{ "generated_at": "2026-08-24T23:22:16+00:00", "note": "Every measurement Isobath has published, with its sample, method and stated limits. REFUTED entries are never removed. Each has a page at /findings/.html", "count": 19, "findings": [ { "id": "F-0001", "title": "Free collateral cannot be derived from position and margin data; naive arithmetic misclassifies one wallet in five", "status": "MEASURED", "observation": "withdrawable == accountValue - totalMarginUsed for 19% of wallets; 20% misclassified as able to defend", "sample": "74 Hyperliquid wallets holding open BTC positions, queried live 2026-08-20", "method": "compare venue-reported `withdrawable` against marginSummary arithmetic on the same response", "confidence": "single venue, single moment, wallets already known to hold positions; would be strengthened by repetition across days and by a second venue", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "research/withdrawable-is-not-derivable.md", "market_gap": "no provider checked (HyperTracker, Coinglass, 0xArchive, Tardis, QuickNode, Dwellir, PurrData) exposes this field", "supersedes": "none", "url": "/findings/F-0001.html" }, { "id": "F-0002", "title": "Order-book depth falls as price moves get larger, so a static-book cascade estimate is optimistic", "status": "MEASURED", "observation": "near-book depth falls to 0.8462 at a 5-minute horizon during the largest moves (p25 0.6573), against 1.0015 in quiet markets", "sample": "BTCUSDT, 1,324 days (2023-01-01 to 2026-08-19), 3,733,943 snapshots, 0 missing days; largest bucket n=313 across 160 distinct days", "method": "notional within +/-1% after a window divided by before, bucketed by absolute move, de-overlapped at the horizon", "confidence": "quiet-market control reads 1.000 to four decimals across 1,320 days, so the effect appears only where claimed; one venue, one asset", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "research/evaporation-result.md, market/evaporation_run.py, ~/genesis-evidence/bookdepth/evaporation.json", "market_gap": "no liquidation product models a non-static book", "supersedes": "none", "url": "/findings/F-0002.html" }, { "id": "F-0003", "title": "A stratified estimate puts the discovered wallet universe at 53.3% of Hyperliquid BTC open interest, not 20%", "status": "MEASURED", "observation": "estimated 53.3% full-universe coverage, 95% CI [40.9%, 70.8%]; this is not the coverage observed in a live scan", "sample": "31,349-wallet universe; stratified estimate with a random sample of 1,200 of the 25,954 newly discovered wallets", "method": "known stratum from the last deep scan plus mean-per-wallet extrapolation over the new stratum; bootstrapped interval", "confidence": "still a lower bound \u2014 the universe is activity-derived and the recording keeps discovering wallets", "first_recorded": "2026-08-20", "last_updated": "2026-08-24", "evidence": "research/coverage-measured-53-percent.md", "market_gap": "no provider publishes a coverage figure for position data", "supersedes": "F-0004", "url": "/findings/F-0003.html" }, { "id": "F-0004", "title": "The wallet universe is exhausted; more wallets of the same kind would add nothing", "status": "REFUTED", "observation": "claimed 20.24% observed coverage was near the ceiling; a later stratified estimate put the wider discovered universe at 53.3% [40.9%, 70.8%]", "sample": "the claim rested on a universe frozen after FOUR HOURS of recording, documented as 21 days", "method": "n/a \u2014 the error was a premise never measured, not a calculation", "confidence": "refuted because newly discovered wallets added substantial measured notional; 53.3% remains an estimate, not a directly observed scan", "first_recorded": "2026-08-19", "last_updated": "2026-08-24", "evidence": "research/DEFECT-universe-was-four-hours-not-21-days.md, research/coverage-measured-53-percent.md", "market_gap": "n/a", "supersedes": "none", "url": "/findings/F-0004.html" }, { "id": "F-0005", "title": "Daily-horizon directional tests on one instrument cannot detect realistic edges; the region is unpowerable, not empty", "status": "MEASURED", "observation": "detecting a 52% hit rate needs 4,900 independent days (13.4 years); 51% needs 19,600 (53.7 years)", "sample": "standard power arithmetic applied to the project's own recorded sample sizes", "method": "80% power, alpha .05 two-sided, on a proportion, with observations counted as INDEPENDENT rather than raw", "confidence": "arithmetic, not