Issues

• estimation: method = gn and optimizer = trust_region do not converge on the joint PK + TTE example

  P3
  FeRx-NLME/ferx-core#2438 · Filter by author carldelfin opened on Oct 27, 2026

• estimation: the non-Gaussian FOCEI objective depends on the FD η-Hessian step (0.73 OFV on CTMM), which kept CTMM out of #1778

  P2
  FeRx-NLME/ferx-core#2437 · Filter by author carldelfin opened on Oct 27, 2026
  

• levels: rank-check every level block at bind time, not only one that rescues a random effect (#1991 follow-up)

  P2
  FeRx-NLME/ferx-core#2421 · Filter by author roninsightrx opened on Oct 26, 2026
  

• a profiled covariance step's held-layout re-solve (#2112) skips the #2214 EBE guard

  P3
  FeRx-NLME/ferx-core#2420 · Filter by author TeunP opened on Oct 26, 2026

• bug(estimation): run_covariance on a profiled fit reports the caller's level estimates, not the re-solved ones its matrix is at

  P2
  FeRx-NLME/ferx-core#2419 · Filter by author TeunP opened on Oct 26, 2026

• fix(estimation): uncertainty draws hold a profiled level the covariance step held on its bound (#2416 review row 3)

  P2
  FeRx-NLME/ferx-core#2418 · Filter by author TeunP opened on Oct 26, 2026

• test(changelog): check_rejects_each_malformed_fragment_and_accepts_its_twin races on a fixed CARGO_TARGET_TMPDIR path

  P3
  FeRx-NLME/ferx-core#2409 · Filter by author TeunP opened on Oct 26, 2026
  

• investigate(fd): the |θ₀| floor on the reconverged FD gradient divides inner noise by a smaller step on an O(1)-sensitivity θ

  P3
  FeRx-NLME/ferx-core#2407 · Filter by author carldelfin opened on Oct 26, 2026

• bug(fd): the |θ₀| FD step floor has no lower clamp (also [covariate_nn] weights and the σ forward differences)

  P2
  FeRx-NLME/ferx-core#2406 · Filter by author carldelfin opened on Oct 26, 2026

• profile: W_PROFILE_FD_GRADIENT code, verbose profile-work summary and S1 docs numbers (after #2401)

  P3
  FeRx-NLME/ferx-core#2405 · Filter by author TeunP opened on Oct 25, 2026

• diagnostics: the reported final gradient does not say which components were one-sided or missing

  P2
  FeRx-NLME/ferx-core#2392 · Filter by author TeunP opened on Oct 25, 2026
  

• refactor(estimation): one masked finite-difference component for central_diff_packed and gauss_newton::masked_fd_component

  P3
  FeRx-NLME/ferx-core#2391 · Filter by author TeunP opened on Oct 25, 2026
  

• inner: two warm-start edge cases in find_ebe_iov — a nested start that fails to compact drops η̂, and a κ-only joint start can arm the ODE+IOV reject that SIR scores

  P3
  FeRx-NLME/ferx-core#2382 · Filter by author TeunP opened on Oct 25, 2026

• perf(mixture): the per-class warm starts carry κ̂ with η̂ under IOV (#2261 follow-up)

  P3
  FeRx-NLME/ferx-core#2380 · Filter by author TeunP opened on Oct 25, 2026

• Adaptive IOV: an ADDL-implied dose takes kappa from its own window, covariates from its governing record's

  P2
  FeRx-NLME/ferx-core#2371 · Filter by author carldelfin opened on Oct 24, 2026

• perf(reader): the ADDL governing-record search is O(implied x records)

  P3
  FeRx-NLME/ferx-core#2370 · Filter by author carldelfin opened on Oct 24, 2026

• The #614 record-order pass ranks an ADDL-implied dose by its parent's row

  P2
  FeRx-NLME/ferx-core#2369 · Filter by author carldelfin opened on Oct 24, 2026

• ADDL: TTE, count, discrete-state and W_MISSING_DV rows never govern an implied dose

  P2
  FeRx-NLME/ferx-core#2368 · Filter by author carldelfin opened on Oct 24, 2026

• ADDL: a route's own lag follows the governing record, while ALAG follows the parent row

  P2
  FeRx-NLME/ferx-core#2367 · Filter by author carldelfin opened on Oct 24, 2026

• FREM: FREM-aware closed-form fixed-EBE gradients instead of the central-FD fallback

  P2
  FeRx-NLME/ferx-core#2365 · Filter by author carldelfin opened on Oct 24, 2026

• test(estimation): after #2280 lands, re-check that no profiled A2 start reports converged outside the 1e-5 band ((48, 12, 3.2) is +1.03e-5 on main)

  P3
  FeRx-NLME/ferx-core#2340 · Filter by author TeunP opened on Oct 23, 2026

• eval(estimation): the profiled MBMA fit (A2) converges to a second basin 171.9 OFV above NONMEM's optimum from TVE0 = 55

  P2
  FeRx-NLME/ferx-core#2339 · Filter by author TeunP opened on Oct 23, 2026

• diagnostics: W_NONFINITE_OBJECTIVE still says the full subject list is "not yet in the R package"

  P3
  FeRx-NLME/ferx-core#2337 · Filter by author TeunP opened on Oct 23, 2026
  

• feat(sim): covariate-conditional η draw for FREM simulations, so each subject's PK follows its own covariates (follow-up of #2245) [#391]

  enhancementP2
  FeRx-NLME/ferx-core#2335 · Filter by author TeunP opened on Oct 23, 2026

• A RATE=-1 infusion's rate does not follow R{n} when it changes during the infusion (NONMEM re-reads R{n} on every advance)

  P2
  FeRx-NLME/ferx-core#2332 · Filter by author carldelfin opened on Oct 23, 2026