# bitcoin-24-hour-change-baseline: review findings (2026-10-24) I checked the article against five sources: - the published charts in `images/` and the manuscript `originals/BLOG_24h-change-baseline_0922.txt`; - a full new download of Binance's public BTCUSDT perpetual hourly candles (8,784 candles, opening 2025-09-21 08:00 to 2026-09-22 07:00), saved in `requery/`; - the current public tickers of Bybit, OKX, Bitget, Gate and Binance; - Bybit's public open-interest history for 2026-09-21 and 2026-09-22; - the repo's market-aggregator code, read only. To reproduce: `python3 -I recheck.py`. Its output is saved in `requery/recheck-output.txt`. ## Checklist lines 92 (en), 93 (ko), 94 (tr), 95 (ru), 96 (vi), 97 (de): CORRECTED All six rows contain the same four alt texts joined together (elements 14, 20, 25 and 33). In every non-English locale these alts are still in English, so they need translating from the reviewed English alts. - **Alt 14 (gap by hour): VERIFIED by the new download.** - The medians 1.298, 0.848 and 0.168 match exactly. - The 90th percentile, 3.731 at 00:00 and 0.640 at 23:00, matches when computed by nearest rank. - The middle-half band, 0.527–2.255 at 00:00 and 0.075–0.354 at 23:00, matches the (n+1)p quartile method. - With linear interpolation the 90th percentile would be 3.684 and 0.625. The gap comes from the method, not from an error. - Every hour bucket holds 365 observations, and the image shows these values. The alt is unchanged. - **Alt 20 (opposite signs by hour): VERIFIED by the new download.** - 01:00 is 186 of 365 days (51.0 %), 12:00 is 108 of 365 (29.6 %) and 23:00 is 18 of 365 (4.9 %). - The 00:00 bucket is empty, and the bars match the image. The alt is unchanged. - **Alt 25 (spread across five venues): PARTIAL.** - 0.026, 1.292 and 0.166 come from 28 live ticker readings that cannot be fetched again after the fact. - The chart is consistent with them: the spread peaks at about 1.29 near 08:41 and bottoms at about 0.03. The five trailing figures at 08:39:25 span 1.129, which matches the article's table. - The alt now says "as captured". - **Alt 33 (open interest): CORRECTED.** - The published numbers add up among themselves: the venues sum to 24.67, the shares to 99.9, and the venue changes combine to +6.96 and −0.80. - The basis is wrong. Before commit 74e51775a5 (2026-10-22), the market aggregator used Bybit's `openInterestValue` field and Bitget's `size` field. That code had not changed since 2026-03-25, so it was the version running on 2026-09-22. The commit's own documentation says both fields count both the long and the short side of every contract. - Bybit's v5 tickers documentation labels `openInterest` and `openInterestValue` "(both sides)" and `singleOpenInterest` "(single side)". - Independent check: Bybit's public ticker today returns `openInterest` 64,975.35 and `singleOpenInterest` 32,487.675, a ratio of exactly 2.0. - Bybit's public history at 2026-09-22 07:00 shows 60,151 BTC × 85,340 = $5.13B. That matches the published $5.11B (an hourly average), so the article used the two-sided figure. - For Bitget, the two-sided `size` rests on its "Get Open Interest" documentation as quoted in the repo. I could not load that page myself. - Binance, OKX, Hyperliquid and Deribit count one side. - Counting every venue on one side gives a total of about **$20.63B**, with Bybit and Bitget halved: | Venue | One-sided | Share | |---|---|---| | Binance | $9.39B | 45.5 % | | Bybit | $2.56B | 12.4 % | | OKX | $2.60B | 12.6 % | | Bitget | $1.49B | 7.2 % | | Hyperliquid | $3.76B | 18.2 % | | Deribit | $0.84B | 4.1 % | - The aggregate becomes **about +6.2 % against 24 hours earlier and about −0.6 % against 00:00 UTC**, instead of +6.96 and −0.81. Rounding each venue's published figures independently to the edge of their last digit gives +6.218 to +6.233 and −0.640 to −0.624, so +6.22 to +6.23 and −0.64 to −0.62. - The two baselines end up about 6.9 points apart, not 7.77. - Each venue's own percentage change is unaffected, because halving cancels out in a ratio. "Four of the six flip sign" still holds. ## Other body numbers checked - **Block 1 (reproduction note):** 0.662510, 3,251 and 1,427, 2,240 opposite signs (25.5708 %), 367 zero readings and 8,393 readings giving 26.6889 % all match exactly. - **Block 13:** - The middle half runs 0.261–1.498, and 172 readings (2.0 %) exceed 5 points. - The maximum is 14.041 at 2026-02-06 00:00, where the trailing figure reads −14.04 %. - No reference price is missing. - **Block 16:** the four small rises in the median, at 05:00, 07:00, 08:00 and 10:00, are confirmed. - **Block 22:** the two zero readings outside midnight fall at 07:00 and 13:00. The 33.8–40.0 % range for 06:00–09:00 matches. - **Blocks 10 and 42:** - The 3.677 and 4.58 point differences are confirmed. - The step-1 example works out to 83,644.49, which is 21 cents below Binance's 83,644.70. - Bitget's and OKX's implied last prices agree across their two baselines. - **What the exchange fields mean today:** consistent with the article. - Bitget still returns `openUtc` and `changeUtc24h`. - OKX returns `sodUtc0` and `sodUtc8`. - Gate's futures ticker has `change_percentage` and `change_price` but no since-midnight field. - Binance's window is about 86,400 seconds plus a few dozen. - I could not check the claim that Bitget's ticker documentation does not define its UTC fields, because the page would not load. - **Block 37:** the figures quoted from the linked realized-volatility article (6.42, 6.14 and 44 %) match that article's copy as reviewed on 2026-10-24. ## Other confirmed defects (English indices) - **Block 2 (reproduction note):** added two facts. The CSV uses linear-interpolation quantiles, so at 00:00 it reads 0.538–2.254 and 3.684 where the text reads 0.527–2.255 and 3.731; the medians are identical. The open interest example was corrected on 2026-10-24, with its per-venue changes as recorded and its totals, shares and aggregate recomputed. Block 2 has no links in any locale. - **Blocks 5, 32, 34, 49 and 50:** the open-interest totals, shares and aggregate changes rest on the mix of one-sided and two-sided counts. These blocks are replaced with the one-sided figures, each marked "about". Block 49's limitation now says that Bybit and Bitget were halved, and that the feed's own total field was built on the mixed count. - **Block 51:** "the 34 free calculators" is out of date. The site has 37 free tools: there are 37 tool routes under `apps/marketing/app/[locale]/tools/`, matching sibling corrections. Changed to "the 37 free tools", and the call-to-action link is kept. - Not claimed: the hourly open-interest rows are averages within each hour, which the article already says. Before the fix, a failed read from a venue counted as zero. I cannot tell whether that affected these three hours, so I did not claim it. ## Figures `figures.txt` gives the full data for redrawing the open-interest chart (`athenum-two-panel-chart-of-bitcoin-perpetual.png`) under a new filename, covering both panels, the footer, the title, the subtitle, the axis labels and the colours. Its upper panel and footer change. Its lower panel values are correct. The other four images, including the cover, are correct. ## Locale structure All six locales have 52 elements in the same tag order. The pictures are at 14, 20, 25 and 33, and there is one table. The referenced indices 2, 5, 32, 34, 49, 50 and 51 point to the same paragraphs in every locale, so no `indexmap.json` is needed.