common_voice
Harbor: pendingAvailable in EleutherAI lm-evaluation-harness as task family common voice (1 tasks).
Incomplete14%Multimodal
- Samples
- N/A
- Released
- N/A
- Scores
- 0
Common-TSQA
A unified evaluation benchmark for time-series QA systems that aggregates public datasets to assess evidence use and reasoning reliability under interventions.
How to run
ltb · common-tsqa
$ npm i -g lettokenburn-cli$ ltb bench info common-tsqa
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 | |||
|---|---|---|---|---|---|---|
| ChatTS | 89.0 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| TimeOmni-1 | 88.5 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| TimeOmni-VL | 81.8 | N/A | Independent | medium | Oct 5, 2026 | arxiv.org |
| Time-MQA | 62.6 | 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
Common-TSQA collects and unifies public time-series evaluation datasets into a common format with standardized task definitions and answer schemas. It evaluates systems through their native interfaces using original conditions and six numerical interventions.
Scoring uses task-specific criteria to measure performance stability and rationale faithfulness. Key findings reveal that aggregate accuracy often masks reliance on non-series cues and that rationales frequently contain unsupported numerical claims or invalid inferences.