Issues

• design(parallel): caller-supplied executor for batch slices (gated on the fleet executor concept)

  architectureenhancement
  OldCrow/libstats#285 · Filter by author OldCrow opened on Oct 27, 2026 · Post-v3 — Architecture Refactor

• cleanup(parallel): deprecate the unused parallel reduction helpers (thread-count-dependent partitions); remove at the next major

  cleanuptechnical-debt
  OldCrow/libstats#284 · Filter by author OldCrow opened on Oct 27, 2026 · Accuracy, contracts & kernel hygiene patch

• reproducibility(gaussian): fit() chunk partition depends on the platform grain, so mean/variance bits differ across machines

  accuracyquality
  OldCrow/libstats#283 · Filter by author OldCrow opened on Oct 27, 2026 · Accuracy, contracts & kernel hygiene patch

• refactor(dispatch): one slice callable per batch site; PARALLEL and WORK_STEALING behind one internal scheduling seam

  architectureRefactortechnical-debt
  OldCrow/libstats#282 · Filter by author OldCrow opened on Oct 27, 2026 · Post-v3 — Architecture Refactor

• TruncatedNormal median of a symmetric window is −1.4e-16, not 0, on Zen 4 (UCRT)

  OldCrow/libstats#277 · Filter by author OldCrow opened on Oct 14, 2026 · Accuracy, contracts & kernel hygiene patch

• TruncatedNormal variance 0 where the window is narrower than ~1e-154 σ (standardized c2 underflows)

  OldCrow/libstats#273 · Filter by author OldCrow opened on Oct 13, 2026 · Accuracy, contracts & kernel hygiene patch

• NaN or −inf where a product or sum of two parameters overflows (Pareto, Weibull, Beta, NegativeBinomial, Gumbel)

  OldCrow/libstats#272 · Filter by author OldCrow opened on Oct 13, 2026 · v3.0.0 — corvus adoption
  1

• Gamma CDF at tiny shape is 5–10 ε off on its large side; F + S − 1 up to 5 ε

  OldCrow/libstats#271 · Filter by author OldCrow opened on Oct 12, 2026 · v3.0.0 — corvus adoption
  1

• beta_i loses accuracy at very large b (NegativeBinomial with tiny p)

  OldCrow/libstats#270 · Filter by author OldCrow opened on Oct 12, 2026 · v3.0.0 — corvus adoption

• Moved-from delegate distributions disagree with their delegates (Bernoulli, Geometric, ChiSquared, Erlang)

  OldCrow/libstats#269 · Filter by author OldCrow opened on Oct 12, 2026 · Accuracy, contracts & kernel hygiene patch

• ExponentialDistribution::setParameters(double) declared but never defined

  OldCrow/libstats#268 · Filter by author OldCrow opened on Oct 12, 2026 · Accuracy, contracts & kernel hygiene patch

• libstdc++ std::cyl_bessel_i writes signgam: residual of #173 in von Mises

  bugplatformtechnical-debt
  OldCrow/libstats#267 · Filter by author OldCrow opened on Oct 12, 2026 · v3.0.0 — corvus adoption

• accuracy: direct prefactor and density forms below shape 20 lose min·ln(max)·ε; values jump across the Stirling switch

  accuracy
  OldCrow/libstats#266 · Filter by author OldCrow opened on Oct 12, 2026 · v3.0.0 — corvus adoption
  1

• TruncatedNormal CDF non-monotone by up to 10 ulps where F moves less than one ulp per input ulp

  OldCrow/libstats#263 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• beta_i rounds the survival up to ~16 ulps high at exact dyadic ties (Binomial(10, ½), NB(2, ½))

  OldCrow/libstats#262 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption
  1

• Reference provenance: kTemmeCoefficients and Student-t ν ≥ 1e20 references come from unchecked-in scripts

  OldCrow/libstats#261 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• Table generators: two self-checks too loose to catch low-order coefficient errors

  OldCrow/libstats#260 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• NEON exp ULP gate drops its edge rows; generator double-rounds subnormal results

  OldCrow/libstats#259 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• Oracle: per-row accuracy budgets and a row diff against the previous sweep

  OldCrow/libstats#258 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption
  1

• Fisher-F batch sample() stream differs from repeated scalar sample() at shapes < 1 (document)

  OldCrow/libstats#253 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• von Mises entropy below κ = 50 limited by Tier 2 Bessel I0/I1 (up to 3.2e-5 abs; 3.2e-5 step at the κ = 50 cut)

  OldCrow/libstats#252 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption

• Gamma CDF up to 433 ε below β·x = DBL_MIN; F flat over 6 ulps, 10-ulp step at DBL_MIN (a = 0.5)

  OldCrow/libstats#251 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption
  1

• Fisher-F CDF steps ~6 ulp at the d1·x overflow seam

  OldCrow/libstats#250 · Filter by author OldCrow opened on Oct 11, 2026 · Accuracy, contracts & kernel hygiene patch

• Student-t CDF up to 45 ε in the side-switch region (t ≈ 1.6–1.75, ν 1e3–1e16)

  OldCrow/libstats#249 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption
  1

• Student-t CDF ~330 ε just below ν = 40 (a < 20, no BGRAT branch)

  OldCrow/libstats#248 · Filter by author OldCrow opened on Oct 11, 2026 · v3.0.0 — corvus adoption
  1