**EML's predicted native home:** recover f(x) from (xᵢ, yᵢ) samples. ## Sub-goals - [ ] **9.0** SR task spec: numeric samples → expression in-grammar; synthetic benchmark from v1/v2 corpus (known ground truth) + a restricted Feynman-style subset - [ ] **9.1** Search over expression space guided by the shared encoders — EML-space search vs AST-space search (the SR version of the representation question) - [ ] **9.2** Baselines: PySR/GP reference, transformer-SR - [ ] **9.3** Metrics: exact-recovery rate, accuracy–complexity Pareto, wall-clock; analysis + report - [ ] **9R** Reserved repair pass ## Gate G9 Exact recovery on the synthetic benchmark above the GP baseline at matched budget, **or a documented negative**.
EML's predicted native home: recover f(x) from (xᵢ, yᵢ) samples.
Sub-goals
Gate G9
Exact recovery on the synthetic benchmark above the GP baseline at matched budget, or a documented negative.