an estimate; the only judgement is what effect size counts as realistic", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "research/POWER-AUDIT-the-daily-lane-is-unpowerable.md", "market_gap": "n/a", "supersedes": "none", "url": "/findings/F-0005.html" }, { "id": "F-0006", "title": "Binance order-book behaviour under stress transfers to Hyperliquid", "status": "ASSUMED", "observation": "none \u2014 nothing has been measured on Hyperliquid", "sample": "none", "method": "n/a", "confidence": "untested and load-bearing; hl2 began recording Hyperliquid l2Book at nSigFigs=3 on 2026-08-20 to settle it", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "research/ASSUMPTION-binance-physics-may-not-transfer.md", "market_gap": "n/a", "supersedes": "none", "url": "/findings/F-0006.html" }, { "id": "F-0007", "title": "No provider sells per-wallet account state, so an archive of it is a durable moat", "status": "REFUTED", "observation": "HyperTracker sells it \u2014 complete market coverage, 15-20 minute snapshots, history from 2025-04-04, $179-1,999/month", "sample": "their own endpoint documentation, fetched", "method": "n/a \u2014 the claim was asserted from landing pages rather than docs", "confidence": "refuted from the vendor's own field list", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "research/VERDICT-the-collection-thesis-is-dead.md, research/CORRECTION-the-moat-is-thinner-than-i-said.md", "market_gap": "n/a", "supersedes": "none", "url": "/findings/F-0007.html" }, { "id": "F-0008", "title": "The position map is up to an hour stale relative to the order book, which manufactures phantom triggers", "status": "MEASURED", "observation": "clock gap of 3,055 s (51 min) between the latest l2Book snapshot and the latest position snapshot", "sample": "first end-to-end cascade run, 2026-08-20", "method": "compare observation timestamps of the two joined sources", "confidence": "structural, not incidental -- the fast position scan runs hourly while the book updates every 5.3 s", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "market/cascade_live.py", "market_gap": "no liquidation product discloses the age of the positions behind its map", "supersedes": "none", "url": "/findings/F-0008.html" }, { "id": "F-0009", "title": "Visible forced flow on Hyperliquid BTC is small relative to visible book depth", "status": "PRELIMINARY", "observation": "$63.6M total forced selling in the visible map against $222.5M of bids within 2.71% of mid; a $10.6M trigger moved price 0.069%", "sample": "one snapshot pair, 2026-08-20, fast-tier map at 24.2% coverage", "method": "cascade sweep of the observed ladder against the observed book, evaporation not applied", "confidence": "ONE observation, low coverage, static book, and a 51-minute clock gap (F-0008)", "first_recorded": "2026-08-20", "last_updated": "2026-08-20", "evidence": "market/cascade_live.py", "market_gap": "no product publishes forced flow as a ratio to available depth", "supersedes": "none", "url": "/findings/F-0009.html" }, { "id": "F-0010", "title": "In CASCADE-1, forced-liquidation episodes did not move price more than volatility-matched random minutes in the same hour", "status": "MEASURED", "observation": "15m mean +40.07 bps in the forced direction against a matched control of 44.52 bps; no horizon clears both benchmarks", "sample": "228 independent episodes >=$250k across 28 symbols (n=106 with price data), Binance market-wide, 2026-08-18 to 2026-08-20", "method": "CASCADE-1, contract frozen sha256 7dee22eed9cdaecb before any outcome; episodes not events; permutation null plus same-symbol same-hour control", "confidence": "K1 met at 228 episodes, so this is not an underpowered null; three days, one venue, and only two of four declared benchmark tiers were implemented", "first_recorded": "2026-08-21", "last_updated": "2026-08-24", "evidence": "research/cascade-1-result.md, market/cascade1.py", "market_gap": "no liquidation product publishes a volatility-matched control", "supersedes": "none", "url": "/findings/F-0010.html" }, { "id": "F-0011", "title": "The hourly map systematically under-represents the positions closest to liquidation", "status": "PRELIMINARY", "observation": "deep scan (5,395 wallets): median distance to liquidation 15.5%, 42.8% of positions within 10%. Fast scan (top 300 by notional) at the same period: median 31.6%, 23.0% within 10%.", "sample": "one deep snapshot of 1,327 BTC positions against