Add detector: GPTTrace (Ettin-150M)

[andrewnakas]

Adds leaderboard/submissions/gpttrace-ettin150/ (predictions.json + metadata.json) for all 672,000 test ids.

Detector. jhu-clsp/ettin-encoder-150m (MIT) fine-tuned as a 2-class sequence classifier for one epoch on ~680k RAID-train rows: every human text plus a capped sample per (domain, model, decoding, repetition penalty, attack) cell, with paraphrased AI texts seen twice. Max 512 tokens, lr 5e-5, cosine schedule, 2×T4 fp16.

Input. Before tokenisation, invisible format characters (Unicode Cf, incl. zero-width spaces) are removed and Cyrillic/Greek lookalike letters are mapped to Latin in mostly-Latin text. No whitespace or case changes.

Score. softmax P(machine) from the fp16 model. (It exports to a 151 MB int8 ONNX for in-browser use; it is a research model and is not yet the detector deployed on gpttrace.com.)

🤖 Generated with Claude Code

[github-actions]

Eval run succeeded! Link to run: link

Here are the results of the submission(s):

GPTTrace (Ettin-150M)

Release date: 2026-10-10

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 99.89 and a TPR of 99.73% at FPR=5% and 98.64% at FPR=1%.
Without adversarial attacks, it achieved AUROC of 99.96 and a TPR of 99.90% at FPR=5% and 99.59% 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!

[liamdugan]

Hi @andrewnakas congrats on the strong performance. Please let me know if you'd like me to merge this onto the leaderboard.

[andrewnakas]

Hey! Thanks @liamdugan, yeah I'd love to be merged onto the leaderboard. I'm going to keep on pushing on this and get this as good as I possibly can!

[liamdugan]

Sounds good! I'll merge you in then. Thanks for submitting to RAID!

In the future if you'd like to update the scores of this system feel free directly edit the predictions.json file in a PR. If you'd like to submit a second system to the leaderboard then you can also do that, just make a brand new folder in the /submissions subdirectory.