One entry per figure or figure-cluster in the brief, reachable from the marker printed beside it in the text. Each says how the number came to exist: reported read off a named source, derived arithmetic we performed on published inputs, own calc computed by us from raw feeds saved in this repository, own sample computed from the 420-account Hyperliquid fills sample, estimate an analyst range with its reasoning stated, unverified reported but not confirmed to a primary source. Where a figure is calculated, the arithmetic is written out with its inputs rather than described.
≈ 1 : 1 initiating side; market makers on ~49% of all sides
MEMX's retail-trading dataset for the side-level split; Cboe for the 'roughly half' cross-check. 'Sides' counts both legs of every trade, so market makers at 49% of sides and retail at ~46% of sides are not additive with each other in the way a volume share would be.
Confidence: High
Sources: MEMX — Retail Trading Insights · John Lothian News — Cboe's Henry Schwartz, May 2026
$2.5-2.7B notional / $72M premium per 30 days; ≈1-2% of crypto options; Derive 85-90%
The 30-day notional and premium are DefiLlama's, read on 9 Sep 2026 (the ranking page and the API disagree, which is the $2.5B vs $2.7B spread). The 1-2% is ours: annualise the 30-day notional (x365/30 = $30.4-32.9B) and divide by the whole-market $1.8-2.5T from P1-06, giving 1.2-1.8%. Derive's 85-90% is $2.25-2.27B over the $2.5-2.7B category total; summing the per-protocol rows of the §4.3 table instead gives 84-85%.
Check: D-04a, D-06, D-09a, D-09b · Confidence: Med
Sources: DefiLlama — options notional volume · DefiLlama — options premium volume · DefiLlama — options overview API · Derive — public stats page · DefiLlama — Derive
~10-30M active crypto derivatives traders monthly; on-chain 2-5M lifetime addresses, 0.3-1M monthly
Ours, and the weakest headcount in the brief. No exchange - Binance, Bybit, OKX, Bitget, Deribit - publishes how many of its users trade derivatives, so the range is inferred from registered-user bases times observed active ratios (OKX disclosed ~4% active on a 60M base in 2024) and cross-checked against the on-chain counts that do exist (Hyperliquid 1.2M addresses, Lighter 188k, edgeX 191k). Treat as an order of magnitude.
Confidence: Low
Sources: OKX statistics (Business of Apps, third-party compilation) · Binance 323M users (CryptoPotato) · Bitget statistics (SQ Magazine, third-party compilation) · Hyperliquid stats feed — cumulative_new_users · BlockEden — Perp DEX wars 2026
1.2-1.7M addresses, ~45-50k daily actives (Q1-26); top 100 addresses = 81% of volume
The address and daily-active figures are ours, summed from Hyperliquid's own stats feeds saved in this repo: cumulative_new_users reaches 1,199,467 on 3 Apr 2026, and daily_unique_users averages 45,459 over 1 Jan-3 Apr 2026 (51,434 in March alone, peak 98,960 on 10 Oct 2025). The 1.73M is a third-party Mar-26 trading-address count on a wider definition. The 81% is Hyperliquid Hub's, not ours. The feed stops on 3 Apr 2026 - nothing after that date.
Check: C-13..C-19 · Reproduce: python3 verify_own_calcs.py (rows C-13 to C-19) · Confidence: Low
Sources: Hyperliquid stats feed — cumulative_new_users · Hyperliquid stats feed — daily_unique_users · This repo — hyperliquid_feeds/ (saved Hyperliquid stats-feed snapshots) · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds) · Hyperliquid Hub — top 100 addresses (via WEEX news) · TechFlow — 43,618 Hyperliquid addresses
Median 57 minutes; 50.7% within 1h, 88.9% within 24h, 1.8% open >7 days
Ours, from 80,501 completed position episodes reconstructed out of 686,363 fills for 420 Hyperliquid accounts (stratified by lifetime volume, all active in the last 30 days, median 138-day window). An episode runs from the fill that takes a coin's position off zero to the fill that returns it to zero or flips its sign; duration is close minus open. Positions opened before the fill window or still open are dropped, so multi-week holds are under-represented.
Reproduce: python3 sample_fills.py && python3 analyze_fills.py - fills_sample.json (~130 MB) is NOT shipped; re-fetch the same accounts from sample_addresses.json · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — sample_fills.py / analyze_fills.py · Hyperliquid leaderboard endpoint
~90-200k people whose job is trading; ~35-90k of them trade derivatives
Ours, built bottom-up from registration and filing counts because no regulator publishes a trader census: sell-side desks 30-60k, hedge funds 15-35k, buy-side dealing desks 10-25k, prop/market-making 10-25k, CTAs 10-25k, crypto-native 3-10k. Each band is anchored on a citable count (FINRA 639,723 registered reps, BLS 531,000 sales agents, SEC Form PF 9,940 hedge funds, NFA 1,164 CTAs, and whole-firm headcounts for the 11 largest prop firms) but the split of those populations into 'people whose job is to trade' is a judgement.
