She pulled a socket wrench from her backpack, knelt beside the SD9's base unit, and began unbolting the harmonic drive housing. 14 bolts. She removed them in a specific sequence: outside corners first, then alternating inward, the way you're supposed to if you want to avoid warping the gasket surface. Two of the senior engineers noticed; they glanced at each other, but said nothing.
The housing came off. Inside was the harmonic drive, the precision gear mechanism that translated motor rotation into the robotic arm's movement: three components—the wave generator, the flex spline, and the circular spline. Together, they allowed the arm to move with sub-millimeter accuracy, when they worked.
Ivory reached in and carefully removed the flex spline, a thin-walled steel cup with teeth machined around its outer edge. She held it under the magnification lamp and turned it slowly. Then she reached into her backpack and pulled out a second flex spline, this one from a working unit, borrowed from the replacement parts inventory that Nathan had quietly set aside for her. She held them side by side under the lamp.
"Everybody look at this," she said—not loud, not performative, the voice of someone who had done this a hundred times in her basement and was simply doing it again in a bigger room. "The teeth on the left, from the failed unit, show a fracture pattern. See the micro-cracks along the root of every third tooth? That's not random wear. That's not a manufacturing defect. That's resonant fatigue."
Brent Calloway crossed his arms. "Resonant fatigue doesn't happen in harmonic drives under normal operating conditions."
Ivory didn't look at him; she kept her eyes on the flex spline. "You're right, it doesn't, under normal conditions. But 8 months ago, someone pushed a firmware update that changed the drive frequency from 62 hertz to 68 hertz."
She flipped open her notebook to a page dense with calculations—neat handwriting, no margins wasted.
"68 hertz is the natural resonant frequency of this specific steel alloy, 17-4 PH stainless, heat-treated to H900 condition. At that frequency, every rotation cycle causes the flex spline teeth to deflect a fraction of a micron beyond their design tolerance. One cycle, no damage. A thousand cycles, no visible damage. But after 200,000 cycles, the cumulative micro-stress exceeds the fatigue limit. The teeth crack. The arm seizes." She pointed at her calculations. "The math is here: resonant frequency derivation, cross-referenced with the alloy spec sheet from the manufacturer and the firmware change log that Nathan pulled from the system archive."
The room was quiet. Brent spoke again, louder this time. "We tested that firmware update exhaustively before deployment. It passed every benchmark."
Ivory looked at him for the first time. "You tested the software. You verified that the code executed correctly. You checked the output signals and the response times." She paused. "But you never tested what the software does to the metal. You ran digital diagnostics on an analog problem. The firmware works perfectly. It's just telling the machine to destroy itself."