Cross-family bivariate models are a development route for a later
release, not a drmTMB 0.6.0 analysis workflow. The native
TMB engine supports the released one-response families and bivariate
Gaussian models; use it for analysis and reporting.
The Julia bridge is halted/deferred and does not make Julia,
JuliaCall, or DRM.jl a package dependency. A future bridge may fit two
unequal response families through a shared latent-scale correlation.
That quantity is rho_latent, not the native
bivariate-Gaussian residual correlation rho12. Neither a
point-estimate nor an interval-inference claim for cross-family
rho_latent is part of the 0.6.0 release surface.
For the released bivariate workflow, see Changing residual coupling with
rho12. For the future bridge boundary, see Julia engine: future support; keep
engine = "tmb" as the default.
Development boundary
The following syntax is retained only to name the intended future interface. It is not a supported 0.6.0 fitting or reporting recipe.
# Deferred beyond 0.6.0: do not use for release analysis.
# fit <- drmTMB(
# bf(mu1 = y1 ~ x, mu2 = y2 ~ x),
# family = c(gaussian(), poisson()),
# data = dat,
# engine = "julia"
# )
# rho_latent(fit)The 0.6.0 capability ledger and release scope manifest are the source of truth for any later promotion decision.