In a pitch-black tunnel the drone stopped trusting its own eyes, flipped RED, and flew itself home instead of into the wall. That refusal is the whole point of Lodestar, our navigation integrity engine for GPS-denied flight. When jamming kills GPS, most systems keep reporting a position with full confidence even when the scene has gone unobservable and the position is simply wrong. In contested airspace a confident-but-wrong fix is more dangerous than an honest "I don't know." Watch it on an 80-second seeded synthetic mission (no real drone, no real flight log, the R-330Zh is only a named jamming scenario). The flight crosses into an EW bubble at t=18s and GPS drops. Lodestar holds to about 0.9% drift over the 245.3 m path, where plain dead-reckoning blows out to 31%. Then the path enters a dark tunnel and the visual features collapse to near zero. Now the decision you can watch. A Navigation Integrity Monitor that lives outside the estimator reads feature count, position uncertainty, and innovation consistency, then flips the state GREEN, AMBER, or RED. When the tunnel goes truly unobservable it latches RED and triggers return-to-home at t=52.5s. It does not bluff a fix. On this run it flagged the unobservable stretch every time, with near-zero false alarms on the healthy GPS-denied leg. Why it holds: trust must not depend on the thing being trusted. The gate is deterministic code outside the estimator, so it stays valuable no matter how good the underlying VIO gets. One click exports a Flight Integrity Report with an honest scope disclosure of what is simulated versus deferred. If your team is building autonomy that has to survive GPS denial, we would genuinely like to hear how you think about the moment a system should admit it is blind. 🧭 #GPSDenied #DroneAutonomy #DefenseTech #BlueUAS #SLAM