Using the information available to you and not excluding a person's judgment in situations where they could reasonably be called 'biased' is the standard practise in places like Y Combinator and OpenPhil. OpenPhil writes about this in the classic post Hits-based Giving. A relevant quote (emphasis in original):

We don’t: put extremely high weight on avoiding conflicts of interest, intellectual “bubbles” or “echo chambers.”

...In some cases, this risk may be compounded by social connections. When hiring specialists in specific causes, we’ve explicitly sought people with deep experience and strong connections in a field. Sometimes, that means our program officers are friends with many of the people who are best suited to be our advisors and grantees.

...it sometimes happens that it’s difficult to disentangle the case for a grant from the relationships around it.[2] When these situations occur, there’s a greatly elevated risk that we aren’t being objective, and aren’t weighing the available evidence and arguments reasonably. If our goal were to find the giving opportunities most strongly supported by evidence, this would be a major problem. But the drawbacks for a “hits-based” approach are less clear, and the drawbacks of too strongly avoiding these situations would, in my view, be unacceptable.

It seems fairly obvious to me that being in a [...] active collaboration with someone should require recusal

This seems plausibly right to me, though my model is that this should depend a bit on the size and nature of the collaboration.

As a concrete example, my model is that Open Phil has many people who were actively collaborating with projects that eventually grew into CSET, and that that involvement was necessary to make the project feasible, and some of those then went on to work at CSET. Those people were also the most informed about the decisions about the grants they eventually made to CSET, and so I don't expect them to have been recused from the relevant decisions. So I would be hesitant to commit to nobody on the LTFF ever being involved in a project in the same way that a bunch of Open Phil staff were involved in CSET.

My broad model here is that recusal is a pretty bad tool for solving this problem, and that this instead should be solved by the fund members putting more effort into grants that are subject to COIs, and to be more likely to internally veto grants if they seem to be the result of COIs. Obviously that has less external accountability, but is how I expect organizations like GiveWell and Open Phil to manage cases like this. Disclosure feels like the right default in this case, which allows us to be open about how we adjusted our votes and decisions based on the COIs present.

In general I feel CoI policies should err fairly strongly on the side of caution

I don't think I understand what this means, written in this very general language. Most places don't have strong COI policies at all, and both GiveWell and OpenPhil have much laxer COI policies than the above, from what I can tell, which seem like two of the most relevant reference points.

Open Phil has also written a bunch about how they no longer disclose most COIs because the cost was quite large, so overall it seems like a bad idea to just blindly err on the side of caution (since one of the most competent organizations in our direct orbit has decided that that strategy was a mistake).

The above COI policy is more restrictive than the policy for any other fund (since its supplementary and in addition to the official CEA COI policy), so it's also not particularly lax in a general sense.

It seems fairly obvious to me that being in a close friendship [...] should require recusal

I am pretty uncertain about this case. My current plan is to have a policy of disclosing these things for a while, and then allow donors and other stakeholders to give us feedback on whether they think some of the grants were bad as a result of those conflicts.

Again, CSET is a pretty concrete example here, with many people at Open Phil being close friends with people at CSET. Or many people at GiveWell being friends with people at GiveDirectly or AMF. I don't know their internal COI policies, but I don't expect those GiveWell or Open Phil employees to completely recuse themselves from the decisions related to those organizations.

There is a more general heuristic here, where at this stage I prefer our policies to end up disclosing a lot of information, so that others can be well-informed about the tradeoffs we are making. If you err on the side of recusal, you will just prevent a lot of grants from being made, the opportunity cost of which is really hard to communicate to potential donors and stakeholders, and it's hard for people to get a sense of the tradeoffs. So I prefer starting relatively lax, and then over time figuring out ways in which we can reduce bad incentives while still preserving the value of many of the grants that are very context-heavy.

being personal friends with someone should require disclosure.

I think this comment highlights some of the reasons for why I am hesitant to just err on the side of disclosure for personal friendships.

I didn't see any discussion of recusal because the fund member is employed or receives funds from the potential grantee?

Yes, that should be covered by the CEA fund policy we are extending. Here are the relevant sections:

Own organization: any organization that a team member

is currently employed by

volunteers for

was employed by at any time in the last 12 months

reasonably expects to become employed by in the foreseeable future

does not work for, but that employs a close relative or intimate partner

is on the board of, or otherwise plays a substantially similar advisory role for

has a substantial financial interest in

And:

A team member may not propose a grant to their own organization

A team member must recuse themselves from making decisions on grants to their own organizations (except where they advocate against granting to their own organization)

A team member must recuse themselves from advocating for their own organization if another team member has proposed such a grant

A team member may provide relevant information about their own organization in a neutral way (typically in response to questions from the team’s other members).

Which covers basically that whole space.

Note that that policy is still in draft form and not yet fully approved (and there are still some incomplete sentences in it), so we might want to adjust our policy above depending on changes in the the CEA fund general policy.

Separately, you mentioned OpenPhil's policy of (non-) disclosure as an example to emulate. I strongly disagree with this, for two reasons.

This sounds a bit weird to me, given that the above is erring quite far in the direction of disclosure.

The specific dimension of the OpenPhil policy that I think has strong arguments going for it is to be hesitant with recusals. I really want to continue to be very open with our Conflict of Interest, and wouldn't currently advocate for emulating Open Phil's policy on the disclosure dimension.

We're a pretty small team, and for most grants, there usually are only one or two people on the fund who have enough context on the project and the grantees to actually be able to assess whether the grant is a good idea. Those are also usually the fund members who are most likely to have conflicts of interest.

