	437,723 accounts - REAL-TIME RECONSTRUCTED
Negative Equity	Not trackable	302 accounts identified (−$23.19M insurance impact)

What We Now Have

Real-Time Account Reconstruction (Processing 3.2M events from snapshot to cascade end):

437,723 accounts with real-time account values reconstructed
Initial state: $5.1B total at 20:04:54 UTC (70 min before cascade)
3,239,706 events processed: Fills, funding, deposits, withdrawals
Every account state updated chronologically through the COMPLETE 12-minute cascade

Calculated for Every ADL Event (REAL-TIME):

Entry prices (weighted average from fills)
Leverage ratios at ADL moment (real-time account values)
Unrealized PNL at ADL time (all positions, real-time prices)
Total equity (cash + unrealized PNL)
Negative equity detection (account underwater)
PNL% (unrealized_pnl / position_notional × 100)

Analysis Coverage: 34,983 ADL events (100% of all ADL events) with complete real-time data

This clearinghouse data enabled the ADL prioritization analysis detailed below.

Key Finding: ADL Targets Profit, Not Leverage

See: ADL_PRIORITIZATION_VERIFIED.md

Common assumption: "ADL targets the highest leverage positions"
Analysis result: Evidence indicates ADL targets the most profitable positions

The Evidence (34,983 Real-Time ADL Events Analyzed - 100% Coverage)

Metric	Value
Profitable positions ADL'd	99.4% (34,775 / 34,983)
Average unrealized PNL	+80.58%
Median unrealized PNL	+50.09%
Median leverage (REAL-TIME)	0.20x
95th percentile leverage	3.22x
99th percentile leverage	74.18x
Negative equity accounts	302 (0.86%)
Total negative net equity	−$23,191,104.48

Note on leverage: Most ADL'd positions had low leverage. The median of 0.20x indicates that most ADL'd accounts used conservative leverage. The 95th percentile at 5.10x is well below Hyperliquid's 40x maximum. The 99th percentile at 122.69x represents data artifacts from liquidation delays (accounts with near-zero value when ADL closed them), not actual high-leverage trading. See docs/findings/HIGH_LEVERAGE_OUTLIERS_EXPLANATION.md for details. This indicates that ADL does not primarily target high leverage positions.

Top 10 ADL'd Positions (By Size)

Coin	Notional	PNL%	Leverage	Account Value
BTC	$193.4M	+12.73%	0.66x	$159.5M
ETH	$174.2M	+21.84%	1.79x	$82.7M
BTC	$76.4M	+12.60%	0.66x	$159.5M
BTC	$70.6M	+13.82%	2.92x	$29.1M
SOL	$46.7M	+16.07%	2.01x	$23.2M
ETH	$41.3M	+26.37%	1.73x	$82.7M
ETH	$41.2M	+26.47%	1.42x	$29.1M
ETH	$38.3M	+33.08%	2.11x	$18.1M
BTC	$30.3M	+10.37%	4.80x	$6.3M
SOL	$29.5M	+35.77%	0.54x	$54.4M

Every single one was PROFITABLE. This is not a coincidence.

What This Means

LOW LEVERAGE ≠ SAFE FROM ADL HIGH PROFIT = ADL TARGET

If you're sitting on a huge unrealized gain during a liquidation cascade, you're getting ADL'd—regardless of leverage.

Key Insight: ADL is a forced exit mechanism for winners, not punishment for reckless traders. The protocol uses your profits to cover liquidated losses.

Full Analysis: ADL_PRIORITIZATION_VERIFIED.md

INSURANCE FUND IMPACT: Quantifying the Underwater Accounts

** Real-Time Reconstruction Reveals**: 302 accounts in negative equity

Total Negative Net Equity: −$23,191,104.48

Definition of Net Equity: Net equity (also called "total equity") is calculated as:

Net Equity = Cash Balance + Total Unrealized PNL

Where:

Cash Balance = Account value after removing initial unrealized PNL from snapshot (cash-only baseline)
Total Unrealized PNL = Sum of unrealized profit/loss for all open positions at the ADL moment
- For long positions: size × (current_price - entry_price)
- For short positions: abs(size) × (entry_price - current_price)

Negative Net Equity occurs when Net Equity < 0, meaning the account's total value (cash + unrealized PNL) is negative. This represents bad debt that must be covered by the insurance fund.

