▲ 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
Confluence: the failures, together
Research, not a recommendation to trade · updated 30 Oct 2026 22:00 UTC
Five signals we tested on their own, each of which failed: none called Bitcoin's next move better than chance. This page asks a different question: when several of them lean the same way at once, does that say anything? Each is read every day at 00:00 UTC and counted. The rules were written down before the first reading, and the answer comes from the days after 27 October 2026.
2 of 5 lean up, 2 lean down, 1 neither · 2026-10-30 00:00 UTC
The market's mood (the engine's reading on the daily chart, shown, not counted): ranging
| signal, and what it reads | now | leans |
|---|---|---|
| Purple whales the $100k–1M size band's six-month reading on Binance spot, 0 to 100, a day late (the archive's day arrives the next day) | 4 of 100 | down |
| Last week's trend Bitcoin's change over the last 7 days | -4.2% | down |
| Fibonacci structure the daily trend by its swing highs and lows, as the Fibonacci study reads it (derived: that study traded 15-minute reactions) | up | up |
| QFL, daily the latest QFL buy or sell on the daily chart in the last 7 days (turned round: it pointed the wrong way in 2018–2022) | none | neither |
| Day of the week whether this weekday rose on average over 2018–2022 | Friday | up |
What it shows: Each signal's reading today and which way it leans, and the count of up against down.
How to read it: A lean is the direction the signal pointed in 2018–2022, the years its sign was fixed on. Five up would be every signal agreeing on a rise; it is a reading of the signals, not a forecast. Each signal's own result is on its study page.
Shown, not counted
- Liquidation distance (young): a 25x long from the last daily swing low would be liquidated 7.2% below; a 25x short from the last swing high 16.0% above.
- Crowding (young, Binance): funding over the last three payments +0.022%, open interest -5.3% over 7 days.
The record
Log scale: equal steps up are equal percentage rises.
Before 27 October 2026 the rows are backfill, worked out afterwards; only the days after count.
The test
For every day, Bitcoin's change over the next day, week and month, set against the count. The score is a rank correlation: +1 if the days with the most "up" votes were always followed by the best moves, 0 if there is no link. Each is checked against two kinds of luck, 1,000 times each: the count slid to other dates, and the votes redrawn at random. p is the share of those that did as well, the larger of the two shown; small is better.
| stretch | days | next day | next week | next month |
|---|---|---|---|---|
| 2018–2022, where the signs came from (backfill) | 1,459 | +0.08 p 0.00 | +0.12 p 0.04 | +0.20 p 0.04 |
| 2022–2026 (backfill) | 1,579 | -0.00 p 0.51 | -0.03 p 0.85 | +0.06 p 0.23 |
| from 27 October 2026, the test | 4 | – | – | – |
The looks, fixed in advance: 25 January 2027 (a progress report only), 27 October 2027 (the next day and week: it holds if p is under 0.025 against both kinds of luck), 27 October 2028 (the next month, the same bar). The 2018–2022 rows look good because the signs were chosen on them; 2022–2026 showed nothing.
The latest rows
| day | up | down | count | BTC open |
|---|---|---|---|---|
| 10-30 | 2 | 2 | +0.0 | 77,992 |
| 10-29 | 1 | 3 | -0.4 | 81,344 |
| 10-28 | 2 | 2 | +0.0 | 80,910 |
| 10-27 | 2 | 2 | +0.0 | 80,151 filled later, 10-28 00:30 UTC |
| 10-26 backfill | 2 | 2 | +0.0 | 80,154 |
| 10-25 backfill | 2 | 2 | +0.0 | 80,888 |
| 10-24 backfill | 2 | 2 | +0.0 | 81,189 |
| 10-23 backfill | 2 | 2 | +0.0 | 81,416 |
| 10-22 backfill | 2 | 2 | +0.0 | 84,093 |
| 10-21 backfill | 2 | 2 | +0.0 | 83,028 |
Educational research from public market data; any trading discussed is paper, never real. Nothing here is advice. The full notice.