mobility-destination-sequence-sampler

Rust/Python bounded destination-plan top-K search with an exact validation oracle. Mobility prepares inputs; this package owns destination choice.

Transport modeller? Start with MODELLER_GUIDE.md. It explains the algorithm with a small activity-plan example, makes the search guarantees explicit, translates the terminology, and gives a one-context debugging workflow.

Public API

DestinationPlanSearch is the only exported class.

top_k() shortlists plausible destinations, grows partial plans from both ends, joins them near the middle, improves the best complete plans, and ranks them with the full rigidity-aware utility. Repeated anchors remain the same destination throughout a plan. Its report contains top_k_is_proven=False; a successfully completed exact report contains top_k_is_proven=True. See DESIGN.md for schemas and invariants.

Callers can supply an optional utility_profile_id on OD costs and initial locations. This searches several population-specific cost profiles in one call: destination metadata and parallel scheduling are shared, while each context uses its assigned OD graph. Calls without the column keep the original single-profile behavior.

Only those paths are in the working tree. Particle, exhaustive-sampling, and second-order research is preserved by the research-archive-2026-07-29 Git tag (for example, git show research-archive-2026-07-29:<path>). Start with ACTIVE_SEARCH.md for the active algorithm and tuning contract.

Development

mamba env create -f environment.yml
mamba run -n mobility-destination-sequence-sampler python -m maturin develop --release
mamba run -n mobility-destination-sequence-sampler python -m pytest

Current quality/performance: BENCHMARKS.md. Experiment routing and decisions: experiments/README.md.