Release Notes

0.5.4

Maintenance

• CI pipeline and deployment workflow improvements.
• README and documentation updates.

0.5.3

New features

• pyworkforce.staffing.MultiSkillStaffing — minimum-cost agent-mix optimiser for multi-skill contact-centre staffing. Given per-skill requirements (from ErlangC / ErlangA) and a catalogue of agent skill profiles, solves the integer programme that minimises total weighted headcount. Flexible (multi-skilled) agents count towards every skill they hold, so the solver automatically balances dedicated vs. flexible hiring. Supports an optional max_agents budget cap and is now exported from the pyworkforce root namespace.
• pyworkforce.queuing.ErlangB / MultiErlangB — Erlang B (M/M/c/c) pure-loss queue for trunk and channel sizing. Models systems with no waiting room where blocked calls are shed. Provides blocking_probability, achieved_occupancy and required_positions (with max_blocking and max_occupancy targets). MultiErlangB sweeps a parameter grid in parallel, mirroring MultiErlangC.
• pyworkforce.breaks.BreakScheduler — CP-SAT solver that assigns break start times to agent slots within each shift while enforcing break windows (earliest start, latest end), preventing overlapping breaks on the same slot, and guaranteeing that simultaneous breaks never exceed the coverage slack. Accepts multiple break types per shift.
• ErlangB, MultiErlangB, BreakScheduler, and MultiSkillStaffing are all exported from the pyworkforce root namespace.

Dependency changes

• Lowered numpy floor to >=1.23 (code uses only stable basic operations; aligns with the pandas>=2.2 minimum and unblocks the conda-forge package).
• Lowered ortools floor to >=9.6 (all CP-SAT APIs in use are stable at this version; matches the conda-forge ortools-python feedstock).

0.5.2 — previous release

New features

• pyworkforce.queuing.ErlangA — M/M/c+M queue with customer abandonment. All metrics computed exactly from the birth-death stationary distribution.
• pyworkforce.queuing.MultiErlangA — parameter-grid variant of ErlangA.
• pyworkforce.shifts — shift coverage builders (shift_coverage_from_hours, shift_coverage_from_spans, shift_coverage_from_periods, validate_shift_coverage, coverage_to_dataframe).
• pyworkforce.utils.results_to_dataframe — tidy DataFrames from Multi* results.
• All estimators expose get_params(), a readable repr() and store their last result as solution_.

Bug fixes

• MultiErlangC now correctly populates achieved_occupancy_params and required_positions_params.
• Validation helpers accept integers where floats are expected and reject booleans, NaN and infinities.
• MinHoursRoster rejects duplicate resource names.

Packaging & tooling

• Migrated to PEP 621 pyproject.toml; upgraded dependency floors.
• Added ruff linting, Dependabot config, and a PyPI trusted-publishing workflow.
• Documentation moved to VitePress on GitHub Pages.

For older releases, see the GitHub releases page.