A signal too weak to cross a detection threshold is undetectable. Adding noise makes detection worse — that's the expected story. Stochastic resonance is the case where noise makes it better.
In a threshold system, noise occasionally boosts a subthreshold signal above the detection line. The key is that the timing of those boosts is not uniform with respect to the signal: when the signal is near its peak, combined amplitude is more likely to exceed threshold. When the signal is at trough, it isn't. The detector's output therefore carries information about signal phase that a noise-free subthreshold signal cannot provide. Some noise recovers structure. More noise destroys it again.
First demonstrated in crayfish mechanoreceptor afferents by Douglass, Wilkens, Pantazelou, and Moss (1993). The crayfish's hydrodynamic sense — used to detect prey and predators — works better with a small amount of added noise than without it. The optimal noise level is not zero. Replicated in human vibrotactile thresholds and balance control: subthreshold mechanical noise applied through insoles reduces postural sway in older adults.
What this can't show: from inside the detector, a noise-assisted crossing is identical to a noise-alone false alarm. The improvement is statistical — visible across many trials, invisible in any individual event. The system receives more information without any mechanism to identify which detections the signal caused.
The crayfish doesn't know noise is helping. The older adult's vestibular system doesn't know the insole is doing something. The information gain is real; its source is inaccessible.