AgentDoxx
Harbor: pendingAn evaluation suite of 822 synthetic interview transcripts grounded in public info to measure LLM re-identification risks via web search.
- Samples
- 822
- Released
- 2026
- Scores
- 9
Ego-SpaCR
Spatially grounded Conversational Reasoning (SpaCR) benchmark for Egocentric Assistants
Egocentric video QA benchmark with 620 sessions, 95 conversations, and 3,100 queries measuring spatially grounded conversational reasoning.
How to run
ltb · ego-spacr
$ npm i -g lettokenburn-cli$ ltb bench info ego-spacr
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 | |||
|---|---|---|---|---|---|---|
| Gemini 3.7 Flash | 41.4 | native video | Independent | medium | Oct 4, 2026 | arxiv.org |
| GPT-5.6-terra | 38.5 | uniform | Independent | medium | Oct 4, 2026 | arxiv.org |
| GLM-5.3 Flash | 27.6 | conversation only | Independent | medium | Oct 4, 2026 | arxiv.org |
| Qwen3.8 27B | 22.3 | native video | Independent | medium | Oct 4, 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
Comprises 620 ScanNet video sessions augmented with 95 task-oriented dialogues and 3,100 evaluation queries requiring cross-scene, recall-oriented, and counterfactual spatial reasoning.
Evaluated via answer accuracy; proposed SpaC-MEM method outperforms baselines in accuracy and object recall while reducing token usage.