How to comment

This is a draft. We’d like it checked, criticized and improved.

We’d like comments from people who know this literature, from people at organizations that use models like this to allocate money, and from anyone who tries the tool and finds it confusing.

Ways to comment

  1. Hypothes.is, for anything specific. Open the sidebar with the < tab at the top right of any page, select the sentence, number or table cell you want to comment on, and add an annotation. Annotations are public by default and anyone can reply, so don’t post anything confidential. A free account takes about a minute.
  2. Email contact@unjournal.org, for longer points or anything you’d rather not post publicly.
  3. Submit a scenario. If you’ve set up values you think are defensible, send them from the share section of the model page with a few words on why. With your permission we’ll add it to the shared scenarios.

We read the annotations and reply in the same thread. Replies drafted by our AI assistant start with “Claude:”; when a comment leads to a change, the reply says what changed.

What we’d most like feedback on

  1. The replication. Does the replication page reproduce the report correctly? We match its Table 3 and closed forms, but only after adopting conventions from the companion notebook that the PDF doesn’t state. Three summary sentences in the report don’t match its own tables; we may have misread them.
  2. The “beyond the paper” switches. Are these sensible ways to formalize the evaluator’s points? We’re least sure about the rebound form (switch 3) and the Monte Carlo defaults (switch 2), which have no empirical basis.
  3. The Rethink Priorities 2026 preset. It approximates the central values in RP’s current model of global catastrophic risk funds, using this report’s simpler structure. Does it represent their model fairly? The departures page lists what it leaves out.
  4. Presentation. What’s confusing, what should come first, and what would someone deciding where to give need to see that isn’t here? In particular: can you tell, from the source tags under the results, which numbers come from the report, which from Rethink Priorities, and which are our illustrative choices? Do the quick-start questions mean what their answers do?
  5. Better inputs. For any setting tagged “illustrative”, what evidence would you use instead? We’d especially like sources for how long an intervention’s effect on risk lasts, and for the chance that it backfires.
  6. Anything we’ve got wrong about the report, the evaluation, Rethink Priorities’ models, or the related work on the Context page.

Who built this and how it’s checked

The Unjournal built this tool with an AI coding assistant (Claude), working from the report, its companion notebook and our evaluation. Automated tests check the engine against the notebook’s printed output, and we spot-checked ten Table 3 cells by hand against the PDF. No outside expert has reviewed it yet, and the report’s author hasn’t reviewed or endorsed it.