Every signal is written to a ledger at the close that produced it —
before anyone knows what happens next. Outcomes are computed mechanically afterwards and
never revised. Every recorded signal appears here, including the losers.
Nothing is chosen after the fact — though the headline average counts finished
trades, and losers that have not finished are still open. Both figures are shown.
How to read this — and why it is set up this way
the point
A backtest can be tuned until it looks good. A forward test cannot, because the signal is published before the outcome exists. This page is the strategy trying to prove itself in public, at the pace the market allows.
R
Profit or loss in units of the risk taken. R is a volatility unit fixed at the signal — so a later data revision cannot quietly improve a past trade.
entry
The next session's open after the signal. Never the signal bar itself — that shortcut inflated a different indicator here by 0.71R per trade.
exit
A short-term strength signal, or a hard time limit — whichever comes first. A tight stop was tested on identical entries and made results worse (1R binds 12.3% of the time and costs a third of the edge); only the far 8R backstop is carried.
expectation
Backtested edge is about +0.09R per trade at a ~67% win rate on a ~4-day hold, net of costs — but that is a full-history average and the edge is smaller lately: +0.126R across 2010-15 against +0.063R across 2023-26. Plan on the recent-era figure, about +0.06R. Anything much larger here would be luck, not vindication.
the tail
Risk is concentrated in the extreme tail: the 5th percentile is only about −1.1R, but with no stop the worst of 74,689 backtested trades was −18.2R. From 31 Jul 2026 the rule carries a catastrophe stop 8R below entry, which binds on ~0.03% of trades and cuts that worst case to −8.0R for ~1.2% of expectancy, with maximum drawdown unchanged. It is a cap with slippage, not a floor — a gap can fill through it. Signals recorded before that date were logged without a stop and are filled without one; they are not rewritten.
what is withheld
The indicator, its periods and its thresholds are not published — that is the part subscribers pay for. Everything needed to check this ledger is here: every ticker, every date, every fill, every outcome. Nothing needed to clone it is.
how independent
Trade count overstates the evidence. Signals arrive in bursts and exit together, so several "trades" can be one market move. The exit days figure counts distinct exit mornings, and independence.clustered_t in the feed treats each as a single observation — the honest denominator.
how many at once
This ledger fills every signal, and the strategy fires roughly 21 a day. At a $50 risk unit that is a ~$34,000 account carrying 40+ positions at once — not the portfolio most readers would run. The board is ranked, and what a subscriber actually trades is the top few names each day. The feed publishes that separately as cohorts.top_2_per_day; the headline figures above are the all-signals book. Neither is adjusted for the other.
honest limit
A few hundred trades are needed before this says anything. Early numbers are noise, and will be left standing either way.