Every flip we have recorded, and what the price did next.
A flip is written when its candle closes and is never edited afterwards.
What happened after DeepBook's weekly flips
There are too few flips here to say anything about what follows them.
| Which flips | Days later | Flips counted | Median change | Range of the median | Ordinary days | Against ordinary days |
|---|---|---|---|---|---|---|
| After bullish flips | 7 | 0 of 1 | not yet | – | – | not yet |
| 30 | 0 of 1 | not yet | – | too little history | – | |
| 90 | 0 of 1 | not yet | – | too little history | – | |
| After bearish flips | 7 | 1 of 1 | +9.4% | – | −14.3% to +33.5% | within the range of ordinary days |
| 30 | 1 of 1 | +9.5% | – | too little history | – | |
| 90 | 1 of 1 | −4.9% | – | too little history | – | |
| Any day since 7 Jun 2026, for comparison | 7 | 111 days | −0.3% | – | – | – |
| 30 | 88 days | too little history | – | – | – | |
| 90 | 28 days | too little history | – | – | – |
Flips counted: 4 of 5 means four of the five flips are old enough for that many days to have passed. Each median uses only those; with an even number of flips it is the lower of the two middle results.
Range of the median: worked out from the flips' own results, with no assumptions about them. The share in brackets is how often a range built this way holds the true median; with fewer than 6 flips it runs from the lowest result to the highest.
Ordinary days: the median of the same number of ordinary days from the same period lands in this range 95 times in 100. A median is called within the range of ordinary days unless it falls outside it. About 1 cell in 20 falls outside by chance alone.
The latest flip is too recent to count in the 7, 30 and 90 day columns.
Prices are weekly closes on Coinbase. Returns are measured from the close of the flip candle.
What we tested
We tested 56 different settings and filters for the trend line, on daily and weekly candles (100 rule-and-timeframe combinations), on 129 coins that have at least three years of history (128 excluding Bitcoin for the setting grid). Hourly (4h) and monthly lines and the Bitcoin-priced lines were not tested. Out of sample means January 2022 to October 2026: one mostly falling stretch, with survivors only.
For altcoins on daily candles, after 2022, settings from factor 2 to 3 gave similar results (30-day gap between what followed bullish and bearish changes 2.4 to 5.0 points, against 3.4 for the published setting; none of the 90-day differences was clear). Before 2022 the published setting was at the low end of that range, and a factor 1.5 line was clearly worse in both periods. For Bitcoin, one series with 45 daily flips since 2022, the evidence is too thin to defend or replace its setting. That the published setting sat in a flat area does not show it was chosen without hindsight.
Rules that wait for extra closes, demand a margin beyond the line, or use a slower factor changed the label less often or later. They did not clearly raise how well it separated what followed after 2022, and the margin rules fell further from the peak before turning bearish. Rules that go neutral when the trend looks weak (ADX, Choppiness, efficiency ratio) made the label change more often, not less (daily: 15 to 86 changes per coin-year against 14), and the flips they let through were followed by much the same results as the rest.
For altcoins on weekly candles the published setting sits next to a cliff: a wider band changed state later and separated what followed less (90-day gap down 12 to 26 points; the calendar-based intervals are wide, 19 quarters). Bitcoin's weekly line has only 4 flips since 2022 and shows no such pattern.
Coins change state together: on 69% of coin-days a coin's daily state matched Bitcoin's (about 50% would be expected by chance; 63% to 65% in 2025 and 2026, 80% in 2022). After 2022 the 30-day median gap between bullish and bearish changes was +3.3 points, but 'about zero' (-0.2, range -0.7 to +0.2) once each change is compared with all coins on the same dates. That holds for medians; by averages the same-date gap is +2.4 points [0.2, 5.2], driven by a few large rises. The calendar-based range of the raw gap includes zero.
Only coins in today's list have a record here; coins that left the list before 1 Oct 2026 are missing, which tends to make past results look better than they were for the coins of that time. Each coin's ordinary-day row shares this, so compare the two.
The study's notes and tables are kept with the project's research files.
How soon flips were reversed
| Which flips | Flips | Reversed within 3 candles (3 weeks) |
|---|---|---|
| After bullish flips | 1 | not yet |
| After bearish flips | 1 | 0 of 1 |
A flip counts as reversed when the next flip came within 3 candles, 3 weeks. A flip is only counted once 3 candles have closed after it.
There is less than a year of flips here, too little to count flips a year.
DeepBook, weekly flips: 1 bearish flip of 2
Each figure is the price change from the flip price: 7, 30 and 90 days later, and at the next flip. Green means the price rose and red that it fell, whatever the flip said.
| Candle closed | Flipped to | Price at the flip | 7 days later | 30 days later | 90 days later | Until the next flip |
|---|---|---|---|---|---|---|
| 7 Jun 2026Lasted 16 weeks | Bear | $0.01659 | +9.4% | +9.5% | −4.9% | +44.8% |
How to read this
Closed candles only
Flips are computed on closed candles only, so a flip never appears and then vanishes while a candle is still open. The ledger is append-only. When the rules change, the rule set gets a new version number and the old entries stay exactly as they were.
Compare with any day
The comparison rows show what the price did after any day since 7 Jun 2026, from the first flip on these candles on, so they cover the same period as the flips. A row that rests on fewer starting days than four times its number of days says too little history.
A small sample
Two flips are a small sample. The numbers describe the past and are not a forecast, and nothing here is advice.