Aider Polyglot
Harbor nativeAider Polyglot: 6种语言diff编辑
Verified71%Coding
- Samples
- 225
- Released
- 2024
- Scores
- 61
From Triage to Impact
From Triage to Impact: A Multi-Hop Reasoning Benchmark for Explainable Persian Hate Speech
A 1,200-instance benchmark for explainable Persian hate speech analysis using a seven-hop reasoning framework covering triage, target identification, tactics, moral foundations, evidence, narrative, and severity.
How to run
ltb · from-triage-to-impact
$ npm i -g lettokenburn-cli$ ltb bench info from-triage-to-impact
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
8 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Grok-4.6 | 82.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| GPT-5.2-chat | 79.7 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Claude Sonnet 4.6 | 79.2 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| o3 | 77.8 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| DeepSeek R1 | 75.0 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| DeepSeek-V3 | 68.5 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Qwen3 Max (2025-09-23) | 66.4 | N/A | Independent | medium | Oct 14, 2026 | arxiv.org |
| Qwen3 Inst | 48.0 | N/A | Independent | medium | Oct 14, 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
The dataset contains 1,200 Persian social media posts from X and Instagram, annotated via a seven-hop reasoning pipeline including content triage, target identification, and severity assessment. Annotations were generated by Gemini 3.5 Flash and validated by human annotators.
Scoring evaluates LLM performance across reasoning hops using task-specific metrics like accuracy and F1. Proprietary models generally outperform open-weight ones, and intermediate reasoning steps significantly improve severity prediction compared to one-shot assessment.