What Pairwise Evaluation Misses: Reward Scores across Prompts
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
This paper proposes a human-referenced audit of reward models, testing how much their scores vary between prompts compared with human ratings of the same responses.
Score
0.70total under rubric v2.0
- Significance0.66
- Reader interest0.70
- Audience breadth0.53
- Practical relevance0.67
- Substantiated0.90
- Technical depth0.93
- New information0.90
- Counts against: Promotional0.03
- Counts against: Overclaim0.19
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| Significance | 0.68 | 0.55 | 0.74 | 0.66 | 0.19 | +0.25 | +0.164 |
| Reader interest | 0.61 | 0.60 | 0.88 | 0.70 | 0.28 | +0.20 | +0.139 |
| Audience breadth | 0.51 | 0.55 | 0.52 | 0.53 | 0.04 | +0.10 | +0.053 |
| Practical relevance | 0.69 | 0.60 | 0.72 | 0.67 | 0.12 | +0.15 | +0.101 |
| Substantiated | 0.99 | 0.80 | 0.91 | 0.90 | 0.19 | +0.10 | +0.090 |
| Technical depth | 0.99 | 0.85 | 0.94 | 0.93 | 0.14 | +0.10 | +0.093 |
| New information | 0.90 | 0.80 | 0.99 | 0.90 | 0.19 | +0.10 | +0.090 |
| Promotional | 0.04 | 0.03 | 0.01 | 0.03 | 0.03 | −0.15 | −0.004 |
| Overclaim | 0.24 | 0.25 | 0.08 | 0.19 | 0.17 | −0.15 | −0.028 |
| Sum of contributions | +0.697 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.697 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN —
Recorded for context only, checked 24 Oct 2026, 08:53 UTC. It played no part in the ranking.