How alerts are graded
The same rules grade the house alerts and yours. In use since 22 Oct 2026.
01Entry
When an alert fires, the entry is locked at the screener's last price, never older than 120 seconds. If no fresh price exists, the alert does not fire.
02Stop and virtual target
Each timeframe gets its own stop at 3 × ATR(14) of that timeframe from entry, against the direction of the alert, and a virtual target the same distance in its favour (1R). The target exists only for grading; nothing is traded.
03The 20-bar window
Each firing is followed for 20 bars of each timeframe: 20 hours on 1h, 80 hours on 4h, 20 days on D. Every firing is graded on all three, independently.
04Settlement order and the same-bar rule
Price is walked in order from the firing, in 5-minute bars (1-hour bars for D), and the first level touched decides the result. If the stop and the target both sit inside the same bar, it counts as the stop: we never assume the better outcome. A gap through a level is graded at the level.
05In profit or in loss at window end
If neither level is touched within 20 bars, the firing settles at the window's last close: in profit counts as a win, in loss as a loss, and a close exactly at entry counts as a loss. The outcome split shows these separately from target and stop hits.
06Episodes
Firings of one alert on one market that overlap in time form one episode, graded once: a firing joins the episode when it fires before the previous firing's window ends. An episode is a win when more than half of its graded firings won; a tie counts as a loss. The episode win rate stops a burst of firings in a single move from inflating the record. Both rates carry a 95 % Wilson range.
07Validation tags
A house alert earns a tag on its headline timeframe, from all settled data since 22 Oct 2026. It needs the settled firings and both range bottoms below, and its first and last settled firing at least 14 days apart. The tag is the highest tier it passes; tags move up and down as firings settle. Below Initial's count or span it is Collecting; with the count and span but not the bars, Not passing. We publish house alerts from Initial up; one that falls below Initial leaves the public table automatically.
| Tag | Settled firings | Firing range bottom | Episode range bottom |
|---|---|---|---|
| Initial | ≥ 30 | > 50 % | > 45 % |
| Medium | ≥ 100 | > 52 % | > 48 % |
| Strong | ≥ 200 | > 55 % | > 50 % |
08Headline timeframe
Each alert has one headline timeframe (1h unless chosen otherwise). Its record and its tag use that timeframe. Wherever 4h or D numbers are shown, they appear without a tag. Your own alerts show the same range and sample size, never a tag.
09Versioning
A definition change starts a new version with a fresh record. Changing the trigger, the direction or the conditions is a definition change; renaming, changing markets or the headline timeframe is not. Old versions keep their record.
Firings on many markets in the same move are not independent, so ranges are narrower than they should be. Read a tag as a reason to look closer, not as proof.
10Corrections log
Threshold changes and any re-settlement are recorded here with their date.
Information, not advice. Past alerts don't predict the next one.