PhysicsReview

The problem Simulation software has a class of bugs that ordinary testing does not catch. A missing ½ in kinetic energy. An angle passed in degrees where radians are expected. A temperature in Celsius where the ideal gas law needs Kelvin. A time step too large for the system's natural period. None of these crash. The code compiles, the tests pass and the output looks plausible. It is simply wrong. We are physics and computer science students working in computational physics, and we hit these bugs in our own research. Today, catching them depends on a manual review by someone who understands both the code and the physics. That expertise is scarce, reviews are slow, and the bugs are usually found late: after a wrong result, a failed prototype or hours of debugging. Teams building robotics, engineering simulation, game physics and scientific software all face the same gap. The solution PhysicsReview is a physics-aware code reviewer. A developer uploads source files or pastes a git diff, and PhysicsReview runs three passes: Law mapping. It identifies which functions implement which physical laws, such as Newton's second law, Bernoulli's equation, the ideal gas law or an RC circuit. Test generation. It writes the tests experts rarely have time for: analytical benchmarks against closed-form results, conservation checks, unit-consistency tests, validity-range guards and sign-convention tests. Review. It flags physics violations in plain language, each with a file, a line, a severity and a suggested fix. The model does not invent physics. Every pass is grounded in a curated knowledge base of 14 laws across 6 domains: mechanics, fluids, thermodynamics, circuits, electromagnetism and modern physics. Each law carries its equations, SI units, validity conditions, known implementation bugs and test templates. Outputs are schema-validated JSON, and source lines are numbered so findings cite real locations