Confidence: Low
Sources: FINRA — 2026 Industry Snapshot · BLS — securities, commodities and financial services sales agents · SEC — Private Fund Statistics Q3 2025 · NFA — membership statistics · CFA Institute
Global 88.7B contracts (2025); US 15.2B (2025); US premium ≈$598B/month in Mar-2024 → ≈$7T/yr run-rate
Contract counts are FIA's and OCC's, read directly. The ≈$7T/yr is ours and is the weakest number in the row: FIA published US option premium traded of $598B for the single month of March 2024; $598B x 12 = $7.18T. That assumes March is a representative month, which it is not (March carries a quarterly expiry). The brief says 'volumes 25-50% higher now' rather than restating the run-rate for 2026, because no official annual premium series exists at all.
Check: D-03 · Confidence: Low on the premium figure
Sources: FIA — ETD Volume December 2025 · OCC — Annual 2025 volume report · FIA — Premium turnover in Indian options hits $150 billion
No annual volume series; interest-rate options $46.2T notional outstanding (end-2024); all OTC derivatives $846T (Jun-25)
ISDA and BIS, read directly. The row exists mainly to record what is NOT measurable: BIS and ISDA publish notional OUTSTANDING, which is a stock, not turnover, so it cannot be compared with any volume figure elsewhere in this brief. The $846T covers swaps, forwards and options together - not options alone.
Confidence: High
Sources: ISDA — Key trends in OTC derivatives, H2 2024 · BIS — OTC derivatives statistics, end-June 2025
~90-200k people; ~35-90k trade derivatives
Ours - see P1-26 for the build. What is citable is the anchor column, not the range: FINRA 639,723 registered reps (YE-2025), BLS 531,000 securities/commodities sales agents (2025), SEC Form PF 9,940 hedge funds across 1,834 advisers (Q3-25), NFA 1,164 CTAs and 36,859 associates (Jul-26). Every one of those counts a wider population than 'people whose job is to trade', so the range is a haircut on them, not a sum of them.
Confidence: Low
Sources: FINRA — 2026 Industry Snapshot · BLS — securities, commodities and financial services sales agents · SEC — Private Fund Statistics Q3 2025 · IOSCO — investment funds statistics · NFA — membership statistics · CFA Institute
The 11 largest employ ≈19-20k in total; the nine listed sum to ≈18.5-19k
Ours, as addition, and the inputs are weak on purpose: these are WHOLE-FIRM headcounts across all asset classes, mostly from Wikipedia and company about-pages rather than filings. Jane Street ~3,000-3,500 + SIG 3,500+ + Optiver 2,112 + IMC 2,070+ + Jump 2,000+ + DRW 2,000 + Citadel Securities 1,800 + HRT ~1,000 + Virtu 969 = 18,451-18,951. Only a fraction of those people are traders, which is why the 10-40k options-professional range in §4.4 is wider than this sum rather than derived from it.
Check: D-13, D-12 · Confidence: Low
Sources: Citadel Securities — Options · SEC options market-structure roundtable, Apr 2026 (Delta Strategy Group summary)
2024 $1.185T total / $743B options; 2025 $1.875T total / ≈$1.354T options; Jan-Jul-26 $666B total / ≈$493B options
Deribit's own monthly statistics page plus its 2024 year-end release. The options split within the total is Deribit's own breakdown (BTC $1.131T, ETH $212B, USDC-linear $10.9B for 2025). Company-published, unaudited.
Confidence: High
Sources: Deribit — Monthly Statistics · Deribit — Q4 and 2024 year-end volumes
2024 $53.3T · 2025 $64.9T (+22%) · 2026 YTD $27.2T over 8 months
The monthly and annual figures are The Block's 15-venue series, read directly. The '≈$41T run-rate' for 2026 is ours: $27.2T x 12/8 = $40.8T, a linear extrapolation that assumes the remaining four months look like the first eight - which the July print ($2.89T, the lowest since late 2023) argues against.
Check: D-23 · Confidence: High
Source: The Block — total futures volumes (15 CEXs)
≈$56-58T (2024) · $85.7T (2025, +47%) · H1-26 $35.1T (−16% YoY)
Coinglass' own annual and H1 reports. Two cross-checks in the cell are ours. First, $85.7T/365 = $234.8B/day, which does not match Coinglass' own '$264.5B/day' - that figure implies a 324-day window ($85.7T/$264.5B), so the two are not the same measurement. Second, derivatives at ≈76% of CEX activity implies ≈3.2x spot (76/24), which is why the brief tags the separately reported '9.6x spot' unverified: it is on a different, unreconciled basis.
Check: D-20, D-21, D-22 · Confidence: High
Sources: Coinglass — 2025 annual report · Coinglass — H1 2026 market report · CoinGecko — 2025 annual crypto report
3.4% (2024 avg) → ~10-13% (2025) → 11-13% (2026 YTD), peak 21.6% Nov-25
The Block's DEX-to-CEX futures ratio: 17 DEX protocols against 15 CEXs. The series was transcribed from The Block's dashboard into make_charts.py as hard-coded literals; the underlying pull was NOT saved, so the chart is traceable to its source but not regenerable from this repo. Chart A carries the same limitation.
Confidence: High
Sources: The Block — DEX-to-CEX futures trade volume · The Block — perpetual protocol trade volume
$14.7B on 9 Sep 2026 (DefiLlama); The Block $14.3B for 8 Sep; series max $15.9B (Aug-25)
DefiLlama and The Block, read directly. The series max is tagged unverified for a concrete reason: no OI series was pulled or saved into this repo, so nothing in the bundle can confirm or refute it. Every OI statement in the brief - series max, 2025 average, 'halved after 10 Oct', and the funding-transfer dollar figure - inherits that limitation.