three fast snapshots of 267-270, 2026-08-21", "method": "distance to liquidation as |liquidationPx - spot| / spot, per position, compared across scan tiers", "confidence": "one deep snapshot against three fast ones, same day, single venue, single asset. Needs several deep/fast pairs across different market conditions before it is more than PRELIMINARY.", "first_recorded": "2026-08-21", "last_updated": "2026-08-21", "evidence": "market/liqmap.py (tier definitions), ~/genesis-evidence/liqmap/snapshots-liq2.jsonl", "market_gap": "no liquidation product we surveyed states which population its map is drawn from, let alone how that population biases the result", "supersedes": "none", "url": "/findings/F-0011.html" }, { "id": "F-0012", "title": "The vulnerability figure Isobath publishes is very nearly a constant, so it cannot discriminate between clusters", "status": "MEASURED", "observation": "across 433 published clusters, cannot_defend_pct has median 100.0 and 71% sit at exactly 100%. Splitting at the median leaves 29% below and 0% above. thinly_defended_pct behaves the same: median 100.0, 27% below, 0% above.", "sample": "433 clusters published across 23 hourly and deep snapshots, Hyperliquid BTC, 2026-08-19 to 2026-08-21", "method": "contrast pre-condition P1 of market/CONTRACT-calibration.md, run before any outcome was examined", "confidence": "three days, one venue, one asset, and drawn largely from the fast scan tier whose selection bias is F-0011. A market regime with materially lower leverage could restore the spread. What is not in doubt is that no spread exists in the period recorded.", "first_recorded": "2026-08-21", "last_updated": "2026-08-21", "evidence": "market/CONTRACT-calibration.md \u00a73, product/calibration.py, ~/genesis-evidence/calibration/calls.jsonl", "market_gap": "no liquidation product publishes a defensibility figure at all, so none has had occasion to discover that theirs would not vary either", "supersedes": "none", "url": "/findings/F-0012.html" }, { "id": "F-0013", "title": "In four scan transitions, wallets holding free collateral disappeared more often; disappearance cannot be read as liquidation", "status": "PRELIMINARY", "observation": "across 4 deep-to-deep transitions, of wallets within 5% of liquidation: 62.7% of those WITH free collateral were absent at the next scan, against 45.1% of those with none. The sign is opposite to the hypothesis that free collateral aids survival.", "sample": "2,302 wallet-transitions within 5% (3,331 within 10%) across 5 deep scans of the frozen 5,395-wallet universe, Hyperliquid BTC, 2026-08-19 to 2026-08-21", "method": "track each wallet's BTC position between consecutive DEEP scans only; classify as absent, survived-improved or survived-worsened; split by whether withdrawable was above zero", "confidence": "only 4 transitions, 12.9h apart, so wallets are counted repeatedly and these are not independent observations. One venue, one asset, 54 hours. The sign is clear only within this sample; neither its persistence nor its magnitude is established.", "first_recorded": "2026-08-21", "last_updated": "2026-08-24", "evidence": "product/calibration.py, ~/genesis-evidence/liqmap/snapshots-liq2.jsonl", "market_gap": "no provider publishes wallet-level survival at all, because none archives per-wallet account state over time", "supersedes": "none", "url": "/findings/F-0013.html" }, { "id": "F-0014", "title": "Both venues' \u00b11% BTC book held roughly a quarter of a billion dollars; measured retail-size sweeps cost about 1 bp and typical clusters were small relative to standing depth", "status": "PRELIMINARY", "observation": "standing notional within \u00b11% of mid is $247.6M median on Binance futures and $225.6M median on Hyperliquid \u2014 within 9% of each other. At burst level, a same-side sweep worth 0.5\u20131% of that book costs the taker 0.8\u20131.1 bps of slippage. The median published liquidation cluster is 0.44% of the Hyperliquid book; the p90 cluster is 5.1%.", "sample": "221,778 aggressive bursts \u2265 $50,000 over 2 Binance futures days (2025-03-10/11), joined to bookDepth \u00b11%; 5,112 Hyperliquid l2Book snapshots 2026-08-20/21; 433 published clusters", "method": "contiguous same-side aggTrades within 200 ms grouped into one burst; cost = executed VWAP against the price at which the burst began, signed against the taker; conditioned on the PREVIOUS minute's range so that volatility already in progress