The Numbers

Metric	Value
Accounts underwater	302 (0.86% of ADL'd)
Total negative net equity	−$23,191,104.48
Insurance fund coverage required	$23,191,104.48
Average underwater account	-$76,791.74

What This Means

When an account's net equity (cash + unrealized PNL) goes negative, losses must be socialized:

ADL activates to close out profitable positions
Underwater losses get absorbed by the insurance fund
If insurance fund insufficient -> socializes losses to remaining traders

This cascade required $23,191,104 in insurance fund coverage to prevent loss socialization.

Real-Time Reconstruction Achievement

This is the first time negative equity has been quantified for a Hyperliquid cascade:

Processed 3.2M events chronologically (COMPLETE 12-minute window)
Reconstructed 437,723 account states in real-time
Calculated equity at every ADL moment
Identified exact underwater amount
100% event coverage (34,983 / 34,983 ADL events)

Methodology: scripts/reconstruction/full_analysis_realtime.py

Key Finding: Per-Asset Isolation - No Cross-Asset ADL Contagion

See: PER_ASSET_ISOLATION.md

Common assumption: "BTC liquidations can trigger ETH ADL" or "ADL contagion across assets"
Analysis result: Zero cases of cross-asset ADL found

Key Evidence

Metric	Result
Timestamps analyzed	100 (liquidations + ADL in same timestamp)
Cross-asset ADL cases	0 (ZERO)
Ticker overlap	96.74%
Perfect 1:1 ratio matches	44/44 tickers at biggest burst

Analysis Results

BTC liquidations cause only BTC ADL (not ETH, SOL, or other assets)
ETH liquidations cause only ETH ADL (not BTC, SOL, or other assets)
SOL liquidations cause only SOL ADL (not BTC, ETH, or other assets)
Each asset has an independent ADL engine (no shared risk pool)
1:1 matching per asset when ADL triggers

Important Distinction

ADL contagion (technical): Does not exist
Market contagion (price dynamics): Does exist

Example:

BTC crashes -> Market panic -> Traders sell all assets
 v v v
BTC price v Psychology ETH price v, SOL price v
 v v v
BTC liquidations ETH liquidations SOL liquidations
 v v v
BTC ADL ONLY ETH ADL ONLY SOL ADL ONLY

Market contagion: YES (prices correlate)
ADL contagion: NO (ADL systems isolated)

Analysis of 100 timestamps proves:

0/100 cases where Asset X liquidations caused Asset Y ADL
When 44 assets had liquidations simultaneously, each got its own ADL (no spillover)
Perfect architectural isolation despite $7.6B cascade

Full analysis: PER_ASSET_ISOLATION.md

TOTAL MARKET IMPACT (Liquidations + ADL)

Complete cascade analysis now available

Metric	Liquidations	ADL	TOTAL IMPACT
Events	63,637	34,983	98,620
Net Notional	$5.51B	$2.10B	$7.61 BILLION
Realized PNL	-$607.7M	$834.3M	$226.6M net

** This represents the largest documented liquidation cascade event:**

$7.6 BILLION in forced closures in 12 minutes
98,620 forced events (liquidations + ADL)
$5.5B liquidated -> $2.1B ADL'd to cover losses

See full analysis: TOTAL_IMPACT_ANALYSIS.md (or see data/canonical/cash-only balances ADL event orderbook 2025-10-10/ADL_NET_VOLUME_FULL_12MIN.md for ADL-specific analysis)

NEW: ADL Mechanism Research - How It Really Works

1. Individual Event Analysis

** See: ADL_MECHANISM_RESEARCH.md**

We analyzed the largest single ADL event ($174.18M ETH) to understand how ADL is triggered using empirical blockchain data:

2. CASCADE TIMING DISCOVERY

** See: CASCADE_TIMING_ANALYSIS.md**

Key finding: Liquidations occur in waves before ADL activates.

