AgentMonBench
Harbor: pendingA software-engineering benchmark evaluating monitors' ability to identify consequential agent decisions and locate evidence for user oversight across three subsets.
- Samples
- 300
- Released
- 2026
- Scores
- 8
DesignReviewBench
A benchmark of 72 Linux kernel patch series and 79 reference design trade-offs for evaluating design-aware code review capabilities.
How to run
ltb · designreviewbench
$ npm i -g lettokenburn-cli$ ltb bench info designreviewbench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
4 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| DeepSeek V4.1 Flash | 0.3 | designlore | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT-5.6-terra | 0.3 | designlore | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT-5.6 Luna | 0.3 | designlore | Independent | medium | Oct 15, 2026 | arxiv.org |
| Claude Haiku 4.5 | 0.2 | designlore | Independent | medium | Oct 15, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
Derived from real Linux kernel patch-review histories, comprising 72 patch series paired with codebase snapshots and 79 reference design trade-offs. It evaluates the ability to identify design trade-offs rather than just functional defects.
Scoring uses F1 score for design trade-off detection. The strongest baseline (Sashiko) achieved an F1 of 0.213, indicating the task remains challenging for current agents.