Skip to content

0.4.3 - 2026-04-13

Choose a tag to compare

@github-actions github-actions released this 13 Apr 13:32
99b6c0e

Release Notes

Added

  • HistoricalResiduals noise method — new NoiseMethod::HistoricalResiduals
    variant that copies pre-computed eta vectors from HistoricalScenarioLibrary
    into the backward-pass opening tree, preserving empirical cross-entity
    correlation from actual inflow observations. Uses hash-based deterministic
    window selection and skips Cholesky decorrelation.
  • Season ID consistency validation (rules 27–30) — four new semantic
    validation sub-rules: season range coverage, observation coverage per season,
    resolution consistency across seasons, and contiguity of defined seasons.
  • Observation alignment validation (rule 31) — validates observation-to-season
    alignment with a three-tier season mapping fallback chain. Replaces hardcoded
    month0() in historical window discovery and standardization with
    SeasonMap-aware logic.
  • SLURM MPI test infrastructure — replaced mpi_smoke.sh with
    mpi_slurm.sh for Dockerized SLURM cluster testing.

Changed

  • Timing instrumentation columns — renamed for accuracy:
    forward_solve_msforward_wall_ms, backward_solve_ms
    backward_wall_ms, mpi_broadcast_mscut_sync_ms,
    rayon_overhead_msbwd_rayon_overhead_ms. Added lower_bound_ms
    and fwd_rayon_overhead_ms. Removed stub columns forward_sample_ms,
    backward_cut_ms, io_write_ms.

Fixed

  • LP determinism across MPI configurations — reload LP model per
    scenario/trial-point to prevent HiGHS internal state carry-over (basis, RNG
    position) from making results depend on rank/thread count. Verified
    bit-identical lower bounds across 1r/1t, 1r/2t, 2r/1t, and 1r/4t.
  • Forward pass sampler count — pass global total_forward_passes to the
    sampler instead of the per-rank local count, fixing LHS/QMC determinism
    across MPI configurations.
  • Timing overhead double-counting — fix overhead_ms always being zero
    due to cut_sync_ms included in both backward_ms and the attributed sum.
  • Historical window year normalizationbuild_observation_sequence now
    normalizes year offsets so the first study stage gets year_offset=0,
    fixing incorrect window matching for studies not starting in January.
  • MPI handling — fix MPI communication edge cases.

Install cobre-cli 0.4.3

Install prebuilt binaries via shell script

curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cobre-rs/cobre/releases/download/v0.4.3/cobre-cli-installer.sh | sh

Install prebuilt binaries via powershell script

powershell -ExecutionPolicy Bypass -c "irm https://github.com/cobre-rs/cobre/releases/download/v0.4.3/cobre-cli-installer.ps1 | iex"

Download cobre-cli 0.4.3

File Platform Checksum
cobre-cli-aarch64-apple-darwin.tar.xz Apple Silicon macOS checksum
cobre-cli-x86_64-apple-darwin.tar.xz Intel macOS checksum
cobre-cli-x86_64-pc-windows-msvc.zip x64 Windows checksum
cobre-cli-aarch64-unknown-linux-gnu.tar.xz ARM64 Linux checksum
cobre-cli-x86_64-unknown-linux-gnu.tar.xz x64 Linux checksum