Failures first
A retired idea gets the same write-up as a live one. You see the losses before you see anything else.
Opening the research
Opening the research
Most trading products show you the winners. We show you the whole record: every idea we tested, the rules it ran on, what the frozen test said, and why it failed or earned a place. If a strategy survives this page, you can see exactly how hard it was tried.
The engines' own state, read from the running record: what is open, what closed, and what the market filter is doing right now.
Read the noteA retired idea gets the same write-up as a live one. You see the losses before you see anything else.
Rules are fixed before the unseen data is opened. A failed test is not retuned until it passes.
Figures come from the test outputs and the tamper-evident signal log, not from a marketing deck.
Read from the running engines at 14:15 UTC. Market data is connected, last message 0 min ago. Every position below is paper.
The research started with Bitcoin lead–lag at second scale, then with cost-aware delayed entries. Both were retired, and both set the standard the later work had to meet.
v007 bought 15-minute RSI extremes inside a four-hour trend on BTC, ETH and SOL futures. The unseen test was negative, and the live paper record agrees.
v004 shorted altcoins that held up while Bitcoin fell, betting they would catch down. Across 349 unseen trades it lost 0.46% each.
Two frozen versions hedged altcoin futures against Bitcoin and lost after costs. The mechanism, the numbers, and the honest conclusion.
The engines' own state, read from the running record: what is open, what closed, and what the market filter is doing right now.
v008 runs two halves on 37 futures and cannot be fairly backtested, because its rules were developed on data that had already seen every past period. So it is judged live.
New notes land when a test finishes, whichever way it goes. Live paper signals post to the channel as they happen.
These notes explain research. They are not signals, advice or forecasts. Crosswake publishes paper signals only: no order is ever placed. A positive test result is not a promise, and the record's own standard says so — the main strategy's unseen-data confidence interval includes zero. Every version is in the changelog.