Add kp-detect-v20 submission#159
Conversation
|
Eval run succeeded! Link to run: link Here are the results of the submission(s): kp-detect-v20Release date: 2026-05-25 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 95.31 and a TPR of 86.77% at FPR=5% and 70.87% 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! |
1 participant
kp-detect-v20 — Kareem Elsamadicy (Independent Researcher)
Release date: 2026-05-25
Detector: kp-detect-v20
Summary
Refines v15 baseline on RAID test rows by combining v15 score with an ensemble of 3,445 'pure-true' proteins — three-atom AND-products that have been validated to have precision = 100% on RAID training data (never voted wrong yet). These voters were filtered from 1M random candidates using a strict falsification rule (Popper-style: keep only proteins where every doc that fires the protein has the same target class; drop the moment a single counter-example fires).
Train OOF (5-fold, balanced n=27,900)
Expected lift on RAID test: modest AUROC lift over v15 baseline (96.44 → ~97), with substantial log_loss reduction from the strict-precision voter ensemble.
Contact: kelsamadicy@gmail.com