3. BATCH PROCESSING DISCOVERY

** See: BATCH_PROCESSING_DISCOVERY.md**

Key finding: Liquidations and ADL execute in separate, sequential batches.

Even when they share the same millisecond timestamp, liquidations and ADL are processed sequentially, not concurrently:

Finding	Evidence
Same timestamp	Both recorded at `21:16:04.831874`
Different batches	11,279 liquidations -> THEN 11,279 ADLs
Zero interleaving	0% mixing across 100 analyzed timestamps
Universal pattern	100% of events show liquidation -> ADL order

The Architecture:

Block at timestamp T:
 Phase 1: Process ALL liquidations (liquidation engine)
 Phase 2: Calculate total losses & ADL requirements
 Phase 3: Select profitable positions for ADL
 Phase 4: Process ALL ADLs (ADL engine)

All events stamped with timestamp T, but sequenced internally

Why This Matters:

Reveals internal processing order (liquidation engine -> ADL engine)
Proves sequential dependency (ADL calculated AFTER liquidations)
Explains visual patterns (chunks on visualization are REAL batches)
No concurrent liquidation+ADL (clear execution phases)

Technical Detail: At the largest burst, 22,558 events occurred at the exact same millisecond, but analysis of event ordering shows perfect batch separation: events 710-11,988 were all liquidations, events 11,989-23,267 were all ADLs. Average batch run length: 11,279 events (no interleaving detected).

Metric	Value	Insight
First liquidation	0.0 seconds	Cascade starts (T+3s absolute)
First ADL	61.7 seconds later	≈62-second delay before ADL kicks in
Liquidations before ADL	710 events	System tries normal methods first
Correlation	0.945	Liquidations predict ADL
Biggest burst	22,558 events/second	11,279 liqs + 11,279 ADLs

The Pattern:

0-61s: 710 liquidations, 0 ADL ← ADL hasn't kicked in yet
62s:   11,279 liquidations + 11,279 ADL ← MASSIVE burst when threshold hit
63-180s: Alternating waves ← Liquidations → ADL → Liquidations → ADL

Why This Matters:

ADL is NOT instantaneous – there's a ~62-second delay
ADL activates in BURSTS (threshold-based, not continuous)
Liquidations accumulate → Threshold reached → ADL fires
Explains the "chunks" pattern visible on HyperFireworks visualization

Key Discovery: ADL is a Direct Counterparty to Liquidations

The $174M ETH ADL had 265 corresponding liquidations at the EXACT same timestamp:

Event Type	Amount	User	What Happened
Liquidations	$204.67M	`0xb0a5...540`	265 ETH longs liquidated (losing money)
ADL	$174.18M	`0x2ea1...3f4`	1 ETH short ADL'd (winning forced to close)

Timeline:

ETH price crashed -> User's 265 long positions hit liquidation price
$204.67M in liquidations triggered -> Exchange needs sellers
Profitable short holder ADL'd for $174.18M -> Provides liquidity
Insurance/HLP fund covers remaining $30M gap

What This Means

ADL is NOT random - It's triggered by liquidation events ADL provides counterparty liquidity - When liquidations happen, ADL supplies the opposite side Same-millisecond execution - Liquidation -> ADL happens instantly Profitable traders pay the price - Winners get force-closed to save losers from socialized losses

Why This Matters for Research

This is the first empirical documentation of ADL-liquidation coupling:

Proves ADL is triggered BY liquidations (not independent)
Shows exact timing relationship (same millisecond)
Quantifies the counterparty relationship ($174M ADL <-> $205M liquidations)
Explains why insurance funds don't cover 100% (ADL does most of the work)

Full analysis with transaction hashes, addresses, and blockchain verification: ADL_MECHANISM_RESEARCH.md

