Struct DmlPartialLinearReference 

pub struct DmlPartialLinearReference {
    pub available: bool,
    pub backend: String,
    pub theta: f64,
    pub se: f64,
    pub ci_lo: f64,
    pub ci_hi: f64,
}

A Double Machine Learning (DML) reference estimate of the average linear effect θ = E[∂E(Y|D,X)/∂D] of a treatment/dose D on outcome Y after partialling out confounders X, computed by a mature Python DML library (DoubleML’s partially-linear model, with EconML’s LinearDML as fallback).

This is the Neyman-orthogonal scalar-target baseline used by #461’s Sim C: the cross-fitted DML estimator is, by construction, first-order insensitive to first-stage nuisance estimation error, so its theta/se are the reference bias/coverage that gam’s orthogonalized marginal-slope target θ = E_x[β(x)] must match-or-beat under x-dependent Stage-1 miscalibration.

Fields

available: bool

Whether a DML library was importable in the reference interpreter. When false, theta/se/ci_lo/ci_hi are NaN and the caller should emit a clear skip message rather than asserting against them — DoubleML/ EconML are heavier optional dependencies than scipy/mgcv, so their absence is treated as a genuine environmental gate (mirroring the CUDA-only skip in tests/common/gpu_gate.rs) rather than the hard failure that a missing scipy/R would be.

backend: String

Which backend produced the estimate: “doubleml”, “econml”, or “none”.

theta: f64

Point estimate of the average linear treatment effect θ.

se: f64

Standard error of θ̂ reported by the DML library.

ci_lo: f64

Lower end of the library’s 95% confidence interval for θ.

ci_hi: f64

Upper end of the library’s 95% confidence interval for θ.