600 valid runs.

Nested polynomial interactions and sigmoid transforms feed a ridge-GCV head; the receipt describes the executable structure without assigning a causal explanation.

Dataset

174 train · 35 test · 7 features

Search

17.578s

Finalist CV

1.183s

Fit + predict

0.480s

Total workflow

19.527s

Strongest stored foundation

TabFM Ensemble · 5.13× lower RMSE

(pipe X (poly_interact append=true) (sigmoid append=true) (poly_interact append=true) (sigmoid append=false) (sigmoid) (sigmoid) (ridge))

Limitation. The ablation is post-hoc and the small held-out split does not establish causal mechanism or prospective frequency.

Foundation comparator wall-clock was not receipted. The displayed clocks are candidate workflow only.

A Gaussian-process regressor combines radial-basis and periodic components for the two-feature table; the match is case-specific, not a climate claim.

Dataset

250 train · 50 test · 2 features

Search

15.955s

Finalist CV

1.677s

Fit + predict

1.224s

Total workflow

19.614s

Strongest stored foundation

TabFM · 3.40× lower RMSE

(pipe X (gpr kernel=rbf*periodic m=300 steps=40))

Limitation. This is an inductive-bias match on one encoded table, not a climate-modeling claim.

Foundation comparator wall-clock was not receipted. The displayed clocks are candidate workflow only.

A compact ridge-GCV calibration program is the recorded quality outlier; its longer complete workflow keeps it outside the fast focal pair.

Dataset

442 train · 89 test · 101 features

Search

20.440s

Finalist CV

19.012s

Fit + predict

0.563s

Total workflow

100.180s

Strongest stored foundation

TabFM Ensemble · 4373.89× lower RMSE

(pipe X (ridge_gcv))

Limitation. Diagnostics are consistent with a low-dimensional calibration structure; they do not establish a universal spectral-learning mechanism.

Foundation comparator wall-clock was not receipted. The displayed clocks are candidate workflow only.