Validation release. The framework now ships with the executable evidence of its calibration claims.
htdv_sim_summary: aggregated output of the pre-registered factorial
Monte Carlo study (1024 cells × 500 replications × 3 inferential
layers, 31 hours wall-clock on 16 cores, sign-corrected coverage).htdv_empirical_benchmarks: three-dataset external validation against
Stock-Watson (2007), Campbell-Shiller (1998), and the iid Welch
baseline on US-Canada 10y yields.HTDV-validation consolidates both validations with
the per-layer reading and the persistence-ladder cross-cut.htdv_simstudy(): end-to-end factorial Monte Carlo runner with
cell-level parallelism via parallel::mclapply and atomic per-cell
caching that survives worker failures and supports resume.htdv_simstudy_summary(): aggregate per-cell results.htdv_simstudy_warnings(): flag cells where the Bayesian
diagnostic-pass rate falls below a user-set threshold (typically
high-persistence × small-sample corners where the AR(1) likelihood
approaches its limit of identification).simstudy_two_sample.stan for two-sample AR(1)
inference with shared autocorrelation.delta = E[X_2] - E[X_1]). Earlier output had spuriously low HAR
and bootstrap coverage for delta > 0 cells; the
recover_coverage.R helper of the companion paper repository
regenerates the corrected aggregates from the raw RDS.<doi:10.48550/arXiv.1604.00695> (CRAN-mandated DOI form).IsadoreNabi/HTDV; ORCID,
email, and BugReports URL aligned in DESCRIPTION and CITATION.Initial public release.
htdv_fit) with TAC, WSC, MPC, Whittle,
and Composite likelihood backends.htdv_envelope).htdv_lrv).htdv_fixedb).htdv_boot).htdv_conformal).htdv_rope, htdv_bf, htdv_waic_lfo,
htdv_stack).htdv_diagnostics, htdv_ppc).htdv_equivalence_constants).