Add Linda-RAID submission

[Lendarixon]

Adds leaderboard/submissions/Linda-RAID/ (predictions.json for all 672,000 test ids + metadata.json).

Model: https://huggingface.co/Lindarixon/Linda-RAID

Detector: an ensemble of two models trained on a sample of the RAID train split (all 8 domains, 11 generators, 12 attacks; human texts of every attack labelled human), with harder strata oversampled (sampling without repetition penalty, paraphrase attack):

• DeBERTa-v3-large fine-tuned on ~590k RAID train texts (256-token windows covering the whole document, mean logit margin);
• a stylometric model (no neural network): ~230 hand-crafted style features + hashed character/word/function-word n-grams, LightGBM + logistic regression, trained on 200k RAID train texts (hard strata oversampled) on CPU.

Scores of the two models are z-normalized on human texts and averaged. Before scoring, input text is normalized against character-level attacks (homoglyphs, zero-width characters, unusual whitespace).

Local validation: 15% of RAID train source documents were held out (no text of those documents was used in training); 1,928 clean human texts + 40,000 AI texts from them, scored with raid.evaluate.run_evaluation (per-domain thresholds, FPR 5%): 99.7% accuracy with adversarial attacks, 99.6% without (DeBERTa-v3-large alone 98.4%, stylometric model alone 98.4%).

🤖 Generated with Claude Code

[github-actions]

Eval run succeeded! Link to run: link

Here are the results of the submission(s):

Linda-RAID

Release date: 2026-09-27

I've committed detailed results of this detector's performance on the test set to this PR.

On the RAID dataset as a whole (aggregated across all generation models, domains, decoding strategies, repetition penalties, and adversarial attacks), it achieved an AUROC of 99.83 and a TPR of 99.48% at FPR=5% and 97.66% at FPR=1%.
Without adversarial attacks, it achieved AUROC of 99.89 and a TPR of 99.65% at FPR=5% and 98.46% at FPR=1%.

If all looks well, a maintainer will come by soon to merge this PR and your entry/entries will appear on the leaderboard. If you need to make any changes, feel free to push new commits to this PR. Thanks for submitting to RAID!