Confidence: Med
Sources: DefiLlama — perps · The Block — Hyperliquid OI climbs to $14.3B (8 Sep 2026) · The Block — Hyperliquid open interest
$972M in 2025 (revenue $818M); $542M Jan-Aug 2026; gross revenue $457M (Q3-25) → $202M (Q2-26)
DefiLlama's Hyperliquid fees API, read directly. ~97% of fees going to HYPE buybacks is Hyperliquid's own published policy, not a measurement.
Confidence: High
Source: DefiLlama — Hyperliquid fees API
$188M TVL (from a $604M Sep-25 peak); lifetime PnL $138M; ≈44% earned in two liquidation events
Ours, computed from a vaultDetails API snapshot saved in this repo as hlp_vault.json (9 Sep 2026). TVL is the last accountValueHistory point ($187.9M) against its maximum ($603.9M on 3 Sep 2025), a 68.9% fall. Lifetime PnL is the last cumulative pnlHistory point ($137.98M). The 44% is the two largest fortnightly steps (+$41.44M for 1-15 Oct 2025 and +$18.82M for 21 Jan-4 Feb 2026) over that lifetime total. Note the granularity: pnlHistory is sampled roughly every 14 days, so these are fortnight buckets containing the events, not the events themselves.
Check: C-20, C-24, C-25, C-26, C-29, C-30, C-31 · Reproduce: python3 verify_own_calcs.py (rows C-20 to C-31) · Confidence: High
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — hyperliquid_feeds/hlp_vault.json (vaultDetails snapshot, 9 Sep 2026) · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds) · CoinGecko — Hyperliquid HLP vault analysis
291k (end-24) → 917k (end-25) → 1.2M (Apr-26); daily uniques 33k (2025), ~45k (1 Jan-3 Apr 2026), ≈51k in March
Ours, read straight off Hyperliquid's own cumulative_new_users and daily_unique_users feeds, saved in this repo: 291,419 at 31 Dec 2024, 917,192 at 31 Dec 2025, 1,199,467 at 3 Apr 2026; daily-unique means of 33,310 (2025), 45,459 (1 Jan-3 Apr 2026) and 51,434 (March 2026), peak 98,960 on 10 Oct 2025. The feed ENDS on 3 Apr 2026 - there is no later data in the bundle, and the liquidation feed's zero rows after that date are a dead feed, not real zero-liquidation days.
Check: C-13 to C-19 · Reproduce: python3 verify_own_calcs.py (rows C-13 to C-19) · Confidence: High
Sources: Hyperliquid stats feed — cumulative_new_users · Hyperliquid stats feed — daily_unique_users · This repo — hyperliquid_feeds/ (saved Hyperliquid stats-feed snapshots) · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds)
420 Hyperliquid accounts, 686,363 fills, 80,501 completed position episodes, median 138-day window
Ours, and every §5.2 figure labelled 'own sample' rests on it. Method: draw 420 accounts at random from Hyperliquid's leaderboard, stratified by lifetime volume (active retail $10M-$100M, mid $100M-$1B, whale/MM >$1B) and filtered to accounts with 30-day volume > 0; pull each account's last <=2,000 fills from the userFills endpoint (2,000 is the API's hard cap per account, so high-frequency accounts contribute a shorter window); reconstruct position episodes per coin - an episode opens on the fill that moves a coin's position off zero and closes on the fill that returns it to zero or flips its sign. Spot fills (@index and X/Y pairs) are excluded; HIP-3 builder-market fills are kept and counted separately. LIMITS, stated because they bound every number downstream: dormant and fully wiped accounts are absent from the leaderboard and therefore from the sample; accounts under ~$10M lifetime volume are excluded by construction; episodes open before the fill window or still open at the pull are dropped, which under-represents multi-week holds; the whale tier mixes market makers with directional whales. Exact for Hyperliquid, a proxy for CEX retail - who by Binance's own 2019 disclosure use more leverage, so these figures understate CEX retail's risk.
Reproduce: python3 sample_fills.py && python3 analyze_fills.py - NOT reproducible from this bundle: fills_sample.json (~130 MB) and leaderboard.json (37 MB, a live snapshot) are not shipped. sample_addresses.json records the drawn addresses so the same accounts can be re-fetched, but a leaderboard re-pull gives a different population even with the same seed. · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · Hyperliquid leaderboard endpoint · This repo — sample_fills.py / analyze_fills.py · Binance Futures — leverage and derivatives overview of 2019
All sampled: median 57 min, p25/p75/p90 = 8 min / 6.9 h / 26.7 h, 50.7% <1h, 88.9% <24h, 1.8% >7d, 3.7% liquidated, 47.2% positive
Ours, from the 80,501 episodes of P5-17. Median and percentiles are over episode DURATION (close minus open, in seconds). 'Ends in liquidation' counts only episodes where a fill's liquidation object names THIS account as liquidatedUser - counting either side of the trade would roughly double it. 'Positive PnL' is gross closedPnL; funding is not a fill and is therefore excluded from every PnL figure in this table.
