CritPt
Harbor: pendingCritPt — AI evaluation benchmark listed by opencompass.
Verified71%Knowledge & Reasoning
- Samples
- 70
- Released
- 2025
- Scores
- 138
DrillBench
Autoregressive Drillhole Modelling Under Distribution Shift
A benchmark of 49,671 Western Australian drillholes for next-layer prediction and autoregressive stratigraphic generation across spatial and geological distribution shifts.
How to run
ltb · drillbench
$ npm i -g lettokenburn-cli$ ltb bench info drillbench
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
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini Pro | 59.7 | generative (hosted) | Independent | medium | Oct 1, 2026 | arxiv.org |
| GPT-6 Astra | 53.2 | generative (hosted) | Independent | medium | Oct 1, 2026 | arxiv.org |
| Qwen3 14B | 51.5 | likelihood + pmi | Independent | medium | Oct 1, 2026 | arxiv.org |
| Qwen3-32B | 48.5 | likelihood + pmi | Independent | medium | Oct 1, 2026 | arxiv.org |
| Phi-4 | 47.4 | likelihood + pmi | Independent | medium | Oct 1, 2026 | arxiv.org |
| Qwen3 8B | 35.2 | generative (bridge) | Independent | medium | Oct 1, 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
DrillBench comprises 49,671 mineral-exploration drillholes with layer-level and cell-level representations. It evaluates models on next-layer prediction and free-running generation tasks across a graded transfer spectrum from local to cross-province shifts.
Scoring involves benchmarking classical, geostatistical, and neural models to reveal a 'transfer boundary' where spatial conditioning degrades under shift while autoregressive models remain robust. The benchmark supports evaluating pretraining and retrieval strategies for generalization.