SolvScout

A PhD student in chemical engineering is modelling a reactor. He tries ode45 because it was in the tutorial. It crashes on stiff systems. He tries ode15s because somebody on a forum said so. It works, sort of. He tunes tolerances for two weeks. His advisor asks why the deadline is slipping. The answer is that nobody taught him which solver to use for which problem, and the documentation for all six contenders reads the same.

Imagine a tool that takes his ODE in, fingerprints it in seconds, runs every reasonable solver on it, and shows him a comparison table. The right solver comes with evidence. The wrong solvers come with evidence too. He picks the one that wins on his actual problem.

A defence prime is evaluating solvers for a next-generation flight controller. The team splits into four factions, one per vendor. Three months of benchmark politicking, contradictory results, and dueling spreadsheets. The decision gets made by someone who did not see the data.

Imagine an evaluation that is evidence-first. Everyone in the room sees the same comparison table, on the same problems, with the same statistical thresholds. The conversation moves in an afternoon from “which vendor do you like?” to “which solver wins on our actual workload?”.

Every engineer who writes an ODE solves a classification problem first: what kind of system do I have? Whose solver should I use? Existing tools do not help them. The ecosystem needs a neutral fingerprinter and benchmarker — a tool everyone trusts, that runs every solver fairly, and that wins user trust by saying clearly when the right answer is not us. That on-ramp is what Scout and Tune are.