Reproduce: python3 analyze_fills.py (input not shipped, see P5-17) · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — sample_fills.py / analyze_fills.py
2.5 h
Ours: the same 80,501 episodes sorted by duration and weighted by episode notional (the largest of open, close and running position value x price), taking the duration at which cumulative notional passes half the total. It is 4x the unweighted 57-minute median - bigger positions do live longer - and it is the figure §7 uses to size the addressable slice, because a claim product is sold against notional, not against episode count.
Reproduce: python3 analyze_fills.py (input not shipped) · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — sample_fills.py / analyze_fills.py
13% of the 807 positions that carry a liquidation price sit within 10% of it (entry basis); 9% on a mark basis; 6% of all 1,265 positions
Ours, and the basis needs stating precisely because two conventions give different answers and the brief previously quoted one of them against the wrong denominator. The 13% is the ENTRY-based distance |liqPx - entryPx| / entryPx over the 807 positions that carry a liquidation price - not over all 1,265, and not from the mark. Recomputed from the saved sample: 13.1%. On the MARK-based distance, which the audit (finding F024) prefers because clearinghouseState gives the mark directly as positionValue/size, the same statistic is 9.3% of positions with a liquidation price, or 5.9% of all 1,265. The 458 positions Hyperliquid reports no liquidation price for (36% of the sample) are excluded from both the numerator and the denominator. The sentence in §5.2 now names all three figures rather than the highest one.
Check: C-38, C-39, C-40, C-41 · Reproduce: python3 verify_own_calcs.py (rows C-38 to C-41) · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — leverage_sample.py and hyperliquid_feeds/leverage_sample.json · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds) · This repo — audit/AUDIT-001-market-landscape-audit.md
Lifetime $138M: ≈$49.7M 2024, $67.5M 2025, $19.7M 2026 YTD; +$41.4M 1-15 Oct 2025; +$18.8M around 31 Jan 2026
Ours, from the saved vaultDetails snapshot. Lifetime PnL is the final cumulative pnlHistory value ($137.98M). The per-year splits cut that cumulative series at the SAMPLE NEAREST each 1 January - 27 Dec 2023, 1 Jan 2025 and 7 Jan 2026, because pnlHistory is sampled roughly every 14 days and no sample falls exactly on a year end. 2024 = 50.77 − 1.10 = $49.67M; 2025 = 118.27 − 50.77 = $67.50M; 2026 YTD = 137.98 − 118.27 = $19.71M. The event figures are the two largest fortnightly steps, which is why they are dated as windows (1-15 Oct, 21 Jan-4 Feb) rather than as single days. Losses: −$4.7M for the 29 Oct-12 Nov 2025 bucket (which spans the POPCAT bad debt reported elsewhere as $4.9M) and −$4.0M for 5-12 Mar 2025 (the ETH whale).
Check: C-20 to C-31 · Reproduce: python3 verify_own_calcs.py (rows C-20 to C-31) · Confidence: High
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — hyperliquid_feeds/hlp_vault.json (vaultDetails snapshot, 9 Sep 2026) · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds) · CoinGecko — Hyperliquid HLP vault analysis · Kappalab — HLP vs LLP in the October 2025 crash · Arkham — Hyperliquid whale passes $4M loss to HLP · CoinDesk — Hyperliquid loses $4M after whale's $200M+ ETH trade
ENVY 16.5%; Hyperdash 13.5%; TechFlow ≈25%
Three independent samples with three different designs, all read directly: ENVY's 10,000 random wallets over 30 days (Nov-24), Hyperdash's 1,000 traders (Jun-25), TechFlow's ~43,000 addresses (May-26). The brief corrects one figure in place: the '170 traders >$10M profit and 1,589 >$1M' often attached to the Hyperdash study are analyst DeFi Mochi's leaderboard counts on 499k users, not the 1,000-trader sample.
Confidence: High
Sources: ENVY Protocol — I analyzed 10,000 Hyperliquid traders · Hyperdash via BeInCrypto — Hyperliquid trader profitability · TechFlow — 75% losing · TechFlow — 43,618 Hyperliquid addresses
35,529 accounts with ≥$10M lifetime volume; 40.7% profitable (upper bound); median −$7,510 / −24% ROI; top 1% = 57% of gains, top 10% = 93%
Ours, computed over the full Hyperliquid leaderboard snapshot on 9 Sep 2026, filtered to accounts with at least $10M lifetime volume. It is explicitly an UPPER BOUND and the brief says why twice: wiped accounts drop off the leaderboard entirely (survivorship bias) and HYPE appreciation flatters realised PnL. That is why §8 resolves the conflict with the 14-25% wallet studies in favour of the wallet studies for retail, and treats 40.7% as a ceiling. NOT REPRODUCIBLE: leaderboard.json (37 MB) is a live endpoint and was not saved, so this cannot be regenerated for 9 Sep 2026.
Reproduce: not reproducible - leaderboard.json is a live snapshot and is not shipped · Confidence: Med
Sources: Hyperliquid leaderboard endpoint · This repo — sample_fills.py / analyze_fills.py
47.2% of episodes positive; mean net PnL +$109; whales −$2,105 per episode
Ours, from the 80,501 episodes. 'Net' is closedPnL minus fees. Funding is NOT included, because funding payments are separate ledger entries rather than fills and the userFills endpoint does not return them - so the true net-of-everything figure is worse than shown, by roughly the funding cost in P5-38. The whale row being negative per episode while whales remain in business is the tell that market makers are in that stratum: they lose per trade and earn on size, spread and fee rebates.
