The shadow · 影
The next strategy
earns trust in public first.
Before any rule change is adopted, the candidate runs here in shadow: computed daily from live data, next to the current rules, with every disagreement logged. The spec is frozen. No capital follows it. If it degrades outside the backtest, this page is where that shows.
Today's reading
data through 2026-07-15 (UTC daily closes) · run 2026-07-16 · gates agree
The would-be record
since 2026-07-15 · day 1 · $100,000 base · marked daily at the close
This starts at inception and is never backfilled from the backtest: a shadow that borrows the backtest’s history is the backtest wearing a costume. It is computed each day by the same engines the backtest runs on, so it is a deterministic function of the frozen rules and public price data. It will be flat whenever the gates say cash, which is most of what a regime filter does.
The book today
What the rules would be holding right now, chosen at the last 14-day rebalance. Not today’s ranking: between rebalances those differ, and the held book is the honest answer.
Long book (majors, top-5 by 60d momentum)
In cash: no long gate is on.
Short book (top-5 by bear score)
Empty: the bear gate is off.
The divergence log
1 shadow day observed, 0 divergences between the gates. Divergent days are the test: they are where the candidate either earns its keep or gets sent back.
| date | v1.0 gate | v2.0 ensemble | bear | verdict |
|---|---|---|---|---|
| 2026-07-16 | · | · 0/5 | · | agree |
Data: Binance USDT spot daily klines + perp funding (completed UTC days). The full candidate spec, every backtest number, and the caveats: strategy v2.0 in shadow. The spec (gates, seeds, universes, frictions) is frozen before the shadow starts; any change restarts the clock. A shadow book is a log, not a recommendation. Decisions are yours.
Questions
What is the Koryu shadow?
A public test period for the next version of our rules. The candidate (v2.0: an ensemble regime gate for longs plus a bear-market short book) is computed every day next to the current rules, and both outputs are published here. No capital follows the candidate until the shadow record has done its job.
Why publish a strategy before adopting it?
Because backtests flatter. A frozen spec running forward in public, divergences and all, is the only evidence that survives contact with live data. If the candidate is worse than the backtest suggested, this page will show it, and that is the point.
What are the two gates?
The current gate is deterministic: BTC above its 50-day average with the 20-day above the 50-day. The candidate gate is a majority vote of five hidden Markov models on BTC trend features, refit monthly, strictly causal. Days where they disagree are logged as divergences: those days decide which gate earns the job.
Is the shadow book a recommendation?
No. It is a log of what a frozen rule set would hold, published for verification. It is not investment advice, and short books in particular carry funding costs and liquidation risks the log only models. Decisions are yours.