AfterDeleteBench: Auditing Post-Deletion Information Persistence in LLM Agents
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
AfterDeleteBench is a behavioral audit showing that deleting an original memory record from a fixed-weight LLM agent often fails to remove the information, because derived artifacts such as summaries can still carry it.
Score
0.66total under rubric v2.0
- Significance0.59
- Reader interest0.73
- Audience breadth0.52
- Practical relevance0.69
- Substantiated0.78
- Technical depth0.74
- New information0.90
- Counts against: Promotional0.04
- 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 | 0.64 | 0.40 | 0.72 | 0.59 | 0.32 | +0.25 | +0.147 |
| Reader interest | 0.68 | 0.60 | 0.90 | 0.73 | 0.30 | +0.20 | +0.145 |
| Audience breadth | 0.61 | 0.40 | 0.55 | 0.52 | 0.21 | +0.10 | +0.052 |
| Practical relevance | 0.76 | 0.60 | 0.72 | 0.69 | 0.15 | +0.15 | +0.104 |
| Substantiated | 0.91 | 0.60 | 0.82 | 0.78 | 0.31 | +0.10 | +0.078 |
| Technical depth | 0.77 | 0.60 | 0.86 | 0.74 | 0.26 | +0.10 | +0.074 |
| New information | 0.92 | 0.80 | 0.97 | 0.90 | 0.17 | +0.10 | +0.090 |
| Promotional | 0.05 | 0.05 | 0.01 | 0.04 | 0.04 | −0.15 | −0.006 |
| Overclaim | 0.16 | 0.30 | 0.12 | 0.19 | 0.18 | −0.15 | −0.029 |
| Sum of contributions | +0.655 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.655 | ||||||
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, 08:54 UTC. It played no part in the ranking.