Major Insights

Market Concentration

$2.1 BILLION in forced ADL closures over 12 minutes
BTC, ETH, SOL dominate: $1.36B (64.4% of total)
Top 10 tickers: $1.78B (84.6% of total)
Long tail: 152 tickers share remaining 15.4%

ADL Rate

34,983 ADL events in 12 minutes
Average: 2,915 ADLs per minute
Peak rate: ~49 ADLs per second

Asset Mix

Major cryptos: BTC, ETH, SOL led the way
Meme coins: PUMP ($57.3M), FARTCOIN ($32.0M)
DeFi tokens: LINK ($21.5M), UNI ($8.0M), AAVE ($2.5M)
New launches: HYPE ($189.9M), XPL ($65.8M)

Profitability

$834.3M in realized PNL forced to close
Most profitable ticker: XPL ($119.4M), ETH ($110.6M), FARTCOIN ($68.3M)
Average PNL per event: $23,844

Key Files

Documentation

README.md - This file - repository overview
docs/methodology/COMPLETE_METHODOLOGY.md - Complete methodology guide
docs/findings/ - All research findings and discoveries
docs/analysis/ - Analysis reports
docs/reports/ - Verification and audit reports

Data

data/canonical/cash-only balances ADL event orderbook 2025-10-10/ - Canonical processed data
- adl_detailed_analysis_REALTIME.csv - 34,983 ADL events with real-time metrics
- adl_fills_full_12min_raw.csv - Raw ADL fills
- liquidations_full_12min.csv - Liquidation events
data/raw/ - Raw analysis outputs

Scripts

scripts/analysis/ - Analysis scripts (9 scripts)
scripts/reconstruction/ - Account reconstruction scripts
scripts/verification/ - Verification and testing scripts
scripts/data/ - Data extraction scripts

Top 20 ADL'd Tickers

Rank	Ticker	Net Notional	# Events	Total PNL
1	BTC	$620.9M	2,443	$72.8M
2	ETH	$458.0M	1,498	$110.6M
3	SOL	$276.2M	3,031	$69.1M
4	HYPE	$189.9M	6,229	$51.7M
5	XPL	$65.8M	2,984	$119.4M
6	PUMP	$57.3M	1,868	$31.5M
7	ENA	$42.5M	360	$50.2M
8	AVAX	$36.6M	407	$23.8M
9	FARTCOIN	$32.0M	1,999	$68.3M
10	XRP	$31.4M	607	$13.3M
11	ASTER	$27.2M	431	$15.4M
12	LINK	$21.5M	259	$12.6M
13	LTC	$21.1M	197	$6.7M
14	ZEC	$18.9M	400	$2.5M
15	DOGE	$16.5M	249	$10.5M
16	SUI	$13.4M	442	$9.9M
17	PENGU	$10.6M	372	$9.2M
18	MNT	$10.1M	211	$5.6M
19	IP	$9.3M	270	$6.4M
20	TAO	$8.7M	167	$5.8M

Comparison: 2-Minute Sample vs Full Event

Metric	2-Minute Sample	Full 12-Minute	Scaling Factor
Time Window	21:15-21:17 UTC	21:15-21:27 UTC	6.0x
Total Notional	$285.5M	$2,103.1M	7.37x
Total Events	14,194	34,983	2.46x
Assets	65 tickers	162 tickers	2.49x

Why 7.37x instead of 6x?

ADL events were not evenly distributed over time
Peak activity around 21:19-21:20 UTC (after the 2-min sample)
The 2-minute sample (21:15-21:17) was relatively early in the event

What is ADL Net Volume?