Reproduce: python3 analyze_fills.py (input not shipped) · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — sample_fills.py / analyze_fills.py
BTC 24.1% (2024 avg) / 10.6% (2025) / 4.7% (2026 YTD); ETH 22.0/8.5/5.4; SOL 28.1/5.3/−0.7; HYPE 118.8/22.1/9.1; peaks to 5,145% APR
Ours, from hourly fundingHistory pulls per coin. Annualized = hourly rate x 8,760. 'Hours longs paid' is the share of hours with a positive rate. 'Paid by a 1x long' is the plain sum of hourly rates over the year, which is why it differs slightly from the average x 8,760 (24.2% vs 24.1% for BTC 2024 - the sum runs over actual hours, the average over a nominal year). Peaks are single hours, annualized, which is why they reach four digits. NOT REPRODUCIBLE from this bundle: the hl_funding_*.json inputs are not shipped; regenerate with hl_funding.py.
Check: D-37 · Reproduce: python3 hl_funding.py BTC ETH SOL DOGE HYPE TRUMP FARTCOIN && python3 analyze_funding.py · Confidence: Med
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — hl_funding.py / analyze_funding.py · Hyperliquid docs — funding
Interest component 0.01%/8h ≈11.6% APR; capped at 4%/hour (35,040% APR); +0.05%/8h ≈55%/yr
The mechanics are Hyperliquid's documented parameters. The APR restatements are ours: the interest component compounds, (1.0001)^1095 − 1 = 11.57%; the cap is simple, 4 x 24 x 365 = 35,040%; the rules of thumb are simple too, 0.05 x 3 x 365 = 54.75% and 0.10 x 3 x 365 = 109.5%. The OKX cross-check is ours from a saved pull: BTC-USDT swap averaged 4.43% annualized over 8 Jun-9 Sep 2026, consistent with Hyperliquid's 2026 BTC figure. Note the OKX pull covers only that 93-day window (OKX's retention limit), so it cross-checks nothing from 2024-25.
Check: D-32, D-33, D-34a, D-34b, C-42 · Reproduce: python3 verify_own_calcs.py (row C-42); python3 verify_derived.py (D-32 to D-34b) · Confidence: High
Sources: Hyperliquid docs — funding · OKX — funding-rate history API · This repo — okx_funding.py and hyperliquid_feeds/okx_funding.json · This repo — verify_own_calcs.py (recomputes each figure from the shipped feeds)
Hyperliquid $177M (2024), $768M (2025, 8 coins ≈81% of OI) → ≈$0.95B all coins; market-wide $10-31B est
This is the least solid number in §5.2 and the brief marks it Low. Method: daily open interest multiplied by the sum of that day's hourly funding rates, per coin, over the 8 coins that make up ≈81% of Hyperliquid's OI; scaling to all coins gives 0.768/0.81 = $0.95B. The market-wide $10-31B extrapolates the resulting 8-13% of average OI onto market-wide OI of $124-236B (0.08 x 124 = $9.9B, 0.13 x 236 = $30.7B, where the top end uses the 7 Oct record OI). NOT REPRODUCIBLE - and not by accident: no open-interest series was pulled or saved in this repo, so the OI leg of the multiplication cannot be checked from the bundle at all. No script here produces this figure. Treat it as an order of magnitude.
Check: D-29, D-30 · Reproduce: not reproducible - no OI series was saved; the rate leg comes from hl_funding.py, the OI leg from nothing in this bundle · Confidence: Low
Sources: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState) · This repo — hl_funding.py / analyze_funding.py · Coinglass — 2025 annual report · The Block — futures open interest by asset
Ranking is by demonstrated dollar cost, not by survey
Stated so the ranking is not mistaken for research it is not. No survey systematically ranks perp-trader complaints. The only behavioural data found - 'liquidation checks rose 32%', 'funding-rate re-checks climbed 35%' across 106k trade setups - is a vendor press release on The Block's press-release channel reporting usage of the vendor's own leverage calculator; self-reported, and tagged unverified in place. The ordering below it is ours, by how much money each complaint has demonstrably cost.
Confidence: Low
Source: Leverage.Trading — Sep-2025 leverage risk report (press release on The Block)
FTX: an '$8 billion hole', >1M users
Reported, with the brief's own counter-evidence attached: FTX is the commonly cited motive for self-custody perps, but the brief's own DEX-share series shows on-chain stayed under 6% until mid-2025 before reaching ~11-13% - so the narrative and the data disagree, and the brief says so.
Confidence: Med
Sources: Bankruptcy of FTX (Wikipedia) · The Block — DEX-to-CEX futures trade volume
Every story sourced to an X post, an on-chain tracker or a news write-up; thesis outcomes checked against daily price data
How each story was assembled and what its limits are. Position facts come from an X post, an on-chain tracker (Lookonchain, Arkham, Hyperdash) or a news write-up. The 'thesis outcome' line - what the price did afterwards - was checked independently against Yahoo Finance daily OHLC (BTC-USD, ETH-USD) and CoinGecko market_chart data (ETH, ZEC, HYPE, XPL, ENA, SOL), not taken from the same article that reported the liquidation. IMPORTANT LIMIT: those are INDEX prints. Exchange wicks are routinely deeper, so a liquidation price can sit outside the daily range shown. Anything that could not be confirmed carries an unverified tag in place.
