How reputation works

You cannot review someone because you disliked them in a random dungeon, and you cannot review a stranger at all. A review is only possible when this platform can establish that both players genuinely took part in the same group.

That means a group listing that was filled, played, and marked complete. The participant list is frozen at that moment, so later changes to the group cannot grant or remove anyone’s right to review.

There is no other route. No form, no admin tool and no API can create the record that makes a review possible.

When you review someone, they cannot see what you wrote until they have reviewed you too — or until the review window closes, currently 48 hours after the group finishes.

Without this, the first person to read a review could simply retaliate with a matching one, and the whole system would measure who checked their notifications first rather than what actually happened.

You can always see your own submission, so you never have to wonder whether it went through.

One question carries the most weight: would you play with this person again?

Then five optional dimensions, each rated one to five:

• Communication— How well did they communicate?
• Attitude— Were they respectful and enjoyable to play with?
• Reliability— Did they show up and stay for the whole run?
• Teamplay— Did they work with the team?
• Mechanics— Did they execute the expected mechanics?

You can skip any dimension you have no basis to judge, and doing so does not count against the person. The weight is redistributed, not scored as zero.

The negative options are deliberately factual rather than insulting. Free-text comments are optional, carry no weight in the score, and can be hidden by moderation without touching the rating they came with.

A percentage on its own is close to meaningless. This is why we never show one without the number of groups behind it, and why the two profiles below are not ranked the way their percentages suggest:

The second player ranks higher, because two reviews are not evidence and a hundred and eighty are. New profiles start from a neutral assumption rather than from zero, and they are labelled provisional until enough different people have weighed in.

Diversity counts separately from volume: forty reviews from three people is not the same evidence as forty reviews from forty people.

• Friends boosting each other. The second review you leave for the same person is worth half the first, the third a third, and so on. Ten mutual reviews are worth about three.
• Brigading. The same damping applies to negative reviews. Twenty angry reviews from one account move a score by a few points, not tens.
• Throwaway accounts. Reviewing requires a Battle.net verified account, and the confidence figure counts distinct reviewers rather than raw review volume.
• Old grudges. A review is worth half as much after 180 days — though it never becomes worthless, because a pattern is still a pattern.
• One bad night. Small samples are pulled toward neutral, so a single negative review cannot destroy an account.

We also track internal signals for moderation — things like whether someone’s reviews all come from the same handful of accounts. Those are never shown publicly and never exposed through any API, because publishing them would just be a manual for evading them.

• It does not measure whether someone is a good person. It aggregates a limited number of subjective reports about specific groups.
• It is not precise. We round to whole percentages and always show the sample size, rather than implying accuracy we do not have.
• It is not a ranking of skill. That is what the performance tab is for, and the two are kept apart on every surface.
• It is not permanent. Recent behaviour carries more weight than old behaviour, so a profile can recover.