cannot be credited to the burst", "confidence": "two Binance days and 34 hours of Hyperliquid book. The order of magnitude is not in doubt; the day-to-day variation is not established, and neither is behaviour in a stressed regime, where F-0002 measured near-book depth falling to 0.657", "first_recorded": "2026-08-22", "last_updated": "2026-08-24", "evidence": "market/impact2.py, market/impact.py, market/CONTRACT-impact.md \u00a74 P1/P2, ~/genesis-evidence/hl2/btc-l2book.jsonl", "market_gap": "liquidation heatmaps publish cluster notional without ever stating it as a fraction of the book standing in front of it, which is the only form in which the number means anything", "supersedes": "none", "url": "/findings/F-0014.html" }, { "id": "F-0015", "title": "How a liquidation price responds to added margin cannot be derived from Hyperliquid's published state, so the \"add $500 and your liquidation moves to X\" feature cannot be built from arithmetic", "status": "PRELIMINARY", "observation": "the documented formula reproduces the venue's own liquidationPx to within 10 bps for only 56.4% of cross positions (median error 4.79 bps, n=1,019); inverting it position-by-position within a single account recovers margin_available values that disagree by 3.49% at the median and 73.7% at p90, so only 28.2% of accounts are internally consistent within 1%", "sample": "90 Hyperliquid accounts fetched live 2026-08-22 (1,474 positions, 1,019 cross); 39 accounts holding 2+ cross positions; 1,923 wallet-transitions across 6 deep scans", "method": "liq_px = entry - side*margin_available/|szi|/(1 - l*side) with l = 1/(2*maxLeverage), margin_available = crossMarginSummary.accountValue - crossMaintenanceMarginUsed; then the same formula inverted per position and compared within each account; then the implied derivative compared against observed liquidationPx movement in wallets whose position size was unchanged", "confidence": "one venue, one afternoon, 90 accounts. The three checks disagree with the formula in three different ways, which is stronger than any one of them; but the third conflates unrealised PnL with deposited collateral and cannot yet isolate a real deposit", "first_recorded": "2026-08-22", "last_updated": "2026-08-22", "evidence": "market/liqcalc.py, ~/genesis-evidence/liqmap/snapshots-liq2.jsonl", "market_gap": "no provider offers a margin-response figure at all, and this measurement suggests why -- the obvious arithmetic is wrong in a way that only shows up in the tail", "supersedes": "none", "url": "/findings/F-0015.html" }, { "id": "F-0016", "title": "Across six years, none of nine public market-state variables separates forward trading conditions once trailing realised range is held constant", "status": "PRELIMINARY", "observation": "band-controlled lift for all nine candidates sits between 0.914 and 1.129 in raw form, and z-scoring each against its own 30-day history collapses every one toward 1.000 (1.129 to 1.039, 1.126 to 1.052, 0.914 to 0.964). The raw long/short lift reverses sign by year: 1.170, 0.990, 0.948, 0.869, 1.164, 1.232, 1.178 for 2020 through 2026", "sample": "66,397 hours of BTCUSDT, ~51,000 usable per candidate; Binance 5-minute metrics from 2020-09 joined to 1-minute klines", "method": "outcome is realised range over the following 24h; control is realised range over the previous 24h, banded; inside each band the candidate is split at its own median and the ratio of median forward ranges taken, then averaged across bands by weight", "confidence": "24h windows stepped hourly overlap ~24-fold, so the effective sample is nearer 2,000 than 51,000 and a residual lift of 4-5% is far inside the noise. One venue, one asset, one horizon, one outcome family. A screen, not a pre-registered test \u2014 its authority is only to eliminate", "first_recorded": "2026-08-22", "last_updated": "2026-08-22", "evidence": "market/screen.py", "market_gap": "none. Every variable here is public and anyone could run this; that it has apparently not been run is not the same as it being ours", "supersedes": "none", "url": "/findings/F-0016.html" }, { "id": "F-0017", "title": "In a uniform sample of discovered BTC-trading wallets, ten wallets held 57% of sampled leveraged exposure and the median position was $1,308", "status": "PRELIMINARY", "observation": "of 8,000 wallets drawn uniformly at random from 51,765 discovered, 2,249 held a BTC position totalling $222.9M. The top 1 wallet holds 14.0%, the