Checklist ONIA 2026 Team Selection Camp (Extended Squad) – Test 2 · Task 1
The Archive of Contradictions
Romanian title: Arhiva contradicțiilor
Classify factual claims as SUPPORTS, REFUTES or NOT ENOUGH INFO by retrieving evidence from a Wikipedia-derived corpus.
The task
A fact-checking agency needs an automatic system that, for each public claim, searches an encyclopaedic archive and decides whether the claim is supported, contradicted, or cannot be verified with the available information.
The contestant receives labelled claims (train.csv), unlabelled claims (public_test.csv) and a corpus of Wikipedia documents (wiki_pages.csv with doc_id, text and indexed sentences). Evidence must be retrieved automatically from the corpus; the gold labels or evidence of the original FEVER benchmark may not be used.
Abridged and translated by SOTA from the official Romanian materials. The official statement has the exact rules, and it wins wherever this summary differs.
At a glance
- You get
train.csv(id, claim, label),public_test.csv(id, claim),wiki_pages.csv(doc_id, text, lines); the example claims in the statement are in English. Dataset archive of 134,887,721 bytes.- You submit
submission.csvwith id, subtaskID (always 1) and label, covering every id ofpublic_test.csvexactly once.- Scoring
- Macro F1 over the three classes; 100 points if Macro F1 ≥ 0.6, 0 if ≤ 0.3, otherwise floor((F1 − 0.3) / 0.3 × 100).
- Rules
- On-site, Timișoara (West University of Timișoara and Politehnica University of Timișoara)
- No internet access; only preinstalled offline resources (NLTK, spaCy
en_core_web_sm, scikit-learn, Gensim glove-wiki-gigaword-50, fastText without pretrained models, and the Hugging Face models distilbert-base-uncased, MoritzLaurer DeBERTa-v3-large-mnli-fever-anli-ling-wanli, roberta-large-mnli, all-MiniLM-L6-v2 and whisper-small) - A GPU Jupyter environment hosted by UVT was provided during the tests
- Problem statements unlocked with a password handed out in the room
- Format
- ONIA 2026 Lot Lărgit, test 2, 24 May 2026 (06:25–11:30 UTC on the platform).