engine = "julia" refuses the weights argument at fit time
(drmTMB_julia_bridge() and its structured counterparts in
R/julia-bridge.R abort on a non-missing weights), so a bridge fit is unweighted by construction and its
prior weights are all one. Without this method weights() fell through to
stats:::weights.default() and returned NULL silently, while
weights() on the same model fitted with engine = "tmb" returns a vector
of ones. This reports the ones, matching both the native engine and
DRM.jl's own weights(fit) = ones(nobs(fit)) (src/comparison.jl).
Usage
# S3 method for class 'drmTMB_julia'
weights(object, ...)Details
These are PRIOR (per-observation) weights, not Akaike model weights.
The length comes from nobs(), not from object$nobs: a joint
(drmTMB_julia_joint) fit counts only its OBSERVED rows
(nobs.drmTMB_julia_joint() is sum(observed_y)), so reading the raw field
would return a weight vector longer than the data the fit actually used.