Making Flawed Benchmarks Useful
Comments on “Making Flawed Benchmarks Useful”
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
Blatent researchers examine why open benchmark scoring can misjudge correct model answers, citing cases like PHYBench's algebraically equivalent expressions and QASPER's partial-credit penalties. They describe how exact-match and token-F1 metrics miss semantic equivalence, and propose checking exact-source extraction deterministically before using an LLM judge to assess correctness.
Why it ranked
No. 118 of 139 with a score of 0.57. The panel rated it highest on new information (0.77) and reader interest (0.69), lowest on significance (0.52). The judges disagreed most on new information (spread 0.45), so the story is marked contested. Placed No. 118, No. 116 by score alone: the 47th AI & machine learning story of the day (−0.080). The placement terms move stories; they are not part of the score.
- AI judgeJudge 2claude-haiku-5-5
- Practical engineering writeup on evaluating LLMs with flawed benchmarks; relevant to AI builders, with partly promotional framing for its own product and a weakly substantiated LLM-grader approach.
- AI judgeJudge 3gpt-5.6-luna
- A substantive evaluation write-up showing how lexical benchmark scoring mismeasures extraction quality, with concrete examples and a proposed source-fidelity plus semantic-judging method. Evidence is useful but methodology and validation remain only moderately detailed.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.57total under rubric v2.3Contestedjudges differ by 0.45 on new information
- Significance0.52
- Reader interest0.69
- Audience breadth0.60
- Practical relevance0.69
- Substantiated0.66
- Technical depth0.67
- New information0.77
- Counts against: Promotional0.25
- Counts against: Overclaim0.20
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) |
|---|---|---|---|---|---|---|---|
| SignificanceHow much does this change what people in tech know or can do? | 0.59 | 0.30 | 0.68 | 0.52 | 0.38 | +0.24 | +0.126 |
| Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on? | 0.61 | 0.60 | 0.86 | 0.69 | 0.26 | +0.26 | +0.179 |
| Audience breadthHow many people who build technology does it concern? | 0.65 | 0.40 | 0.76 | 0.60 | 0.36 | +0.13 | +0.078 |
| Practical relevanceHow useful is it to someone who builds software or tech products? | 0.68 | 0.60 | 0.78 | 0.69 | 0.18 | +0.22 | +0.151 |
| SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations. | 0.71 | 0.50 | 0.78 | 0.66 | 0.28 | +0.05 | +0.033 |
| Technical depthHow technically substantive is it? | 0.78 | 0.60 | 0.64 | 0.67 | 0.18 | +0.05 | +0.034 |
| New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)? | 0.86 | 0.50 | 0.95 | 0.77 | 0.45 , above the contested threshold | +0.05 | +0.039 |
| PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional. | 0.07 | 0.45 | 0.22 | 0.25 | 0.38 | −0.19 | −0.047 |
| OverclaimDo the headline or conclusions claim more than the body supports? | 0.18 | 0.35 | 0.08 | 0.20 | 0.27 | −0.11 | −0.022 |
| Sum of contributions | +0.570 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.570 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-11-05/118/"><img src="https://rubric.hrih.si/badge/2026-11-05/118.svg" alt="Ranked #118 on Rubric, 5 Nov 2026" width="245" height="28"></a>Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN 1 pts, not on front page
Recorded for context only, checked 5 Nov 2026, 05:46 UTC. It played no part in the ranking.