▲ Work in progress. This site is built and researched in the open: pages, numbers and findings may change or be wrong, and the record has gaps. Not advice. More
The Lab
Ideas traders use, put to the record
← the Lab · notes · dated note
#43 Bitcoin after the CPI flush: how the levels did, a day later
A follow-up to "Bitcoin after the CPI flush: where the liquidations sit" (14 Oct 2026, odds as of 14:33 UTC). Scored at 14:33 UTC on 15 Oct, one day on. Dated and frozen. Public prices only. Liquidation levels are estimates from leverage and entry, not real liquidation orders.
Last checked: 16 October 2026.
The first note gave each liquidation level a chance of being touched within 1 hour to 1 week. The way this one is scored was fixed before the day closed. The 1-week window stays open until 14:33 UTC on 21 October.
What the price did
From 81,961 at 14:33 UTC on 14 October:
- The highest it went in the day: 82,071 (+0.1%). It never got further up.
- The lowest: 79,750 (−2.7%), at 13:55 UTC on 15 October. That was after the next day's US PPI release, at 12:30 UTC, which also came in hot. Before PPI, the day's low was 80,167 (−2.2%), at 02:25 UTC.
So no level above the price was touched, and six below it were.
The levels nearest the price
| level | who gets liquidated | chance in 1 day | touched? |
|---|---|---|---|
| 83,220 | 100x shorts, opened into the dump | 28% | no |
| 83,078 | 50x shorts, opened at the low | 32% | no |
| 82,259 | 100x shorts, opened at the low | 76% | no, missed by 0.2% |
| price at 14:33: 81,961 | |||
| 81,564 | 50x longs, bought in the dump | 68% | yes, 15:58 UTC (within 2 h) |
| 81,441 | 100x longs, bought at the low | 68% | yes, 22:42 UTC (within 10 h) |
| 80,903 | 25x longs, opened before CPI | 32% | yes, 02:24 UTC |
| 80,622 | 50x longs, bought at the low | 25% | yes, 02:28 UTC |
| 80,064 | 20x longs, opened before CPI | 14% | yes, 13:31 UTC, after PPI |
| 79,907 | 25x longs, opened into the dump | 13% | yes, 13:53 UTC, after PPI |
| 79,079 | 20x longs, opened into the dump | 7% | no |
The maths, step by step
1. Expected against actual. Adding up the chances gives how many levels the note expected to be touched. In one day it expected 4.2, and 6 were.
| window | expected | touched |
|---|---|---|
| 1 hour | 0.4 | 0 |
| 2 hours | 0.8 | 1 |
| 4 hours | 1.3 | 1 |
| 10 hours | 2.5 | 2 |
| 1 day | 4.2 | 6 |
The first 10 hours ran close to what was expected. The day ended above it because PPI pushed the price down to two more levels in its last hour.
2. How close the chances were. Take each level's chance and what happened (1 if touched, 0 if not). Square the gap and average it: 0 is perfect. This is the Brier score (Glenn Brier, 1950), made for weather forecasts. For one day it was 0.14 for our own curve and 0.13 for the textbook formula, the one from Brownian motion. Neither did better.
3. Why this is about one result, not 26. The levels on one side are nested. Reaching 80,064 means passing 81,564 first. So the whole day comes down to two numbers: how high the price went (+0.1%) and how low (−2.7%). That is one result. One day can't show whether the odds are right.
Could more entries have been found?
We took every turn on the 5-minute, 15-minute and 1-hour charts, at the moment it could first be known (a few candles later), not at its exact price. That gave 53 entries. Each was followed for 4 hours, against an entry at every 5-minute close in the same hours.
- Longs at turns (27): a median best move of +0.32% and worst of −0.57%. Random entries: +0.32% and −0.60%. No better than random.
- Shorts at turns (26): a median best move of +0.84% and worst of −0.34%. Random shorts: +0.60% and −0.32%. A little better, on a day that fell. One day is not an edge.
- Liquidation levels are not targets. Hitting them is not a plan to trade toward them.
Limits
- One day, one coin; the levels assume isolated margin, a 0.5% maintenance margin and no fees.
- PPI landed inside the window, so the day's low is partly PPI's move, not CPI's.
- Prices from Bybit's 1-minute candles, with Binance's as a check. Both gave the same counts.
- A fix made at scoring time: our candle reader dropped the first minute of a long request, which here was the 12:30 CPI candle itself. It now asks from one minute earlier. No odds, levels or rules changed.
Update, 16 October 2026: four corrections
Added after this page was published. The text above is left as it was.
- The pre-PPI low's time was wrong. It was 80,167 at 02:28 UTC, not 02:25. The 02:25 came from a 5-minute candle's label. That is why the table has 80,622 first touched at 02:28: it was touched on the way down to the low, in the same minute.
- Bachelier did state the doubling rule. Histories of his work say his 1900 thesis already worked out the chance that a price is reached within a given time, the doubling included. So the credit belongs to him, as our CPI and PPI notes give it, not to no one.
- "Neither did better" means within the noise. The textbook formula scored a little lower (0.13 against 0.14), but with one result to judge (step 3), that gap means nothing.
- Two lows for 15 October. This page gives 79,750 at 13:55 UTC, Bybit's price. Our PPI note gives 79,732 in the 13:45 15-minute candle, Binance's price. Both are right for their exchange.
Update, 27 October 2026: where the idea comes from
Added after this page was published. The text above is left as it was.
- Scoring a forecast after the fact comes from weather forecasting. Glenn Brier's 1950 score is worked through in step 2.
- Calibration is the other half: of the days a forecaster said 70%, did it happen about 7 times in 10? Allan Murphy and Robert Winkler studied it for weather forecasts in 1977.
- For example: this note gave seven levels between 25% and 76% in a day. Their chances add up to 3.3, so if the odds were right, about 3 of the 7 would be touched. Four were.
- The catch: one day is one result. The levels on each side are nested, so the whole day comes down to its high and its low (step 3). Calibration needs many separate forecasts. That is why this page checks the odds and does not judge them.
- Fixing the scoring before the result is the rule clinical trials follow when they register their measures in advance. It stops a scorer from choosing, afterwards, the measure that looks best.
- Theirs and ours. The Brier score and calibration are from weather forecasting, and the touch formula is Bachelier's, all credited below. The levels, the scoring rule fixed in advance and the turn entries are ours.
Credits
- Glenn W. Brier, "Verification of Forecasts Expressed in Terms of Probability", Monthly Weather Review, 1950: the score in step 2.
- Louis Bachelier, "Théorie de la spéculation", Annales scientifiques de l'École Normale Supérieure, 1900: the first model of prices as Brownian motion, which the textbook touch formula comes from. The formula's doubling rule (the "reflection principle") is a standard result. We aren't sure who first stated it, so we don't name anyone.
- Allan H. Murphy and Robert L. Winkler, "Reliability of Subjective Probability Forecasts of Precipitation and Temperature", Journal of the Royal Statistical Society, Series C (Applied Statistics), 1977: calibration.
- The liquidation levels follow the exchanges' published margin rules, simplified (isolated margin, a 0.5% maintenance margin).
- The curve of touch odds is our own, from our note "Bitcoin after the CPI flush: where the liquidations sit".
- Data: Bybit's and Binance's public 1-minute BTCUSDT perpetual prices.
- PerpGremlin and Claude, Bitcoin after the CPI flush: how the levels did, a day later, TerminalGremlin, 2026: the tests, the code and the words.
← Bitcoin after the CPI flush, 14 OctBitcoin after the PPI flush, 15 Oct →
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.