Garden Succession Solver

Garden Succession Solver enters demand rather than dates — pounds per week between two weeks of the year — and solves sow and set dates so harvest windows tile without gaps or gluts. Four constraint families hold at once: bed-week occupancy (a bed is busy from set to pull-out, and two plantings share a bed when their weeks do not overlap or the bed-feet still fit), germination feasibility from a bundled soil-temperature climatology (or your own pasted 52-week probe curve), rotation eligibility, and the fall backstop (75% of fall-adjusted DTM complete by your latitude's 10-hour day, from closed-form solar geometry). Infeasibility is proved separately and cheaply by a sliding-window bed-week capacity bound, so 'no arrangement exists — week 28 is 42 bed-feet short' is a proof, not a search timeout. When the bound passes but the bounded search cannot place everything, the app says exactly that: search found nothing, not proven impossible. Rotation is tiered by evidence and it says so on screen. Family rotation and bed-week occupancy are hard; every remembered year is marked remembered or guessed, and a guessed year downgrades that constraint to a soft warning so fabricated history can never harden into a false proof. Pathogen horizons ship OFF by default, are opt-in per bed, and carry the published range with its source rather than a single integer. Reality diverges, so it re-solves: mark a planting dead and the whole graph is re-derived, with a diff of which dates moved, which beds were reassigned, and which fall plantings just became arithmetically impossible. The bed-week Gantt is clickable, the line-item table cites the rule and inputs behind every date, and per-crop overrides (DTM, harvest window, yield per bed-foot) are remembered. It is useful from three beds and zero history, imports last year's CSV through the front door, and exports the whole plan and history to a local JSON file. Four locales.