Confidence: Med
Source: Hyperliquid info API (userFills, fundingHistory, vaultDetails, clearinghouseState)
$1.25B BTC long at 40x liquidated ~4.8% below entry; ~$37M realized loss on the final long; BTC 14-20% higher within months
Wynn's own X post plus CoinDesk, The Block and Presto Research; Lookonchain for the recap. The brief reconciles three different loss headlines rather than picking one: ~$37M realized on the final BTC long, against Presto's $85M and The Block's >$100M, which are the peak-equity drawdown and the principal-plus-peak-paper-profit measures respectively. Prices afterwards ($110,380 on 10 Jun, $123,092 on 14 Jul, $126,198 on 6 Oct 2025) are from the independent price check in P5-46.
Confidence: High
Source: This repo — prediction_markets.md (research notes with every source URL)
Four liquidation stories with price recoveries afterwards
Lookonchain feed items and per-trader coverage (CoinDesk, CCN, BeInCrypto for Qwatio; CryptoSlate and OurCryptoTalk for Tate). The Garrett Jin item rests on a single source and the brief says so in §9. Recovery prices come from the independent daily-price check, not from the liquidation coverage.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
−$25.8M (CCN) against $37M already reported by CoinDesk for the week to 7 Jul
The two cited sources conflict outright and neither can be reconciled from on-chain data available here, so the brief prints both and tags the figure rather than picking the larger or the more recent. The position sizes quoted alongside it (1,743 BTC + 33,743 ETH + 15M FARTCOIN, ≈$232M at entry for the two positions listed) come from the BeInCrypto/CCN write-ups.
Confidence: Low
Source: This repo — prediction_markets.md (research notes with every source URL)
Event slides where the price came back within days
The event dates and the recovery windows are from the daily-price check in P5-46, which is why this story exists separately: it is the one story in §5.4 whose content is primarily our own price verification rather than a report.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
~$472M ETH long liquidated below $2,500 on 31 Jan 2026, ~$250M loss
Single-sourced to an OurCryptoTalk write-up, which §9 flags as a single source. It is kept in the brief because it is the one case in §5.4 where the thesis was NOT vindicated afterwards - ETH kept falling - and dropping only the counter-example would bias the story set. The price moves around it (BTC $62,354 low to $71,681 within 24 hours; ETH $1,819 to $2,090) come from the independent daily-price check in P5-46 and are solid; the loss figure is not.
Confidence: Low
Source: This repo — prediction_markets.md (research notes with every source URL)
$536k of funding in five to six weeks (11 Oct-19 Nov 2025), $22M drawdown, $6M liquidated — then ZEC fell 70%
EmberCN's on-chain tracking plus crypto.news and Bitcoin.com coverage; the 70% fall afterwards is from the independent price check. This is the brief's cleanest illustration of its own thesis: the trader was right and the funding line and liquidation line still decided the outcome.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
$2.05M of funding paid on a BTC long up $4.4M; $2.275M collected in three days shorting 16 alts; $46M realized loss on HYPE
Lookonchain and Bitcoin.com/Cointelegraph coverage per wallet. B5 is in the brief specifically as the other side of the ledger - funding is a transfer, not a cost to the market, and someone collects it.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
$107M BTC long with $4.42M unrealized after $2.05M paid in funding; ZEC short −$21.9-25.7M
The position facts are EmberCN's on-chain tracking (7 Sep 2026) and are checkable on chain. What is tagged unverified is the ATTRIBUTION - the link to a Jin-connected entity - which rests on the tracker's labelling rather than on anything the brief could confirm. The trade is usable; the name attached to it is not.
Confidence: Med on the position, Low on the attribution
Source: This repo — prediction_markets.md (research notes with every source URL)
P5-55reportedC1-C6 — JELLY, ETH self-liquidation, XPL squeeze, HYPE hunt, ENA wick, POPCAT wall↑ back to the text
Six engineered or oracle-driven liquidation incidents
CoinDesk and The Block for JELLY, Arkham and CoinDesk for the ETH self-liquidation, The Block and PANews for XPL, Lookonchain for the HYPE hunt and POPCAT wall. The '1x hedge, account destroyed' line is a quote from a liquidated XPL hedger, not a summary.
Confidence: High
Sources: CoinDesk — Hyperliquid delists JELLYJELLY · Arkham — Hyperliquid whale passes $4M loss to HLP · CoinDesk — Hyperliquid loses $4M after whale's $200M+ ETH trade · The Block — Hyperliquid safeguards after XPL pre-market liquidations · Incrypted — Chen/Hayes on Hyperliquid
f(x), Derive, Index Coop, Toros, Squeeth, TLX TVL comparison
DefiLlama TVL for each protocol, read on 9 Sep 2026. NOT REPRODUCIBLE from this bundle - those DefiLlama pulls were not saved, which the repo README states. The qualitative point (liquidation-free leverage products exist and have not scaled) does not depend on the exact TVL figures; the exact figures should be re-pulled before use.
