13 × 13 arrow-word grid published on Flexorama

The need

Publishing new arrow-word grids on Flexorama every day requires a large lexicon, good clues and a reliable way to fill valid grids.

The solution

  • A WPF studio to create templates, fill grids, edit clues, run consistency checks and export to the platform.
  • A grid-filling engine based on constraint satisfaction: backtracking search, word domains stored as bitsets, and a computation budget to decide quickly whether a grid is feasible.
  • A Python data pipeline that builds the lexicon from the Lexique 3 French database, generates clues at scale through Anthropic’s Batch API and applies about twenty validation rules before import.

Technical notes

Local SQLite database, batch exchanges in XML, MVVM architecture, and automated tests for both the engine and the Python pipeline.