Since none of us are full-time, we also usually don't have enough time to share all of our models with each other, so it often isn't feasible to just have the contact-person share all of their impressions with the other fund members, and have them vote on it (since we realistically can't spend more than an hour of full-fund time on each individual grant we make).

One of the things that I am most concerned about if you were to just move towards recusal, is you just end up in a situation where by necessity the other fund members just have to take the recused person's word for the grant being good (or you pass up on all the most valuable grant opportunities). Then their own votes mostly just indirectly represent their trust in the fund member with the COI, as opposed to their independent assessment. This to me seems like it much further reduces accountability and transparency, and muddles a bunch of the internal decision-making.

The current algorithm we are running is something closer to: Be hesitant with recusal, but if the only people on the fund who are strongly in favor of a grant also have some potential weak COI, then put more effort into getting more references and other external sources of validation, but usually still taking the vote of the person with the potential weak COI into account.

Maybe we could integrate this officially into this policy by saying something like: If you have a COI of this type, we will give less weight to your vote, but your vote will still have some weight, depending on how strong your COI is judged by the other fund members. Though I am worried that this is too detailed and would require changing every time we change the local dynamics of how we vote on things.

Wanted to +1 this in general although I haven't thought through exactly where I think the tradeoff should be.

My best guess is that the official policy should be a bit closer to the level of detail GiveWell uses to describe their policy than to the level of detail you're currently using. If you wanted to elaborate, one possibility might be to give some examples of how you might respond to different situations in an EA Forum post separate from the official policy.

If I imagine myself in the shoes of a more conservative potential donor, who is checking the fine print as part of their due diligence, I would be put off by phrases like ‘metamour’, ‘consumption of drugs’ and the repeated mentioning of sexual relationships.

The purpose of disclosure is to provide potential donors with information they consider relevant to their decision process. That some of donors will be persuaded not to donate by the information is a feature, not a bug.

Yeah, I was also unsure about this. In the original Google Doc, I had the comment:

I sure feel like this is going to confuse people, but not sure what else to do. Any help would be appreciated.

Highlighting the word metamour. If someone has a better word for that, or a good alternative explanation, that would be good.

I think I would eventually want this policy to be public and linked from our funds page, though it should probably go through at least one more round of editing before that.

Overall, I think if you want a good conflict of interest policy, you have to be concrete about things, which includes talking about things that in reality are often the sources of conflict of interest, which are pretty often messy romantic and sexual relationships and sharing weird intense experiences together. I don't know of a way to both be concrete here, and not be off-putting, while also actually addressing many of these potential sources for COIs.

+1 that requiring disclosure of past intimate relationships seems bad. Especially if the bar is lasting 2 weeks.

grantees might be discouraged from applying due to concerns about publicizing their personal lives

This is a good point. While I think the disclosure policy is correct, some mitigation of this can happen - it's probably good to mention in the policy that all public writeups get seen by grantees before being published, and they will not be blindsided by private information being published about them.

I can't imagine myself being able to objectively cast a vote about funding my room-mate / close friend / partner's boss / someone who I had a substantial romantic relationship with that ended 2 years ago (especially if the potential grantee is in a desperate financial situation!). I'm skeptical that humans in general can make reasonably objective judgments in such cases.

(emphasis added)

This isn't a point about the OP, but I thought I'd mention that I think humans can make these choices, if they have the required discipline and virtue, and I think in many situations we see that.

When you're the CEO of a successful company, you often have very close relationships with the 5-20 staff closest around you. You might live / have lived with them, work with them for hours every day, be good friends with them, etc. And many CEOs make sensible decisions about when to move these people around and fire them - it's not remotely standard practise to 'recuse' yourself from such decisions, you're the person with the most information about the person and how the organisation works, and if you actually care about those things enough and are competent enough to know your own mind and surround yourself with good people and a healthy environment, many CEOs are massively successful at making these decisions. I think this is true in other groups as well - I expect many people are pretty good at deciding e.g. if a close friend is unhealthy for them and that they want to cut ties.

I agree many people make quite unfortunate decisions here, but it is no iron law of psychology that 'humans in general' cannot make 'reasonably objective judgments' in such cases.

I can't imagine myself being able to objectively cast a vote about funding my room-mate

So, I think I agree with this in the case of small houses. However, I've been part of large group houses with 18+ people in it, where I interacted with very few of the people living in it, and overall spent much less time with many of my housemates than I did with only very casual acquaintances.

Maybe we should just make that explicit? Differentiate living together with 3-4 other people, from living together with 15 other people? A cutoff at something like 7 people seems potentially reasonable to me.

Yeah, I am not sure how to deal with this. Currently the fund team is quite heavily geographically distributed, with me being the only person located in the Bay Area, so on that dimension we are doing pretty well.

I don't really know what to do if there are multiple COIs, which is one of the reasons I much prefer us to err on the side of disclosure instead of recusal. I expect if we were to include friendships as sufficient for recusal, we would very frequently have only one person on the fund being able to vote on a proposal, and I expect that to overall make our decision-making quite a bit worse.

You could do early screening by unanimous vote against funding specific potential grantees, and, in these cases, no COI statement would have to be written at all.

Since we don't publicize rejections, or even who applied to the fund, I wasn't planning to write any COI statements for rejected applicants. That's a bit sad, since it kind of leaves a significant number of decisions without accountability, but I don't know what else to do.

The natural time for grantees to object to certain information to be included would be when we run our final writeup past them. They could then request that we change our writeup, or ask us to rerun the vote with certain members excluded, which would make the COI statements unnecessary.