I have searched the meta, and I am surprised that we have not brought up this question yet (as far as I know, perhaps I missed something). It has been expressed that it is hard to tell what consensus is on this site, what should count as consensus?
Consensus is what it is, everyone agrees. Of course we can't check that. I think if no one participating in the discussion disagrees and there is enough participation then it can be taken as consensus.
I am guessing that you are asking this for moderation purposes. I.e. you want to know when you can act based on a meta discussion. Here is what we do on cstheory:
We use meta-discussions with policy tag to pass policies. Users including moderators can propose policies on meta and tag them with it. The post should begin with something like
This is a policy question. The proposed policy in the question (or answers) will be adopted in the case that after 14 days it received at least 5 up votes with a 2/3 majority in favor. Please upvote or downvote this question based on your opinion of the policy statement as presented (not based on whether the premise of the question is reasonable).
[A SHORT CONTEXT]
POLICY: [PROPOSED POLICY]
The time period should be such that people participating on meta will have time to see them. Two weeks should be a reasonable amount.
The minimum number of up votes is to guarantee that enough people have participated and the policy has enough support. It can be different from one site to another one and may need to be increased as you get more active users. 5 seemed to be a reasonable threshold for us.
2/3 majority is somewhat arbitrary. We preferred it to simple majority since for a policy to work you really need significant support for the policy among users. Simple majority didn't seem to be good enough for this reason.
Also I would suggest that you also take into account when there is a strong opposition against a proposed policy. Try to address their concerns with amendments. You don't need to accommodate every single objecting user. However, you don't want to alienate active users even if they are a minority.
By the way, we usually have a preliminary meta post where we discuss the issue before posting a policy question. In the policy question users vote and suggest amendments. Having these primarily discussion posts keeps policy posts cleaner.
After the time period, we lock the question and its answers and look at the votes. If the policy passed we state that on the top of the question, something like:
This policy is now in effect, with 13 votes in favor and 2 against. In addition, three of the amendments suggested in comments passed as well. The new policy on [TOPIC] is as follows:
POLICY: [STATEMENT OF THE POLICY INCLUDING PASSED AMENDMENTS]
If it is rejected we put something similar:
This policy was rejected, with 2 votes in favor and 13 against.
Here is an example from cstheory: Questions about topic for research project
CS uses a similar process. Here is an example from CS: Homework policy
How far an opinion is agreed to on meta. Something with 1-2 upvotes is not really consensus, but one with 5 or more likely is.
Something agreed to on chat should not count, because it goes by so fast and the majority will never have a chance to see/discuss a topic, much less agree on it.