Interactive views of what the pipeline produced — permutation importance, real-data backtest errors, per-category value, and the 12-month tier forecast. Hover any chart for exact numbers.
Feature Importance
Permutation importance shuffles each feature and measures MAE lift. expected_price dominates — every other feature moves the error less than $0.02.
MAE by Model × Category
Mean Absolute Error per category. Lower is better. The Naive baseline (predict expected_price) sits within cents of the best ML model — most of the apparent skill is the linear trend, not the learner.
Real-Data Backtest (out-of-time)
MAE on the last three months of real HardwareDealsCo prices — values the model never saw during training. This is the credible accuracy claim.
Value Metrics by Category
Median price-per-unit for 2025 components. Boxes span the 25-75th percentile. Lower bars are better value.
Hardest Predictions
Components where the model errs the most. Largely enterprise SSDs and legacy hardware with thin training data — the long tail.
12-Month Build Cost Forecast by Tier
Total build cost projected forward across budget, mid, and high tiers. Each tier picks one canonical component per category.