A system rated for a load is a promise; the honest critic keeps piling on until the promise breaks. Below the rating it holds; a little past it, it degrades; keep going and it fails — sometimes gracefully, sometimes all at once. The number that matters is not the rating but the margin between the rating and the break.
Response rises linearly with load while the load stays under the rating, then bends over and collapses past a breaking point set by the brittleness (a brittle system drops sharply, a ductile one sags). The instrument computes the response curve, the breaking load, and the safety margin = (break − rating)/rating live: a system can pass at its rating and still sit one nudge from failure. The honest test is not 'does it meet spec' but 'how far past spec until it breaks' — the margin, quantified.
A one-dimensional load/response toy; real stress testing probes many failure axes (heat, time, adversarial inputs, corner cases) and failure is rarely a single clean point. The principle — margin matters more than the rating, and you only learn it by testing to failure — is the real content.