{"model":{"version":"survival-2026-08-11","horizon_hours":6,"trained_rows":1022,"trained_through":1786391694817,"train_base_rug_rate":0.7162,"validation":{"method":"walk-forward, 6 expanding chronological folds, no shuffling","pooled_n":672,"pooled_auc":0.9541,"worst_fold_auc":0.88,"previous_scorer_auc":0.545}},"the_finding":"On Base, the standard contract-safety heuristics are INVERTED. Across 1,022 graded tokens, 'ownership renounced' tokens rugged 77.5% of the time while NOT-renounced rugged 4.8%; tokens carrying 5+ green flags rugged 85% against 25-63% for those with 4 or fewer; and every proxy, mintable, vanity-address and prior-honeypot-creator token in the set survived. The cause is structural: Base launchpads auto-renounce, auto-verify and auto-lock everything they mint, so a full sweep of green flags is the fingerprint of a disposable memecoin, not evidence of safety.","why_it_matters":"Any scanner that scores 'renounced = safer' is selling the inverse of the outcomes on this chain. We shipped that same inversion until 2026-08-11 and our own public ledger is what caught it.","horizon_warning":"This model predicts survival over 6 HOURS and nothing longer. We sampled tokens this ledger graded 'fine' at 6h and re-checked them two weeks later: 12 of 12 were dead, and even among the deepest-liquidity clean calls only 2 of 12 survived. Almost everything launched on Base dies. The answerable question is whether the pool outlasts your position, which is what this scores.","calibration_out_of_sample":[{"predicted_rug_prob":"0-0.1","n":65,"actually_rugged_pct":3.1,"actually_survived_pct":96.9},{"predicted_rug_prob":"0.1-0.25","n":25,"actually_rugged_pct":28,"actually_survived_pct":72},{"predicted_rug_prob":"0.25-0.5","n":70,"actually_rugged_pct":55.7,"actually_survived_pct":44.3},{"predicted_rug_prob":"0.5-0.75","n":43,"actually_rugged_pct":53.5,"actually_survived_pct":46.5},{"predicted_rug_prob":"0.75-0.9","n":25,"actually_rugged_pct":72,"actually_survived_pct":28},{"predicted_rug_prob":"0.9-0.98","n":238,"actually_rugged_pct":97.5,"actually_survived_pct":2.5},{"predicted_rug_prob":"0.98-1.01","n":206,"actually_rugged_pct":100,"actually_survived_pct":0}],"verify_it_yourself":{"raw_labelled_data":"/track-record/raw","grading_formula":"/track-record","note":"Every row carries the feature vector we scored on and the outcome we later measured. Refit it and check."},"as_of":"2026-09-06T01:07:58.718Z"}