Your report shows two numbers that sound like they should agree, and sometimes they point in different directions. A test can show very high confidence and still show high Risk. Both numbers can be correct at the same time. Here is what each one actually measures.
The short answer
Confidence tells you which variant is ahead. Risk tells you how much you stand to lose if that call turns out to be wrong.
Direction and size are two separate questions. You can be sure of the direction and still know very little about the size.
The two numbers
Confidence appears in your report as Chance to Beat Original, or as Chance to be Best if your test has more than one variant. It is a probability. 95% means that if you called this variant the winner right now, you would be right about 95 times out of 100.
Risk is the average amount you would fall behind by, counting only the outcomes where the other variant is genuinely the better one. It is a size, not a probability. Mida works it out by simulating 30,000 possible versions of your test from the data so far, keeping only the ones where the other variant comes out ahead, and averaging how far behind you land in those. Lower is better.
So Confidence answers "how likely am I to be wrong". Risk answers "if I am wrong, how badly".
Risk is not 100 minus Confidence
This is the most common misreading. If Chance to Beat Original is 97%, Risk is not 3%.
Those 3% of outcomes are not all near misses. Some of them are outcomes where the other variant wins by a wide margin. Risk reports the average size of the gap in exactly those outcomes, so it can sit in double digits while confidence reads 97%. The two numbers are not two views of the same thing, and subtracting one from the other will not give you the other.
Why this usually happens: one arm with very few conversions
The clearest case is a control that has not converted yet. A test in this shape:
Control: 224 visitors, 0 conversions
Variant 1: 240 visitors, 4 conversions
Confidence is very high. 4 beats 0 and there is nothing ambiguous about the direction.
Risk is also high, around 25%. With zero conversions, the control's true conversion rate is still almost unknown. It might be genuinely poor, or it might be about the same as the variant and simply unlucky over 224 visitors. In the minority of simulated outcomes where the control really is the better arm, it is better by a lot, and that is what Risk is picking up.
Read together, the two numbers are saying: we are confident about the direction, and we still know very little about the size. For a test that is a couple of days old, that is normal and expected, not a bug.
A test in this shape will usually also show a very wide credible interval, for the same reason.
The thresholds Mida uses
Below 1% is Very Safe
Below 10% is Safe
10% or above is risky, and the banner will say the variant is likely the winner but risky
These caps are fixed. Your confidence level setting changes how much probability Mida needs before it calls a winner. It does not change how much loss Mida is willing to accept.
What to do when confidence is high but Risk is not
Keep the test running. Risk falls on its own as conversions build up on both arms, because there are fewer and fewer plausible outcomes where the losing arm is secretly much better.
Do not ship on confidence alone while Risk is above 10%. High confidence with high Risk usually means the direction is right but the size is unknown, and the size is what you are actually banking on when you roll the change out.
If you have to move now, treat it as a business call rather than a statistical one. Shipping the leader is reasonable when the change costs nothing to keep. It is a worse idea when you are about to build a plan or a forecast on the size of the lift.