Release Notes
Added
- LP setup timing instrumentation --
SolverStatistics now tracks cumulative
wall-clock time for load_model, add_rows, and set_row_bounds/set_col_bounds
separately from solve time. Three new columns (load_model_time_ms,
add_rows_time_ms, set_bounds_time_ms) in training/solver/iterations.parquet
enable diagnosing LP rebuild overhead vs simplex time.
- LP setup optimisation -- Model persistence across scenarios at the same
stage (S1), active-cut-count caching, incremental cut append, sparse cut
representation, and bound-zeroing deactivation. Reduces LP rebuild overhead
by avoiding redundant load_model + add_rows calls.
- Simulation basis warm-start -- Simulation LPs are warm-started with the
per-stage basis from the training checkpoint. The basis is read-only and shared
across all threads, preserving determinism while reducing simplex iterations.
- Physical evaporation upper bounds -- Evaporation flow (
Q_ev) is now
bounded above by a physical estimate derived from linearisation coefficients
and maximum storage, with a 2x safety margin. Over-evaporation slack
(f_evap_minus) is penalised at 100x the under-evaporation cost to prevent
the solver from inflating evaporation as a dump valve.
- Cost breakdown extraction --
inflow_penalty_cost, hydro_violation_cost
(evaporation + withdrawal violations), and diversion cost are now extracted
from LP primal values into the simulation cost breakdown. Previously these
LP objective contributions were included in immediate_cost but not reported
in any named component field.
Fixed
- Stale z_inflow column offset formulas -- Corrected 8 test column offset
formulas and 10+ comments in lp_builder.rs and noise.rs that still
referenced the old z_inflow-at-end-of-columns layout after the N*(1+L)
refactoring. Tests passed coincidentally because adjacent columns had
identical bounds and objective values.
- Simulation cost extraction with LP prescaling -- Per-variable cost
extraction (spillage, thermal, exchange, NCS curtailment, and the aggregate
cost breakdown) now divides by col_scale[j] to undo column prescaling.
Without this, per-entity costs were inflated by the column scale factor when
LP prescaling was active.
- NCS curtailment cost semantics -- Changed curtailment cost to use
curtailment_mw (available minus generation) instead of generation_mw.
The field now reports the actual penalty for not generating, matching the
curtailment_cost field name semantics.
- HiGHS internal scaling disabled -- Set
simplex_scale_strategy = 0 (off)
in default solver options. Cobre's own prescaler handles conditioning; the
HiGHS internal scaler interfered with basis reuse and dual extraction.
cobre-cli 0.1.11
Install cobre-cli 0.1.11
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cobre-rs/cobre/releases/download/v0.1.11/cobre-cli-installer.sh | sh
Install prebuilt binaries via powershell script
powershell -ExecutionPolicy Bypass -c "irm https://github.com/cobre-rs/cobre/releases/download/v0.1.11/cobre-cli-installer.ps1 | iex"
Download cobre-cli 0.1.11
cobre-mcp 0.1.11
Install cobre-mcp 0.1.11
Install prebuilt binaries via shell script
curl --proto '=https' --tlsv1.2 -LsSf https://github.com/cobre-rs/cobre/releases/download/v0.1.11/cobre-mcp-installer.sh | sh
Install prebuilt binaries via powershell script
powershell -ExecutionPolicy Bypass -c "irm https://github.com/cobre-rs/cobre/releases/download/v0.1.11/cobre-mcp-installer.ps1 | iex"
Download cobre-mcp 0.1.11
