Issues

• Plain binomial probit GAM (3 smooths, ~200k rows) takes 8-90 min per fit; 5 of 12 exceeded 2 h

  SauersML/gam#4595 · Filter by author SauersML opened on Sep 25, 2026
  2

• gam-event-history: the forward kernel's Lagrange interpolant amplifies ln r's approximation error near the hull edge, and the filter diverges

  SauersML/gam#4594 · Filter by author SauersML opened on Sep 25, 2026

• CLI 6-row delayed-entry survival fit: the REML criterion falls to -inf at a Laplace fold (inner curvature crosses zero at theta~2.24), so no optimum exists

  SauersML/gam#4593 · Filter by author SauersML opened on Sep 25, 2026
  3

• survival marginal-slope 2930: outer Hessian column 0 misses FD by 1.5e-6 rel, and the separated seed's inner solve stalls

  SauersML/gam#4592 · Filter by author SauersML opened on Sep 25, 2026
  1

• BMS learned-kappa Matern fit stops at |Pg|=4.6e-2 with lambda_min(H)=-45

  SauersML/gam#4591 · Filter by author SauersML opened on Sep 25, 2026

• survival: a left-truncated Weibull-PH seed reaches an inner Hessian with eigenvalue -1.68e3

  SauersML/gam#4589 · Filter by author SauersML opened on Sep 25, 2026
  3

• gam-models: the measure-jet range screen's analytic V' misses its resolved central difference by 2e-8 relative (#2902 test)

  SauersML/gam#4586 · Filter by author SauersML opened on Sep 25, 2026
  2

• Certificate accepts a marginal-slope fit with all smoothing parameters railed to zero; it scores far below its nested probit at low prevalence

  SauersML/gam#4581 · Filter by author SauersML opened on Sep 25, 2026
  1

• Marginal-slope fit wall time does not scale with threads: multistart lanes bound it

  SauersML/gam#4580 · Filter by author SauersML opened on Sep 24, 2026

• transformation-normal fit on a 300-row 5-D order-0 power-8 Duchon basis (12 centres) does not finish in 900 s, isotropic or anisotropic

  SauersML/gam#4567 · Filter by author SauersML opened on Sep 22, 2026
  9

• A rigid-law BMS fit with linkwiggle() in both formulas does not finish in 30 min on 3000 rows

  SauersML/gam#4564 · Filter by author SauersML opened on Sep 21, 2026
  3

• Multi-scale draw selection in smooth_significance descends to the last bit of the criterion while feeding a 2048-draw Monte-Carlo estimate: no problem-sized cost bound, and no test drives te(x, z) beside a factor

  SauersML/gam#4563 · Filter by author SauersML opened on Sep 21, 2026
  3

• SAE interval latents: pinned active-bound slots lack their half-line Laplace mass; θ-adjoint / direct-ρ traces unaudited at the bound

  SauersML/gam#4077 · Filter by author SauersML opened on Sep 20, 2026
  8

• survival/predict: Royston-Parmar posterior mean collapses to the plug-in with ~zero SE whenever one I-spline coefficient sits on its β≥0 bound

  SauersML/gam#3575 · Filter by author SauersML opened on Sep 20, 2026
  2

• gam-cli: gam predict survival bands are mean ± z·sd clamped to [0,1]; the Royston–Parmar branch computes the library's transformed-η interval and throws it away (miscalibrated both ways, one-tailed)

  SauersML/gam#3560 · Filter by author SauersML opened on Sep 20, 2026
  9

• Survival location-scale calpred fit (19.4k rows) doubled its outer iterations between 89bea03263 and 4ee765aeba (52 -> 108, 265 s -> 1509 s)

  SauersML/gam#3412 · Filter by author SauersML opened on Sep 20, 2026
  7

• gam-sae --lib: 16 manifold tests red on main 3b862200 (ARD rho atom axes mismatch #2822, arrow-Schur proximal correction Armijo rejection)

  SauersML/gam#3355 · Filter by author SauersML opened on Sep 20, 2026
  14

• Block-sparse lane: no device route for block OMP + least-squares codes (GpuPolicy::Required refuses)

  SauersML/gam#3326 · Filter by author SauersML opened on Sep 20, 2026
  5

• BMS flex row jets are dense in the primary width, so each added link-deviation column costs r^3-r^4 per outer evaluation

  SauersML/gam#3290 · Filter by author SauersML opened on Sep 19, 2026
  8

• Binomial flexible(probit) link on y ~ x refuses to fit on 3 of 8 cloglog samples at main: outer optimum not certified (|Pg|=5.7e-4 > bound 2.5e-4)

  SauersML/gam#3257 · Filter by author SauersML opened on Sep 19, 2026
  13

• Two Gaussian location-scale wiggle tests red at main: outer smoothing declines a certified optimum an evaluated state beats and certifies nothing

  SauersML/gam#3228 · Filter by author SauersML opened on Sep 19, 2026
  17

• Survival marginal-slope default fit ends with the outer search non-stationary at an indefinite outer Hessian (|Pg| 1.737 vs bound 0.115) on a 120-event frame

  SauersML/gam#3209 · Filter by author SauersML opened on Sep 19, 2026
  7

• Tracking: exact outer Hessians are withheld at eight driver sites, forcing the gradient-only BFGS plan

  SauersML/gam#3201 · Filter by author SauersML opened on Sep 19, 2026
  10

• Survival marginal-slope at K=6 takes 5x longer than K=16 on the same data: first BFGS cost-stalls for 438 s over 45 recoverable outer evals

  SauersML/gam#3005 · Filter by author SauersML opened on Sep 19, 2026
  13

• Survival marginal-slope on a monotone i-spline Weibull baseline spends 900 s in order-4 row towers with most worker threads idle, in both the anchored and closed-form frames (n = 2,000)

  SauersML/gam#2979 · Filter by author SauersML opened on Sep 18, 2026
  9