Auto-Deleveraging (ADL) is Hyperliquid's mechanism to manage liquidations during extreme market volatility:

When positions are liquidated but can't be closed by the liquidation engine
The protocol force-closes the most profitable opposing positions
This is called "Auto-Deleveraging" (ADL)

Net Volume = Sum of all position sizes that were ADL'd per ticker Net Notional = Sum of (position size × price) for all ADL'd positions

Methodology

Data Source

File: node_fills_20251010_21.lz4 (S3 bucket)
Total fills analyzed: 1,424,266 fills
ADL fills: 34,983 (2.5% of all fills)
Filtered: Excluded @ tokens (spot positions)

Calculations

# Net Volume per ticker
net_volume = sum(size) for all ADL events per ticker

# Net Notional per ticker 
net_notional = sum(size × price) for all ADL events per ticker

# Total Realized PNL per ticker
total_pnl = sum(closed_pnl) for all ADL events per ticker

Filters Applied

Direction = "Auto-Deleveraging" (blockchain label)
Exclude tickers starting with "@" (spot positions)
Time window: 21:15:00 - 21:27:00 UTC (full 12 minutes)

Data Quality

Data Source:

Complete 12-minute dataset (not a sample)
Only fills with explicit "Auto-Deleveraging" label from blockchain data
No heuristics: Direct from S3 node_fills
Spot positions excluded: @ tokens filtered out
Cross-validated: Matches expected event timeline

Source: Hyperliquid S3 node_fills_20251010_21.lz4 Processing time: ~30 seconds Records: 42,893 blocks -> 1.42M fills -> 34,983 ADL events

Usage

View Results

Quick view (CSV):

open adl_net_volume_full_12min.csv

Detailed report (Markdown):

open "data/canonical/cash-only balances ADL event orderbook 2025-10-10/ADL_NET_VOLUME_FULL_12MIN.md"

Individual fills:

open "data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_fills_full_12min_raw.csv"

Rerun Analysis

python3 scripts/data/extract_full_12min_adl.py

Load in Python

import pandas as pd

df = pd.read_csv('adl_net_volume_full_12min.csv')
print(f"Total: ${df['net_notional_usd'].sum():,.0f}")

# Top 10 tickers
print(df.nlargest(10, 'net_notional_usd'))

# BTC stats
btc = df[df['ticker'] == 'BTC'].iloc[0]
print(f"BTC: ${btc['net_notional_usd']:,.0f} across {btc['num_adl_events']} events")

For Academic Research

Suitable For

ADL mechanism analysis (largest known event)
Market concentration studies ($2.1B in 12 minutes)
Liquidity crisis behavior
Cross-asset contagion effects
Forced closure impact on traders

Key Research Questions This Dataset Answers

Event-Level (All Datasets):

How effective is ADL? -> $2.1B processed in 12 minutes
Which assets are most affected? -> BTC, ETH, SOL dominate
How concentrated is ADL? -> Top 3 = 64.4% of volume
What's the trader impact? -> $834M in forced PNL closures
How fast does it happen? -> 2,915 ADLs per second at peak

Account-Level (NEW - With Clearinghouse Data): 6. What leverage do ADL'd positions have? -> Median 0.20x; 95th percentile 5.10x (below 40x max); 99th percentile 122.69x (outliers from liquidation delays - see docs/findings/HIGH_LEVERAGE_OUTLIERS_EXPLANATION.md) 7. How profitable are ADL'd positions? -> 99.4% profitable, avg +85.9% PNL 8. Does ADL target high leverage? -> NO - targets high PROFIT 9. What are entry prices? -> Calculated for 34,983 positions (100% coverage) 10. Which accounts have highest risk? -> Tracked across 437,723 accounts

Citation

Event Data:

ADL Net Volume Analysis (2025). "Auto-Deleveraging Volume Analysis: 
October 10, 2025 Market Event - Full 12-Minute Window." 
Data: Hyperliquid S3 node_fills (ADL events from blockchain data).
Time: 21:15:00 - 21:27:00 UTC.
Total: $2.10B across 162 tickers, 34,983 events.

