Any slicing of statistics will show something. The edge map tests each cut statistically and shows only the ones that hold up.
Slice statistics long enough and some cut will show a profit. You can't tell a finding from chance by eye, and traders lose money trusting a pattern that isn't there.
five out of sixty-one
How it works here
Setup, market regime, session, instrument, weekday, holding time and their combinations. In the demo journal that's 61 cuts over 318 trades.
A confidence interval, a correction for multiple comparisons and a check window by window. In the demo, 46 cuts reached a usable size and five passed the gate.
A cut that held in early windows and gave out in the last one goes to the decay section. 'Worked and stopped' reads differently from 'doesn't work'.
How many trades a claim needs to be reliable, what the interval is now and what happens if your trades are reshuffled a thousand times.
A short interview about your style: how you enter and exit, the risk you hold, what you call a mistake. Facts are built from your answers, files and trades, and the AI judges a trade by them.
On a day with an unusual move, the system looks for what else moved unusually and counts how often that coincidence happened over the year.
Setups, regimes, sessions, instruments and mistakes are nodes. Link thickness shows the number of trades, colour shows the result. A hypothesis can only be recorded together with the cut it refers to.
Where it differs
Typical analytics shows everything it computed and leaves you to decide what to trust. The edge map shows five cuts out of sixty-one, each with a status: established, emerging, hypothesis, not enough data. Twelve trades give no finding, and the screen says so.
Check it yourself
Create the demo journal with one button and open the edge map. The pattern is built into the trade prices, not the labels, and the engine finds it with the same code that runs on your trades.
The order
A couple of minutes about your style. You can skip it and come back later.
While trades are few, the map says 'not enough data'. That's normal for the first weeks.
Each card opens into the list of trades it was built from.
What it does not do
Questions
The rest failed the test. Showing them would make noise look like a finding.
No. The map, the trust figures and the graph come from code under tests. A model can retell the result in words, but *it produces no numbers*.
It depends on the cut and the spread of results. The trust screen names the sample needed for each figure.
A cut on the map has survived the interval, the correction and the window check. Everything else is labelled for what it is.