# how-liquidation-cascades-happen-in-crypto-futures: findings (reviewed 2026-10-24) ## Line 401 (en, alt 16): CORRECTED "$18.0B at 07:00 UTC on 2026-08-03" matches the image (value pill, subtitle, AS OF stamp) and block 17. The flagged token `0` is from "07:00". The defect is what the alt says the number is. In Athenum's market-aggregator code at the time (before commit 74e51775a5, the all-contracts cut of 2026-10-22), `open_interest_total` summed **the main perpetual of each tracked venue, in US dollars**: Binance BTCUSDT, Bybit BTCUSDT, OKX BTC-USDT-SWAP, Deribit BTC-PERPETUAL, Bitget BTCUSDT, Hyperliquid and Lighter. So it is not "aggregated BTC futures open interest". **Bybit and Bitget were also counted on both sides.** Bybit's public hourly history shows openInterest = exactly 2 × singleOpenInterest in all 225 distinct hours re-queried (226 rows, one duplicate; Jul 25 00:00 to Aug 3 08:00). That is +$1.83B at 07:00 for Bybit alone. By the series' own venue split, Bybit + Bitget are $5.71B of $17.63B on Jul 29, so the double count adds roughly $3B, about a sixth (figure from the independent reviewer). The alt is rewritten to state this scope, the traced range and the gap: - Traced range: start $18.26B on 26 Jul about 11:00, low $17.04B late on 29 Jul (about 21:00 UTC), end $17.95B. - Gap: a straight interpolated stretch of 0.94 days from 31 Jul about 03:00 to 1 Aug about 02:00. - Axis: starts at $17.0B. **Unverifiable:** the $18.0B total. Deribit, Bitget, Hyperliquid and Lighter publish no open interest history. Public pieces that were re-queried: - Binance BTCUSDT: $6.89B (data.binance.vision) - Bybit, both sides: $3.67B - OKX, all BTC contracts: ≤ $2.57B - Sum: $13.13B. The remaining ~$4.9B is plausible for the four venues without history. The number is kept "as recorded". ## Other substantiated defects (fixed) - **Block 8**: "20x … liquidated after roughly a 5 percent adverse move, since 5 percent initial margin equals 20x". Also "10x … roughly a 10 percent move, a 50x … about 2 percent". These figures treat initial margin as the liquidation distance. The article itself says the position closes at maintenance margin (blocks 2, 6). On isolated margin the distance is 1/L − MMR. At 0.4% (Binance BTCUSDT tier 1, and the default in Athenum's leverage calculator) that gives 4.6%, 9.6% and 1.6% before fees (recheck.py §2). This matches the correction already made in risk-reward-ratio-crypto-leverage-liquidation. "On isolated margin" was added, because cross margin draws on the whole balance. - **Block 17** (caption): "within a $17.0B to $18.2B range" is wrong. The line opens at $18.26B, above the $18.2B gridline, so the range is about $17.0B to $18.3B ("about 7%": 7.1% of the low, 6.7% of the high). The caption also lacked the scope caveat and the gap. Rewritten. It names what is left out as other contracts on those venues (Binance's and Bybit's coin-margined and USDC perpetuals), dated futures and CME: Deribit's coin-margined BTC-PERPETUAL and the USDC perpetuals of Hyperliquid and Lighter are counted. - **Block 21**: "a position liquidated on one exchange moves the index that marks positions everywhere". This is wrong: - There is no shared index. Each venue builds its own index from spot prices (Binance's mark/index price documentation describes the index as a volume-weighted basket of major spot exchange prices; not re-fetched here). - A futures liquidation reaches an index only through arbitrage into spot. Rewritten. Also added, unlinked (en 21 and de 19 carried no link, and build-family rejects new ones): aggregated liquidation totals are a floor, because some venues, Binance above all, publish only part of their forced orders (Bybit has pushed all liquidations since Feb 2025), citing Athenum's floor post by name. - **Note** added, because the meaning of blocks 8, 17 and 21 changes. No historical cascade figures (e.g. 2021-05-19, 2025-10-10) appear in any locale, so there was nothing to date-check. Title, description, blocks 0–7, 9–15, 18–20 and 22–24 reviewed, no substantiated defect. Two points were left alone: - "Price tends to get pulled toward" bands (block 20) is a soft claim, not a number. - The cover is a product screenshot in a loading state. Its "10 exchanges" tab label versus "14 exchanges" in block 23 is a product-copy matter, not a reviewed figure. ## Locale structure - en has 25 elements, de 23, ko 15, tr 14, ru 16, vi 17. **No locale carries the chart (en 16) or its caption (en 17)**, so both map to null everywhere. The English alt stays English-only. - **de** is a full translation without the chart: en 8→8 and en 21→19. - **vi** is a near-full translation: en 8→7. The vi counterpart of en 21 is merged into vi 15 and lacks the index sentence, so null. - **ko 7, tr 6 and ru 7** are condensed or merged paragraphs (tr 6 and ru 7 also absorb en 9). They are mapped null, and the 5/10/2 percent figures are fixed with verbatim patches in `locale-patches.json`. Each patch occurs exactly once in blocks.json and in the visible page text. In the raw HTML it occurs twice, because the RSC payload repeats the text. - ko 13, tr 12 and ru 13 do not carry the index claim, so null. - Mapping is in `indexmap.json`. ## Figure The image needs relabelling (title and subtitle overstate scope; the gap is drawn as a line). See `figures.txt`. The identical PNG is also used in estimated-leverage-ratio-elr-crypto-explained (line 388), so fix both together.