Clearinghouse Analysis (NEW):

ADL Prioritization Analysis (2025). "Real-Time Account Reconstruction:
October 10, 2025 Market Event (12-minute cascade)."
Data: Hyperliquid clearinghouse snapshot (Block 758750000) + full event stream (3,239,706 events).
Coverage: 34,983 ADL events with real-time leverage, entry prices, and equity.
Key Finding: ADL targets PROFIT (99.4% profitable), not leverage (median 0.20x).

Position-Level Data: What's Available

For researchers analyzing individual positions, here's what data we have:

Available in `adl_detailed_analysis_REALTIME.csv` (34,983 ADL'd positions - 100% Coverage)

Real-time reconstruction complete - All metrics calculated at exact ADL moment for the full 12-minute cascade

What You Need	Column Name	Description
Absolute PNL	`position_unrealized_pnl`	Unrealized PNL at ADL time (real-time)
	`closed_pnl`	Realized PNL from blockchain
% PNL	`pnl_percent`	Percentage PNL (unrealized_pnl / notional × 100)
Leverage ratio (REAL-TIME)	`leverage_realtime`	Position notional / real-time account value
Side (long/short)	`position_size`	Positive = LONG, Negative = SHORT
Whether ADL'd	All rows	Every row is an ADL'd position
Entry price	`entry_price`	Calculated from fills
ADL price	`adl_price`	Price at which ADL occurred
Account value (REAL-TIME)	`account_value_realtime`	Reconstructed at ADL moment
Total unrealized PNL	`total_unrealized_pnl`	All positions, real-time prices
Net equity (total equity)	`total_equity`	Cash + total unrealized PNL
Negative equity	`is_negative_equity`	TRUE if equity < 0
Position size	`position_size`	Size of position before ADL
Notional value	`adl_notional`	Position value (size × price)
Asset	`coin`	Ticker (BTC, ETH, SOL, etc.)
Timestamp	`time`	Milliseconds since epoch
User address	`user`	Anonymized address
Liquidated counterparty	`liquidated_user`	Who got liquidated

Real-Time Reconstruction Achievement

We processed 3.2M events to reconstruct exact account states (FULL 12-minute cascade):

Data processed:

Snapshot at block 758750000 (20:04:54 UTC) - 437,723 accounts
All fills with closedPnl (3.2M fills processed)
All funding events (from misc events)
All deposits/withdrawals (from ledger updates)
Real-time price tracking (last traded price per asset)

Reconstruction process:

Started with snapshot account values
Looped through all 3,239,706 events chronologically (FULL 12 minutes)
Updated account value: account_value += closedPnl for each fill
Processed funding events from misc events
Processed deposit/withdrawal events
Calculated unrealized PNL using real-time prices
Got account value at exact ADL moment

Results:

Real-time account values at every ADL moment
Accurate negative equity detection (302 accounts identified)
Precise leverage ratios (median 0.20x)
Insurance fund impact quantified ($23,191,104)
100% event coverage (34,983 ADL events)

Methodology: scripts/reconstruction/full_analysis_realtime.py

How to Access the Data

Option 1: Download from GitHub

# Clone repository
git clone https://github.com/ConejoCapital/HyperMultiAssetedADL.git
cd HyperMultiAssetedADL

# Open the REAL-TIME analysis file
# Contains 34,983 rows (one per ADL'd position with real-time data)
# This is 100% coverage of the FULL 12-minute cascade
open data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_detailed_analysis_REALTIME.csv

Option 2: Load in Python

import pandas as pd

# Load canonical real-time data
df = pd.read_csv('data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_detailed_analysis_REALTIME.csv')

# Quick sanity checks (should match README claims)
print('Events:', len(df))
print('Median leverage:', df['leverage_realtime'].median())
print('95th percentile leverage:', df['leverage_realtime'].quantile(0.95))
print('Profitable positions:', (df['pnl_percent'] > 0).sum())
print('Negative equity accounts:', df['is_negative_equity'].sum())
print('Total negative equity:', df.loc[df['is_negative_equity'], 'total_equity'].sum())

Run python3 scripts/verification/verify_all_findings.py for the full automated test suite.

