Smooth Surrogate Models Approve Impossible Firing Patterns: A Hidden Flaw in Writing Brain Codes

When scientists design stimulation to make a neuron fire at a desired rate, they often use a smooth, easy-to-optimize mathematical copy (a surrogate) of the neuron's response curve. Here I show that this shortcut has a dangerous blind spot: the smoothness that makes the surrogate easy to work with also erases the neuron's real limits. Using a detailed computer model of a neuron, I measured the set of firing rates the cell can actually produce: either zero, or any rate between about 55 and 115 Hz. There is no way to get rates like 30 or 45 Hz. Yet a standard smooth surrogate, fit almost perfectly to the neuron's own data, claims that 30 Hz and 45 Hz are achievable. When I used this surrogate to design stimulation for 20 cells, 12 of them were wrong; the pipeline reported zero total error, while the real neuron model stayed silent. This error is invisible from inside the surrogate's own calculations. This failure is not about poor fitting; every smooth model I tested has the same blindness, because smoothness itself erases the sharp boundaries that define what the neuron can do. A simple safety check, testing target rates against the neuron's true firing curve before optimization, catches impossible requests and eliminates the failure. The lesson is clear: when writing firing codes through a smooth model, you must first check that your target rates lie within the neuron's actual possible range.