The same bitcoin, quoted in Seoul and quoted offshore, and the gap between them — nine years of it. Above zero Koreans are paying up; below it Seoul trades at a discount.
+91.5 bp BTC · Korea premium on Oct-22 2026, the last day in this build, measured and published by askthetape.com.
Korea's exchanges are walled off from the rest of the world: money cannot move in and out freely, so the price of a coin in Seoul drifts away from the price everywhere else. We read Upbit's won price every hour, turn it into dollars at one public daily USD/KRW close and print the gap to Binance's tether price — so the gap also carries tether's own distance from a dollar. We show whether Upbit’s converted price stands above or below Binance’s. People call a positive gap the kimchi premium. The last two and a half years are our own hourly capture; before that we rebuilt each day from Upbit's and Binance's public daily closes, and the seam is re-measured every build: on the last 200 overlapping BTC days the two legs are IDENTICAL on 51.0 % of them — that zero is arithmetic, not proof, because both legs rebuild the daily point from the same closes — and on the 49.0 % that do differ the gap runs median 8.0 bp, p90 35.8 bp, worst 68.9 bp (worst month 2026-09: 1 of 10 days differ, median 68.9 bp). The marker is measured, not guessed: days whose hourly capture did not hold exactly 24 rows carry a residual 91.7 % of the time (12 days) against 46.3 % on the clean 24-hour days (188 days).
| leg | own hourly capture (`kimchi_premium`: Upbit REST close ÷ one daily USD/KRW close ÷ Binance BTCUSDT close) from 2024-04-12; before that a daily rebuild from Upbit /v1/candles/days, Binance klines 1d and USD/KRW from KRW=X |
|---|---|
| cadence | hourly capture, shown daily (high / mean / low / close); the high-mean-low view exists only for the captured era |
| since | 2017-09-25, the day Upbit opened — 2,406 public days spliced to 909 own days on BTC |
| policy | 🩸 the splice is judged on its RESIDUAL, not on a median. A median of zero would be a tautology here: both legs rebuild the daily point from the same closes, so where they agree they agree by arithmetic. What is left over is the number: on the last 200 overlapping BTC days 51.0 % carry no residual at all and 49.0 % do, and on those the gap runs median 8.0 bp, p90 35.8 bp, worst 68.9 bp — concentrated in 2026-09 (1 of 10 days, median 68.9 bp). The overlap is 868 own BTC days (1,735 coin-days counting ETH), counted on our own leg only; every row keeps both numbers and the source it came from. AVAX, NEAR, SOL and SUI are own capture only — no public rebuild was done for them, so their lines start in 2024, not 2017. · export: from 2024-04-12 (rows before that day are the exchange's own archive and stay here) |
| re-pricing | own hourly leg re-priced at the settled USD/KRW close of the same UTC day (the live poll froze the rate for up to six days between Apr-15 and Aug-03 2026). 🩸 the daily point is priced on Binance SPOT and the intraday range on the PERP, so they do not share a denominator: measured on this build, the point sits outside its own high-low band on 152 of 922 days (16.5 %), median breach 4.0 bp, p90 11.9 bp, worst 44.1 bp. The band is the perp's day, not the spot point's error bar. The seam between the public rebuild and our own leg is measured in «policy» above and in the plain text — it is not repeated here. |
| alt coins | AVAX/NEAR/SUI: the live rate column was wrong for the whole own era — re-priced from fx_daily, and the correction is measured, not asserted: median 116 / 118 / 116 bp, mean 156 / 157 / 157 bp, sign flips on 29 % / 31 % / 28 % of their days (measured on this build against the pre-re-pricing file) |
| fx | 🩸 there is only ONE dollar-won source behind this page and the page no longer pretends otherwise: the public file and our own fx_daily leg are the same KRW=X pool, and on the 927 days they overlap they agree to the bit on 99.8 % of them (worst gap 13.7 bp, on 2026-10-23). No second, independent rate was captured, so no source comparison is claimed. The premium is priced on our own fx_daily leg from 2023-04-16 and from 2026-10-08 on our own USD/KRW close at 24:00 UTC of the same day (the last hourly price before midnight), and that is the rate the «won itself» line shows. |
| hourly window | prod kimchi_premium, our own Upbit capture (hourly bucket) · since 2024-04-12 00:00Z by data, 22,125 hours with data inside a 22,198-hour span → 2026-10-23 21:00Z (in the browser on Tape and Desk the 1h window draws the newest 400 days of that and the max window every hour) · captured 2026-10-23T21:21:56Z · 🩸 the won fix is a DAILY one (fx_daily USDKRW) and a closed day can be revised when that fix is: measured 2026-09-15 23:16Z → 2026-09-16 00:20Z the whole of Sep-14 2026 moved by up to 8.05 bp with the print count per hour unchanged, so an hourly export of a closed day is not reproducible byte for byte the next day |
| this build | data to Oct-22 2026 |
| policy | no backfill · read it here · export: from 2024-04-12 (rows before that day are the exchange's own archive and stay here) |
The chart itself is drawn in the browser: turn JavaScript on to hover a day, pick a window and draw your line. The numbers above come from the last build, Oct-23 2026 21:24Z.
What the tape says today · what's new · all 111 metrics