How to read your results
What the two grades mean, what the colours say, what "direction open" means, and how to firm up the links that matter.
The map in one sentence
Every AI analysis gives you a map of your data. Its links are what's connected, and each link's direction is what causes what. A short summary at the top of the results says how many relationships the AI found; on a sample with a known answer it also counts them for you: correct, reversed, direction open, still to find, and extra.
The two grades
On a sample with a known answer, Ground Truth Validation gives two grades.
- Links grades what's connected: the right pairs found, with any extra link counted against it.
- Direction grades what causes what, among the true links the AI found. A correct direction counts in full, a link whose direction the data leaves open counts half, and a reversed one counts zero.
Both use the same letters: A from 90 %, B from 80 %, C from 70 %, D from 60 %, and F below that. For example, 5 of 6 true links with 0 extra is Links A; 3 correct, 0 reversed and 2 open is Direction B. The precision, recall, F1 and SHD figures sit under the grades for anyone who wants the detail.
What the colours say
On the map, an arrow points from cause to effect: green for an effect that pushes the result up, red for one that pulls it down, and a thicker arrow for a stronger effect. A dashed line without an arrowhead is a link whose direction the data leaves open.
In Ground Truth Validation's lists:
- Green, Correct: confirmed, found with its true direction.
- Grey: the right pair, with its direction still open.
- Amber, Reversed and Missed: worth a look; one points the other way, the other is still to find.
- Red, Extra: check this one; the known answer leaves this pair unlinked.
"Direction open"
Some patterns in data fit either direction equally well. When that happens, the AI shows the link and leaves its direction open, instead of guessing. The link itself is real; a quick test or what you know about the process settles which way it runs.
More rows, clearer results
The AI does its clearest work on tables of 5 to 30 number columns with at least 500 rows. With 2,000 rows or more, links are firmer and fewer directions stay open.
Confirm the links that matter
The map is the most likely explanation of your data. Before you act on a link, confirm it with a quick test: change that factor on a small scale and watch the result. Then you have evidence your team can act on.