Physbound

Validates RF calculations against hard physical limits to catch AI hallucinations in engineering workflows. Exposes four tools: rf_link_budget for Friis transmission equation checks with aperture limit validation, shannon_hartley for channel capacity verification, noise_floor for thermal noise and cascaded noise figure calculations, and radar_range for monostatic detection analysis. Returns structured errors with LaTeX explanations when claims violate Shannon limits, thermal noise floors, or antenna gain bounds. Reach for this when reviewing link budgets, vetting telecom proposals, or designing receiver chains where impossible numbers could slip through. Built on SciPy's CODATA constants with dimensional analysis through Pint. The repo includes a full hallucination test suite showing where LLMs commonly fail on SNR, aperture, and fourth-root radar calculations.