Thornbill is a CP solver designed to run on microcontrollers. The same solver that enters this challenge also runs on a Raspberry Pi Pico and an ESP32, within the few hundred kilobytes of RAM available on these systems. Most of what distinguishes it from a desktop-class solver follows from that target. - **Compact code.** The solver core is a small C++17 kernel with no dependencies; the entire solver (kernel, propagators and search engine) compiles to under 400 KB of machine code on embedded platforms. - **A fixed, pre-sized memory arena.** Thornbill reserves its working and search memory once, at start-up, and never calls the allocator while solving. Its memory use is therefore bounded and known in advance, which is what an embedded board needs. On a desktop or server, it keeps the solver's working set compact. - **Interval domains.** Variable domains are contiguous intervals [l, u] with no holes, and propagation is (at most) bounds-consistent. This gives up some pruning in exchange for a small per-variable footprint and fast, simple propagation. The challenge configuration runs in three modes. For the fd category, it follows the model's search annotation. For free search, it alternates between the search annotation and an activity-based (dom/wdeg) search, committing to one of them according to a heuristic based on progress and time. For the parallel category, it implements a parallel portfolio: the model's own search annotation, several activity-based searches with restarts, and large-neighbourhood search, sharing incumbents and failure weights across the workers. The common global constraints (all-different, cumulative, circuit, regular, global cardinality, bin-packing, and others) have native propagators rather than being decomposed. The free search sequential portfolio is also available on the embedded target platforms, while the parallel portfolio uses OS threads, which are not available on microcontrollers. Thornbill's code is written entirely by AI coding agents (Codex and Claude Code), working under close (human) supervision.