top 10 hold 56.6%, the top 100 hold 87.6%. Median position $1,308 against a mean of $99,108 \u2014 a mean 76 times the median", "sample": "WIDE-1 first scan, 8,000 wallets, Hyperliquid BTC, 2026-08-22", "method": "uniform random sample without replacement, seed 20260822 recorded before drawing, from every distinct wallet seen in the BTC trade recording; positions read from the venue's own clearinghouse state", "confidence": "one scan. Concentration through time is the question worth asking and this is a single point on it. The draw is uniform over wallets that TRADED BTC during the recording window, which is not the same as uniform over all Hyperliquid accounts", "first_recorded": "2026-08-22", "last_updated": "2026-08-24", "evidence": "market/wide.py, ~/genesis-evidence/liqmap/snapshots-wide.jsonl", "market_gap": "every provider reports aggregate open interest. None surveyed reports concentration over a disclosed wallet-sampling frame", "supersedes": "none", "url": "/findings/F-0017.html" }, { "id": "F-0018", "title": "The hourly scan reproduces coverage and forced notional to within a point, but sees only 28% of positions near liquidation where the full scan sees 47%", "status": "PRELIMINARY", "observation": "across 7 deep scans paired with the fast scan nearest in time, coverage agrees to within 1.6 points (36.8% vs 36.2% on the closest pair) and forced sell-side notional within \u00b112%; but median distance to liquidation is 11.6% in the deep tier against 26.5% in the fast tier, and the share of positions sitting within 10% of liquidation is 47.4% deep against 27.8% fast", "sample": "7 deep/fast snapshot pairs within one hour of each other, 10,585 deep positions and 1,896 fast positions, Hyperliquid BTC, 2026-08-19 to 2026-08-22", "method": "pair each deep snapshot with the fast snapshot closest in time; compare the published aggregates directly, and compare the full distance-to-liquidation distribution of the positions each tier actually holds", "confidence": "three days, one venue, one asset, 7 pairs. The direction is unambiguous and consistent across every pair; the exact ratio is not established. The earliest pair sat at 20% coverage while the archive was still filling and its forced-notional gap (71.3M vs 33.6M) is far larger than any later pair's", "first_recorded": "2026-08-22", "last_updated": "2026-08-22", "evidence": "~/genesis-evidence/liqmap/snapshots-liq2.jsonl, research/QUEUE.md Q7", "market_gap": "no provider states which of its published figures survive its own sampling and which do not", "supersedes": "none", "url": "/findings/F-0018.html" }, { "id": "F-0019", "title": "Holding trade size against the standing book fixed, a book thinner than three-quarters of its own daily normal costs about 20% more \u2014 so the ratio absorbs most but not all of the stress regime", "status": "PRELIMINARY", "observation": "across 985,950 aggressive bursts, median cost at a burst/depth ratio of 0.01-0.02 is 1.88 bps when the book is below 0.75x its day's median, against 1.62, 1.54 and 1.62 bps in the three deeper regimes. The pattern repeats at every ratio band: 0.98 vs 0.80-0.83 at 0.005+, 3.24 vs 2.58-2.74 at 0.02+. Only the thinnest band separates; the other three are indistinguishable", "sample": "985,950 same-side aggressive bursts of $50,000 or more, 22 days sampled every 60 days across Binance futures BTCUSDT from 2023-01-01 to 2026-06-14, joined to bookDepth within \u00b11%", "method": "contiguous same-side aggTrades within 200 ms grouped into one burst; cost is executed VWAP against the price at which the burst began, signed against the taker; depth regime is the standing \u00b11% notional divided by the same day's median, so the split is \"thin for this market\" rather than \"thin for 2023\"", "confidence": "the daily-median regime is not the same stress F-0002 measured, which was depth AFTER a large move relative to before \u2014 a sharper and rarer condition this test does not isolate. Bursts within a day are not independent. The high-ratio cells above 0.05 remain empty, so nothing is established about large trades", "first_recorded": "2026-08-22", "last_updated": "2026-08-22", "evidence": "market/impact2.py run_stress, market/CONTRACT-impact.md K4, research/QUEUE.md Q6", "market_gap": "every execution model assumes a static book; none states whether its own estimate survives the regime where it would be used", "supersedes": "none", "url": "/findings/F-0019.html" } ] }