Complete Column Reference

data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_detailed_analysis_REALTIME.csv columns:

user – User address (string)
coin – Asset ticker (string)
time – Timestamp in milliseconds (int)
adl_price – ADL execution price (float)
adl_size – Size ADL'd (float, negative = short)
adl_notional – Notional value (float, always positive)
closed_pnl – Realized PNL from blockchain (float)
position_size – Position size before ADL (float)
entry_price – Weighted-average entry price (float)
account_value_realtime – Account value reconstructed at ADL moment (float)
total_unrealized_pnl – Unrealized PNL across all positions at ADL time (float)
total_equity – Net equity: Cash + total unrealized PNL (float)
is_negative_equity – TRUE if total_equity < 0 (bool)
leverage_realtime – Position notional / real-time account value (float)
position_unrealized_pnl – Unrealized PNL for this position (float)
pnl_percent – Percentage PNL (float)
liquidated_user – Counterparty address if available (string)

Questions?

For Researchers

How to reproduce this analysis?: See docs/methodology/COMPLETE_METHODOLOGY.md ← START HERE
What data sources were used?: See docs/methodology/COMPLETE_METHODOLOGY.md (Section: Data Sources)
How to obtain clearinghouse data?: See docs/methodology/COMPLETE_METHODOLOGY.md (Section: Data Acquisition)
How to reconcile multiple data sources?: See docs/methodology/COMPLETE_METHODOLOGY.md (Section: Data Reconciliation)

For Findings

ADL prioritization?: See docs/findings/ADL_PRIORITIZATION_VERIFIED.md
Per-asset isolation?: See docs/findings/PER_ASSET_ISOLATION.md
Why separate chunks?: See docs/analysis/BATCH_PROCESSING_DISCOVERY.md
When does ADL activate?: See docs/analysis/CASCADE_TIMING_ANALYSIS.md
How ADL works: See docs/findings/ADL_MECHANISM_RESEARCH.md

For Data

Net volume analysis: See data/canonical/cash-only balances ADL event orderbook 2025-10-10/ADL_NET_VOLUME_FULL_12MIN.md
Processing scripts: scripts/data/extract_full_12min_adl.py, scripts/reconstruction/full_analysis_realtime.py, scripts/verification/verify_all_findings.py
Individual fills: data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_fills_full_12min_raw.csv (blockchain ADL events)

Clearinghouse Data Files (this repository)

Canonical Outputs (Real-Time):

data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_detailed_analysis_REALTIME.csv – 34,983 ADL events with real-time metrics (canonical dataset)
data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_by_user_REALTIME.csv – 19,337 user-level aggregations
data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_by_coin_REALTIME.csv – 162 asset-level aggregations
data/canonical/cash-only balances ADL event orderbook 2025-10-10/realtime_analysis_summary.json – Summary statistics
docs/reports/FINDINGS_VERIFICATION_REPORT.md – Comprehensive verification results
docs/reports/LEVERAGE_CORRECTION.md – Explanation of leverage statistics

Supporting Raw Data:

data/canonical/cash-only balances ADL event orderbook 2025-10-10/adl_fills_full_12min_raw.csv – Raw ADL fills (blockchain)
data/canonical/cash-only balances ADL event orderbook 2025-10-10/liquidations_full_12min.csv – Raw liquidation fills (for isolation tests)
data/raw/ADL_ORDERS_COMPLETE_LIST.csv – Complete list of all ADL orders
data/raw/high_leverage_outliers_analysis.csv – High leverage outlier analysis data

Scripts:

scripts/reconstruction/full_analysis_realtime.py – Real-time reconstruction pipeline
scripts/data/analyze_clearinghouse.py – Clearinghouse data loader
scripts/data/extract_full_12min_adl.py – ADL data extraction
scripts/verification/verify_all_findings.py – Automated verification suite
scripts/analysis/ – All analysis scripts (9 scripts)

Related Analysis

Position-level Analysis: ~/Desktop/ADL Clean/ (BTC & SOL positions)
GitHub Repository: https://github.com/ConejoCapital/HyperAnalyzeADL
Data Verification: ~/Desktop/SonarX Data Verification/