Reproduce: not reproducible - the DefiLlama TVL pulls for f(x), Derive, Index Coop, Toros, Squeeth and TLX were not saved · Confidence: Med
Sources: DefiLlama — f(x) Protocol API · f(x) Protocol · DefiLlama — options protocols (TVL) · DefiLlama — perps
≈$2.5B/30d of options against $370-600B/30d of perps — well under 1%
Ours, dividing two figures already established: DeFi options 30-day notional from P4-14 and perp DEX 30-day volume from P5-08. 2.5/370 = 0.68% and 2.5/600 = 0.42%. The point of the ratio is that on-chain options are not a small version of on-chain perps; they are a rounding error against them.
Confidence: High
Sources: DefiLlama — options notional volume · DefiLlama — perps · The Block — perpetual protocol trade volume
f(x) $122.1M (peak $271.3M); Derive V2 $158-159M; Index Coop $14.4M; Toros $7.7M; Squeeth $0.6M; TLX ≈$0
DefiLlama TVL per protocol on 9 Sep 2026. NOT REPRODUCIBLE from this bundle - those pulls were not saved, which the repo README also states. The peaks matter more than the current values here: every one of these products is 60-97% below its own peak, which is the finding.
Reproduce: not reproducible - the DefiLlama TVL pulls for f(x), Derive, Index Coop, Toros, Squeeth and TLX were not saved · Confidence: Med
Sources: DefiLlama — f(x) Protocol API · f(x) Protocol · DefiLlama — Derive · DefiLlama — options protocols (TVL) · DefiLlama — perps
$122M TVL is under 2% of Hyperliquid's average 2025 OI ($7.4B)
The division is ours (0.122/7.4 = 1.6%) but the denominator is tagged unverified for the same reason as everywhere else in the brief: no open-interest series was saved, so Hyperliquid's average 2025 OI of $7.4B cannot be checked from this bundle. The numerator is also not like-for-like - f(x)'s $122M is deposited collateral including the fxUSD/stability-pool side, not open notional - and the brief says so in place. Treat the ratio as an order of magnitude, not a measurement.
Confidence: Low
Sources: DefiLlama — f(x) Protocol API · The Block — Hyperliquid open interest · DefiLlama — perps
2025 12B+ contracts; Q2-26 13.6B, $156M revenue; ~1.15¢ revenue per contract
Robinhood's own results releases and Fortune's Q2 coverage. The per-contract revenue is ours: $156M / 13.6B = 1.15¢. Relevant because Robinhood routes to Kalshi, so its contract counts are not additive with Kalshi's - the same trade can appear in both.
Confidence: High
Sources: Robinhood — Q2 2026 results · Revenue Memo — how Kalshi makes money
Opinion $10.1B notional in 2025 (points-driven), then −90% to $0.75B in Aug-26; predict.fun $6.2B YTD; Limitless $0.86B YTD
DefiLlama per-protocol series. Opinion's collapse after its points program ended is the single clearest evidence in the brief that incentive-era volume is not organic demand - it is why §7 says to keep incentive-era DEX volume off the slides.
Confidence: High
Sources: DefiLlama — Opinion summary API · DefiLlama — predict.fun summary API · DefiLlama — Limitless summary API
£84B matched in 2025 (+~10%)
Single-sourced to a Covers trade-press article about the Betfair Predicts beta; no Flutter or Betfair statement was found. It is in the brief only as a legacy scale reference and is tagged unverified. Do not use it on a slide.
Confidence: Do not use
Source: Covers — Betfair Predicts beta (sole source for the £84B figure)
79% of capital held under 30 days, over 40% under 5 days; 21% patient capital ($73.6M)
Zozulynskyi's study of 1,843 markets and $348M. Critical framing carried in the brief: this measures OPEN INTEREST BY MARKET HORIZON, not how long traders hold. A market resolving in 5 days caps holding at 5 days; it does not show anyone chose to hold for 5 days. It is not comparable with the §5.2 perp holding-time distribution.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
Pew, 11,989 wallets, May-Jun 2026: median 46 trades in six weeks, average trade $6.50, median 10 active days of 42
Pew Research's wallet study, read directly. This is the closest thing to a holding-behaviour measure that exists for prediction markets and it is a tempo measure, not a duration one.
Confidence: High
Source: Pew Research — the typical Polymarket user
Kalshi ≈85-95% of volume resolves within days; Polymarket 5-minute BTC ≥54% of Aug-26 crypto event volume
Both ours. Kalshi: sports 75-88% plus 15-minute crypto ~12%. Polymarket: 1,822 five-minute windows with a median $79k each, out of a top-2,100-event Gamma API sample worth $288M on the dollar basis. The API's 2,100-event cap excludes the smallest windows, so ≥54% is a FLOOR - the true share is higher.
Confidence: Med
Sources: Polymarket — Gamma events API · DefiLlama — Kalshi summary API · Kalshi — homepage
Any L>1 on a 0/1 claim creates a guaranteed adverse-outcome shortfall; stress tests on 13,298 markets
Nechepurenko's paper, read directly: dynamic margin cut drawdowns only 5.1% and staged halts made terminal shortfalls worse. The sample is 13,298 markets that passed adequacy gates out of 61,087 ingested (21-27 Apr 2026) - the brief gives both numbers so the filtering is visible. The conclusion that a bounded scalar payoff is a more natural leverage substrate is the brief's inference from that result, not the paper's claim.
