Lens · manifund-goals
importance-for-epistemic-public-good
importance of this evaluation criterion to a grantmaker whose aim is producing epistemic public goods: research, data, and tools that improve collective understanding
| Rank | Entity | Judged text | Latent score | × median | Percentile | Evidence |
|---|---|---|---|---|---|---|
| 1 | epistemic-integrity | Epistemic integrity of the write-up — honest failure modes, quantified claims, falsifiable milestones; whether the proposal reasons transparently rather than selling. | 0.483 ± 0.538 | 1.27× | 87.5% | |
| 2 | theory-of-change-plausibility | Plausibility of the causal path from activities to claimed impact — whether the mechanism connecting what the team will do to the outcome they promise survives scrutiny step by…Plausibility of the causal path from activities to claimed impact — whether the mechanism connecting what the team will do to the outcome they promise survives scrutiny step by step. | 0.483 ± 0.371 | 1.27× | 62.5% | |
| 3 | team-track-record-evidence | Verifiable track-record evidence the team can execute — concrete, checkable prior work demonstrating the team has shipped comparable things before. | 0.000 ± 0.386 | 0.79× | 37.5% | |
| 4 | impact-per-marginal-dollar | Expected impact per marginal dollar at the stated ask — how much good the next dollar of funding actually buys, given the project size, cost structure, and counterfactual fundin…Expected impact per marginal dollar at the stated ask — how much good the next dollar of funding actually buys, given the project size, cost structure, and counterfactual funding landscape. | 0.000 ± 0.538 | 0.79× | 12.5% |
How to read the score: it is a natural-log ratio. The difference between any two rows, exponentiated, is the ratio the judging model is expected to report between them — × median reads each entity against the field's midpoint. ± is the posterior standard deviation on the same scale, drawn as the band under each score; where two bands overlap, the order between those rows is not settled.
Run metadata
1 run- Model
- anthropic/claude-haiku-4.5
- Temperature
- 0
- Seed
- 2550563098764197778
- Comparisons
- 32 of 32 budgeted
- Stop reason
- budget_exhausted
- Top-k error
- 1.12
- Cost
- $0.07
- Harness
- cardinal-harness 0.9.0
- Scored
- Aug 20, 2026, 3:37 AM