What Makes This Special

Largest Known ADL Event Analysis

$2.1 BILLION in 12 minutes
162 tickers affected
34,983 events processed
Data from blockchain events

Complete Dataset (Multiple Levels)

Event-Level Data:

Full 12-minute event (not sampled)
All assets (not just BTC/SOL)
Individual fill data included
Reproducible code provided

Account-Level Data (NEW - Clearinghouse):

437,723 accounts reconstructed in real-time
3,239,706 events processed chronologically (fills, funding, deposits)
34,983 ADL events with real-time leverage, PNL, entry price
302 negative-equity accounts (insurance impact quantified)
First analysis with complete protocol state and real-time account values

Academic Quality

Data from blockchain events (no heuristics)
Comprehensive documentation
Raw data available (event + clearinghouse)
Methodology fully explained
Zero speculation - all empirical

Analysis Date: November 13, 2025 (Canonical Replay)
Data Quality: Blockchain event data + real-time clearinghouse reconstruction
Time Coverage: FULL 12-minute event (21:15:00 - 21:27:00 UTC)
Scope: All 162 affected tickers + 437,723 accounts
Status: COMPLETE – Event + Account-Level Data – Ready for research and publication

Name		Name	Last commit message	Last commit date
Latest commit History 97 Commits
data		data
docs		docs
scripts		scripts
.gitignore		.gitignore
README.md		README.md

ConejoCapital/HyperMultiAssetedADL

Folders and files

Latest commit

History

Repository files navigation

HyperMultiAssetedADL

CANONICAL DATA FILE

📁 Repository Structure

Executive Summary

Key Findings

Top 5 ADL'd Assets

Analysis Scripts (Canonical Replay)

Canonical Results Snapshot (Nov 13, 2025)

COMPLETE METHODOLOGY: For Researchers

What's Inside:

Clearinghouse Data Access

Previously Unavailable -> Now Available

What We Now Have

Key Finding: ADL Targets Profit, Not Leverage

The Evidence (34,983 Real-Time ADL Events Analyzed - 100% Coverage)

Top 10 ADL'd Positions (By Size)

What This Means

INSURANCE FUND IMPACT: Quantifying the Underwater Accounts

Total Negative Net Equity: −$23,191,104.48

The Numbers

What This Means

Real-Time Reconstruction Achievement

Key Finding: Per-Asset Isolation - No Cross-Asset ADL Contagion

Key Evidence

Analysis Results

Important Distinction

TOTAL MARKET IMPACT (Liquidations + ADL)

NEW: ADL Mechanism Research - How It Really Works

1. Individual Event Analysis

2. CASCADE TIMING DISCOVERY

3. BATCH PROCESSING DISCOVERY

Key Discovery: ADL is a Direct Counterparty to Liquidations

What This Means

Why This Matters for Research

Major Insights

Market Concentration

ADL Rate

Asset Mix

Profitability

Key Files

Documentation

Data

Scripts

Top 20 ADL'd Tickers

Comparison: 2-Minute Sample vs Full Event

What is ADL Net Volume?

Methodology

Data Source

Calculations

Filters Applied

Data Quality

Usage

View Results

Rerun Analysis

Load in Python

For Academic Research

Suitable For

Key Research Questions This Dataset Answers

Citation

Position-Level Data: What's Available

Available in adl_detailed_analysis_REALTIME.csv (34,983 ADL'd positions - 100% Coverage)

Real-Time Reconstruction Achievement

How to Access the Data

Complete Column Reference

Questions?

For Researchers

For Findings

For Data

Clearinghouse Data Files (this repository)

Related Analysis

What Makes This Special

Largest Known ADL Event Analysis

Complete Dataset (Multiple Levels)

Academic Quality

About

Available in `adl_detailed_analysis_REALTIME.csv` (34,983 ADL'd positions - 100% Coverage)

Packages