Confidence: High
Sources: This repo — prediction_markets.md (research notes with every source URL) · Crypto Briefing — Polymarket perpetual futures, 20x
12.6% of addresses control 81.4% of notional; 68 addresses (0.09%) = 28%
Nechepurenko, over 13.4M fills and 77,204 addresses with at least 5 records. The scope limits matter and the brief carries them: a 2.3-day block range (25-28 Apr 2026) on the LEGACY CTF exchange only, so negative-risk markets are absent entirely. It is a snapshot of a subset, not a lifetime measure.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
Retail takers lost $583.5M to professional makers; 63% of retail revenue from 6% of matched orders
Roosevelt Institute, over 400M+ trades and $32B of matched-trade volume on the DOLLAR basis - not notional, which matters when comparing with the $23.8B notional figures elsewhere. Kalshi disputes the study, saying it counts institutional market-maker flow as 'ordinary users' and casual app trades as 'professional'; the brief carries the dispute rather than resolving it.
Confidence: Med
Sources: Roosevelt Institute — Kalshi retail vs professional flow · Kalshi Institutional · Dune — prediction markets dashboard
$40M via within-market rebalancing and combinatorial arb; $1.12M via NegRisk basket arb; arb half-lives under a minute
Saguillo et al. and the other academic microstructure papers, read directly. The 5.6-6.3pp Polymarket-vs-Binance probability wedge is measured on a SINGLE Sep-2023 BTC contract over 214 hourly observations (Portnaya) - a small sample the brief flags in place rather than generalising.
Confidence: Med
Sources: This repo — prediction_markets.md (research notes with every source URL) · Polymarket — Gamma events API
821 manipulator-profile traders (1 in 300 of 243,000) extracted $8.2M from 5-minute BTC contracts; absent in 15-minute contracts
Dai, Jia & Yu. The 5-minute-versus-15-minute contrast is the single most useful finding in §6 for StrikeSimple, and §7 point 5 uses it directly: the settlement window, not the asset, decided whether the market was manipulable.
Confidence: High
Source: This repo — prediction_markets.md (research notes with every source URL)
Insider cases; academic screens classify 3.1% as skilled winners; Théo made $85M on the 2024 election
News coverage for the indictments and referrals, academic screens (Gomez-Cram et al.; Mitts & Ofir over 210,000+ wallet-market pairs) for the classification. The brief marks two citations as MISSING rather than supplying a plausible-looking one: the ~1,950 'insiders' flag and the LBS/Yale attribution in the Solidus row.
Confidence: Med
Source: This repo — prediction_markets.md (research notes with every source URL)
P7-02own calc + estimate2. 'No funding, no liquidation' has a dollar value↑ back to the text
BTC longs paid 10.6-24.2%/yr; ≈$0.95B on Hyperliquid in 2025; market-wide $10-31B est
The per-coin funding costs are ours and reproducible in principle (P5-35). The $0.95B and the $10-31B are ours and are NOT reproducible from this bundle - no open-interest series was saved (P5-38). The brief states the market-wide figure as a range with a Low confidence tag for that reason, and the caveat attached to it is the honest comparison: a prefunded claim replaces funding with upfront time value, so the like-for-like test is total cost of holding a correct view for N days, not 'free vs paid'.
Check: D-29, D-30 · Confidence: Low
Sources: This repo — hl_funding.py / analyze_funding.py · Coinglass — 2025 annual report · The Block — futures open interest by asset
P7-06derived6. On-chain liquidity concentrates in one venue per category↑ back to the text
Hyperliquid 62% of DEX perp OI; Derive 85-90% of DeFi options; Polymarket ≈75% of on-chain prediction dollar volume
All three shares are ours (P2-15). The Rysk and Derive Hyperliquid-collateral slices (≈$45M of Rysk's $52M TVL; ≈$52M of Derive's $158M) are ours too, from the per-protocol chain splits on DefiLlama. The caveat - that distribution rather than product decided every one of these categories - is the brief's reading, not a measurement.
Check: D-19, D-09a · Confidence: Med
Sources: DefiLlama — perps · Derive — public stats page · DefiLlama — Rysk · DefiLlama — Polymarket international summary API
The explicit gap list
What was looked for and not found, recorded so a reader can see the shape of the hole rather than infer it: OKX/Binance/Bybit/Bitget options volumes and user mixes; CME options-only ADV/OI for 2026; BIS FX/equity options notional; a 2025 market-wide liquidation total for 2024; The Block's monthly series behind its JavaScript charts (only current windows were verified); Kalshi trade-level data, paywalled behind Dune since 15 May 2026; Coalition Greenwich front-office headcounts; and any exchange's derivatives-user count. The DefiLlama monthly prediction series in §6.1 was transcribed on 9 Sep 2026 but the pull was not saved, so it is not reproducible from this bundle.
Confidence: High
Sources: Dune — prediction markets dashboard · BIS — OTC derivatives statistics, end-June 2025 · CME — Crypto Catch-Up Q2 2026 · The Block — total futures volumes (15 CEXs)