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
SEMAADB
Systems Engineering Modeling Assistant with AI Dataset and Benchmark
A benchmark of 100 human-verified SysML contexts (5 diagrams each) measuring semantic repair and cross-diagram consistency for LLMs.
How to run
ltb · semaadb
$ npm i -g lettokenburn-cli$ ltb bench info semaadb
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
3 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-5.6 Sol | 80.7 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| GPT-5.6-terra | 77.5 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| GPT-5.6 Luna | 73.0 | N/A | Independent | medium | Oct 5, 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
SEMAADB comprises 3,000 engineering contexts with 15,000 connected SysML diagrams. A subset of 100 contexts is human-verified to form the benchmark test set, ensuring coherence across Requirement, Block Definition, Activity, State Machine, and Sequence views.
It evaluates models on diagram repair and cross-diagram update tasks. The best model achieved 64.3% semantic error repair and 80.7% F1 in change propagation, indicating that while syntax repair is solved, semantic consistency remains challenging.