The hazard nobody thought of. Surface texture harbored bacteria—nobody asked the question. Risk analysis that connects complaints to assumptions.
Surgical device launches after three years of development. Risk analysis covered electrical safety, biocompatibility, mechanical failure, use errors. 510(k) cleared.
Eight months later, first adverse event. Patient developed infection at the surgical site. The device's surface texture, chosen for grip, creates micro-crevices that harbor bacteria during reprocessing. The sterilization process validated for smooth surfaces doesn't reliably sterilize this texture.
This hazard wasn't in the risk file. Nobody asked: "What happens to this surface texture during reprocessing?"
Most risk management happens in Excel. The spreadsheet lies by omission—it shows hazards someone thought to include. It doesn't show what's missing.
When a complaint arrives about an issue not in the risk file, the connection is manual. The post-market team investigates. The risk management team doesn't see it unless someone copies them. The file stays static while reality evolves.
"What's the probability of this hazard causing harm?" Three engineers give three answers. Without defined criteria—what does "remote" mean for this device?—the matrix produces whatever answer the analyst wants.
In Seal, probability criteria are defined per product type. When an analyst selects P3, they must cite the basis. The same hazard evaluated by different teams reaches the same conclusion.
The device overheats during extended use. Engineering's control: a warning label. "Do not use for more than 30 minutes continuously."
Users will ignore it. They'll use it for 45 minutes because they're almost done. The label absolves the company on paper while patients get burned.
ISO 14971 requires controls in priority order: design first, then protective measures, then warnings. If you can design out the overheating—automatic shutoff—a warning label isn't acceptable.
Risk management doesn't end at submission. Seal connects post-market data to pre-market analysis. Complaints link to hazards. When a hazard appears more frequently than predicted, the system flags it for re-evaluation.
