Step 1: rank suspects
Run both scans; list every high finding plus the top 5 features by importance/SHAP. A feature is a suspect if any of these hold:
- single-feature AUC > 0.9 or far above domain expectations;
- outcome-like name, or comes from a system that acts on outcomes (CRM cases, collections, refunds);
- built from a table without history (current-state);
- its importance is much higher than the business can explain.
Step 2: establish the write time
For each suspect, answer with the data owner:
| Question | Leak if |
|---|---|
| Which process writes the field? | a process triggered by the outcome |
| When relative to the event? | at or after the event |
| Is it overwritten? | yes, and you read the current value |
| Is it backfilled? | yes, historical rows were filled later |
Step 3: look at the timeline
Plot the feature against days before the outcome for positives, and against days before the scoring date for negatives:
pos = df[df.y == 1].assign(days_before=(df.event_date - df.feature_date).dt.days)
pos.groupby(pos.days_before.clip(-30, 90))["feature"].mean().plot()Values that change sharply at day 0 or after are written by the outcome.
Step 4: rebuild as-of and compare
Recompute the suspect strictly from data before the prediction time and compare:
| Version | Single-feature AUC | Model AUC (time holdout) |
|---|---|---|
| As delivered | 0.95 | 0.97 |
| Rebuilt as-of | 0.62 | 0.81 |
A collapse confirms leakage; the rebuilt version is the one to keep (if it still helps).
Step 5: ablation on a time-ordered holdout
Drop the feature (or its group), retrain, evaluate on the latest period. Leaky features often add a lot on random CV and little on the future window.
Step 6: shadow comparison
In shadow mode, log the feature's value at prediction time and compare with the value the training pipeline produces for the same entity and timestamp. Any difference is train/serve skew, often caused by leakage.
Step 7: fix and guard
- Remove or rebuild the feature; add it to a
KNOWN_LEAKSlist in the training code with a comment. - Add
leakage_scan.py(data) andcode_leak_lint.py(code) to CI:yaml - run: python scripts/leakage_scan.py data/train_sample.csv --target churned --time-col score_date --drop customer_id - run: python scripts/code_leak_lint.py src/train.py - Data contracts: the owning team documents write times and history semantics for each source.
Decision table
| Evidence | Decision |
|---|---|
| Written after the event | drop |
| Current-state value, history available | rebuild as-of |
| Current-state value, no history | drop, or start snapshotting now and add later |
| Proxy of the label but available at prediction time | keep only if acting on it is legitimate (e.g. "customer requested cancellation" when the decision is a save offer) |
| Pipeline leak | fix the code; re-run evaluation |
anomaly-detection-model-builder
Build and evaluate anomaly detection for tabular, sensor and transaction data with few or no labels: define the anomaly type (point, contextual, collective), choose detectors (robust z-score/MAD and IQR rules, Mahalanobis with robust covariance, k-NN distance, LOF, Isolation Forest, one-class SVM, PCA reconstruction / autoencoders, forecasting residuals for time series), combine them by rank, explain each alert by its deviating features, set thresholds from an alert budget or the score distribution, and evaluate with precision@k, PR-AUC and recall on known incidents, or detector agreement when no labels exist. Includes pure-Python detectors (robust z, Mahalanobis, kNN, Isolation Forest, ensemble) and an evaluation script.
catboost-model-builder
Build CatBoost models that use its strengths correctly: raw categorical columns as cat_features (ordered target statistics, no manual encoding), text_features, Pool construction with NaN-safe categories, ordered boosting and has_time for temporal data, symmetric (oblivious) trees, overfitting detector with eval_set and use_best_model, losses for classification, regression (Quantile, Poisson, Tweedie) and ranking (YetiRank), class weights, monotone constraints, feature importance (PredictionValuesChange, LossFunctionChange) and SHAP values, GPU training and model export. Plans cat/text/numeric columns from a CSV and generates a runnable training script, and checks parameter sets for the conflicts CatBoost rejects (iterations vs n_estimators, Bayesian bootstrap with subsample, multiple class-weight options, depth > 16, use_best_model without eval_set).
categorical-encoding-selector
Choose and implement the right encoding for every categorical column given the model family and the data: one-hot (with infrequent-level grouping and handle_unknown), ordinal with an explicit order, cross-fitted target encoding (scikit-learn TargetEncoder, CatBoost ordered statistics, category_encoders inside CV), frequency/count encoding, feature hashing, native categorical support in LightGBM, HistGradientBoosting, XGBoost and CatBoost, prefix levels for codes (zip3, NAICS-2), and embeddings. Measures cardinality, rare-level share, unseen-level rate on a time-ordered holdout and target signal per column, emits the ColumnTransformer, and quantifies target-encoding leakage (in-sample vs out-of-fold vs holdout AUC).
classification-model-selector
Choose the right classification algorithm for tabular, text or mixed data and prove the choice with a fair bake-off: logistic regression (L2, L1, elastic net), linear SVM, naive Bayes, k-NN, RBF SVM, decision trees, random forest, extra trees, HistGradientBoosting, LightGBM, XGBoost, CatBoost, explainable boosting machines (EBM/GAM), points scorecards and MLPs, matched to data size, width, categorical and high-cardinality share, missingness, class imbalance, text, and constraints such as interpretability, monotonic constraints, calibrated probabilities, latency budgets and tiny deployable models. Ranks candidates with starting hyperparameters and cautions, then generates a bake-off script running every candidate on identical CV folds with per-fold metrics, fit time and prediction latency.
clustering-algorithm-selector
Choose the right clustering algorithm and check that clustering makes sense at all: cluster tendency (Hopkins statistic with a correlated-features caveat, nearest-neighbour isolation), then k-means / MiniBatchKMeans, Gaussian mixtures with BIC, agglomerative (Ward, average, complete, single) with dendrograms, DBSCAN, HDBSCAN, OPTICS, spectral clustering, BIRCH, mean shift, k-prototypes and k-modes for mixed or categorical data, and Gower distance, matched to data size, expected shapes, noise, varying density, known k, soft memberships, hierarchy and streaming. Covers preprocessing (scaling, encoding, dimensionality, distance metrics), validation (internal indices, bootstrap stability, external labels, business profiling) and assigning new points.
collaborative-filtering-implementer
Implement collaborative filtering recommenders that beat popularity: item-based kNN with cosine, Jaccard or asymmetric similarity, shrinkage, neighbourhood size, positive-only or mean-centred rating signals and 'because you liked X' explanations; implicit ALS with confidence weighting, BPR pairwise ranking and the closed-form EASE model in plain NumPy, tuned on a temporal validation split by recall@k and NDCG@k and refit on all data; plus user-kNN, explicit matrix factorisation, the implicit/LightFM libraries, hyperparameter ranges, cold-start fallbacks, scaling with sparse matrices and approximate nearest neighbours. Produces top-k lists in the format of the recommendation evaluator.