How to read a cell without fooling yourself.

Click an hour. Any hour.

Below is EUR/USD, 23 years of real hourly bars, weekdays only, in UTC — the same data the dashboard reads. Each square is one hour of the day, coloured by how often it closed higher than it opened.

The percentage is the headline. It is not the story.

The big number is the up-rate — the share of those hourly candles that closed above their open. 50% is a coin flip. 54% means that in 100 of those hours, 54 finished higher.

That's all it means. It does not say how far price travelled, whether the winners were bigger than the losers, or that the next one will go up.

Look at n before you believe the percentage.

n is how many real bars produced that number, and it decides how much of the percentage is signal and how much is luck. The noise floor falls with the square root of n:

The brass dot answers one question only.

A dot means |z| ≥ 3: for that sample size, the gap from 50% is big enough to be unlikely from chance. The z-score is just the two numbers above combined — z = (2·ups − n) / √n.

It tells you the pattern is probably real. It does not tell you the pattern is worth trading. Those are different questions, and the next step is the one that separates them.

Now the number nobody looks at: average move.

Direction is free to measure. Trading it is not — your broker takes the spread on the way in. So compare the bucket's average move against your typical spread, and most "edges" quietly disappear.

This is why the dashboard shows average move next to every up-rate, and why a strong-looking cell in the 21:00–22:00 UTC rollover — where spreads widen sharply — gets flagged rather than celebrated.

Zoom out before you zoom in.

A single weekday×hour cell is the noisiest thing on the dashboard. The wider views are far more trustworthy because the sample is much larger:

• Hour-of-day — thousands of samples per bar. The most reliable view on the site.
• Sessions (Tokyo / London / New York) — aggregates nine hours at a time; excellent for "is this session directional at all?"
• Hour × weekday — where genuine structure shows up, but check n every time.
• Month filters — divide n by twelve. Treat with suspicion.

Start wide, confirm the story holds as you narrow, and stop the moment n gets thin.

Four ways to misuse this.

• Trading a cell as an entry. It's a filter for when, not a signal for now. Bring your own setup.
• Hunting for the biggest percentage. The most extreme cell is almost always the thinnest sample.
• Ignoring the spread. See step 5. This is the one that actually empties accounts.
• Assuming it holds forever. These are base rates from the past. Market structure changes; rollover times change; brokers change.

The FAQ's "what it can't do" section goes further — including why this data doesn't support 1-minute binaries or swing trading.