Method

One deterministic trading system, live since 2026-09-02, measured against one benchmark: holding Bitcoin. It is run like a registered trial. Every idea is written down with its expected result before it is tested, tested on fixed windows that never move, published whether it passes or fails, and hashed into Bitcoin before anyone can know the outcome.

What is not claimed: that the system beats Bitcoin. It has been live for days. The record starts before it looks like anything, on purpose.

The benchmark

Hold-BTC is 1.000. Every result on this site is a multiple of holding Bitcoin over the same window, after fees. There are no dollar returns, no annualised figures and no Sharpe ratios standing alone. An altcoin that rises 20 % in dollars while Bitcoin rises 30 % is a loss, and is recorded as one.

The benchmark epoch is post-ETF: January 2024 onward. Earlier data is used for stress, never as the standard.

The pipeline

1. A question arrives. A trader’s claim, a vendor’s indicator, an idea of our own. The filter is simple: a named rule can be tested; dated conditions (“if it breaks X by Friday”) are recorded as a voice and scored later; anything without a rule is nothing.
2. Pre-registration. The test is written before it runs: which lines, which windows, at most four free parameters, and what each line is expected to do. The file is hashed into Bitcoin at that moment.
3. Fixed windows. One benchmark epoch, post-ETF (January 2024 to today), cut in two: a fit window (January 2024 to August 2025) that is the only thing a line is ranked on, and the last twelve months as a holdout that is read as pass or fail and never ranked. The post-ETF number is the two joined. Earlier history — from 2020, and from 2015 where the data exists — is a stress test: it can fail a line, it can never promote one. The windows do not move between batches.
4. Verdict and scorecard. The result table, and a line-by-line account of which predictions were right and which were wrong. Published either way, and hashed.
5. Shadow book. A line that survives runs as a paper book against hold-BTC on the same prices and the same fee model as the live book. After 90 days it must beat the benchmark on both multiple and drawdown to be promoted. Nothing goes from a backtest into the live book directly.
6. Decision ledger. Every live proposal, approved or vetoed, is scored at 1, 3, 7 and 30 days against how the asset/Bitcoin ratio actually moved. This is where the risk engine’s vetoes are held to account.

Every trial ever run stays in the ledger and counts as the denominator for the deflated Sharpe ratio. The more things we test, the harder it gets for the next one to look good by chance. That is the point. As of the first publication the ledger holds 301 trials in 26 batches.

What is compared against what

• The live book against hold-BTC, as a multiple, since the day it went live.
• Every altcoin against the Bitcoin ratio, never against dollars.
• The live book against its own shadow twin, which runs the same rules without execution friction; the gap is the cost of being real. The twin was rebased onto the live book on 2026-09-07 (see the scoreboard).
• Approved proposals against vetoed ones, at fixed horizons.

Not compared: other people’s track records. Analysts and sources we monitor are scored privately until they have a record of their own. Publishing that is a separate decision, not yet taken.

The architecture, in principle

The brain is deterministic. There is no language model in the trading path; one is used only to turn analysts’ prose into structured fields, and those fields carry no weight until the source has a scored record. A human holds every irreversible switch. Spot only, one collateralised hedge, no leverage beyond it, no shorting. All data sources are free tier. The tokenised-equity pot on Solana is its own ledger, benchmarked in USDC, and is never folded into the Bitcoin book — an accounting lesson learned on 2026-09-04 and kept.

Kitsunebi

Foxfire, in the old stories, is the light that leads travellers off the path. Kitsunebi is the ledger that does the opposite. Every hypothesis file, verdict, shadow-book definition and daily snapshot of the decision ledger is hashed, and the hash is timestamped into Bitcoin through OpenTimestamps: free public calendars, no wallet, no token, no chain of our own. A verifier needs the file, the proof and any Bitcoin node.

The ledger was first anchored on 2026-09-04. Rows registered that day are marked retro: they prove the files existed on that date, not on the earlier dates the version history shows, and the ledger says so on every such row. From that day on, registration and anchoring are the same event.

What stays private

The rules, the thresholds, the sizing, the universe, the live positions, the capital. Numbers travel; parameters do not. The reason is plain: the mechanics may become a product one day, and a product bought from someone whose failed hypotheses you have read for a year is worth more than one bought from a stranger.