The published miss rate
Every call we publish, scored in public.
Before you let a signal move your money, read the record: how many wicks we published, how many carried driver unknown, and how many verdicts were wrong. Every figure is computed from the archive, never entered by hand.
Wicks published
—
since launch
Driver unknown
—
move published, no cause named
Verdicts wrong
—
broken down by driver class
Judged in private
—
withheld below the publish floor
The engine is not live, so there are no figures yet. When it runs, these come from the same stored events subscribers received, and this page shows them whether they flatter us or not.
Methodology
How the number is produced.
A record is worth something only if you can see how it was made and reproduce it yourself.
Outcomes at three horizons
For every wick the price is recorded 5, 15 and 60 minutes later. The reaction verdict is judged against those, not against a later narrative. A wick with a missing outcome is reported, never dropped.
Computed from the archive
The counts come from the same stored events subscribers received, and the published total must equal the archive total. Sampled operational counters are never the source of a published figure.
Nothing filtered out
No category of wick is excluded to improve the figure, and the miss rate is broken down by driver class. Unknown drivers are counted as what they are.
The publish floor is dated
A verdict publishes only at or above the floor, which starts at 0.70. Every change to it appears here with its date, beside the count of moves judged in private, so abstention cannot hide a weak model.
Why publish it at all
Because a claim you cannot check is worth nothing.
Anyone can say their signals are good. A record computed from the archive lets you verify it before you risk anything, and lets you size positions on measured accuracy instead of a sales page. Replay our history and you can regenerate these numbers yourself, which is the only reason they are worth reading.
Your agent sees the price move. Send it the reason.
Subscribe with an npayload key. The free tier is realtime, capped at five events an hour, and eight lines of code print your first event.
No card. No login. No dashboard.