## 2026-04-30 Morning Voicecast

Good morning. It's Thursday, April 30th, 2026. Let's get to it.

**Propulsion got its own revolution yesterday.** Astrobotic announced record-breaking hot-fire tests of its Chakram rotating detonation rocket engine at NASA Marshall Space Flight Center. Two engine prototypes completed eight successful tests, accumulating over 470 seconds of total run time — including a single 300-second continuous burn — without any discernible hardware damage. The engine generated more than 4,000 pounds of thrust. Now, rotating detonation engines aren't new in theory. A continuous detonation wave spirals through the combustion chamber, generating higher pressure and more thrust per unit of fuel compared to conventional steady-burn engines. But they've historically failed because nobody could build hardware that survives the extreme heat and pressure. Chakram survived a five-minute burn completely intact. If this scales up, RDREs could transform everything from lunar landers to orbital transfer vehicles. Astrobotic is aiming for cislunar operations.

**MIT has a new answer for AI models that lie confidently.** The problem with reasoning models is well-known: the more you train them for better problem-solving through reinforcement learning, the worse their confidence estimates become. A 2026 ICLR paper called "The Reasoning Trap" found that tool-hallucination rates increase in lockstep with task performance gains. MIT researchers have now published a training method that improves calibration of AI confidence estimates without sacrificing performance. It addresses one of the root causes of hallucination in reasoning models — and it does it without multiplying computational costs the way active learning does. In high-stakes domains like chip design or medical AI, knowing when a model is guessing is just as important as getting the right answer. This moves the needle toward the latter.

**NASA JPL fired up a Mars engine that was first conceived in the 1960s.** A prototype lithium-fed magnetoplasmadynamic thruster was tested at record power levels at the Jet Propulsion Laboratory in February. MPD thrusters generate huge electrical currents that interact with a magnetic field to directly accelerate lithium plasma — no chemical combustion needed. A crewed Mars mission would need 2 to 4 megawatts of total propulsion power. Current electric thrusters use inert gases and erode over time. Lithium-fed MPD thrusters don't have the same wear problem. Paired with a nuclear power source, this technology could slash launch mass requirements and enable the large payloads that human Mars missions demand. It's been sixty years of theory. Now they're getting actual data.

**And finally, the Pentagon just admitted the bottleneck.** Flight-test throughput is the binding constraint on America's entire hypersonic and missile development pipeline. The Defense Department awarded $150 million to modernize instrumentation at the Reagan Test Range in the Marshall Islands — the primary facility for validating hypersonic and reentry systems. The investment signals that capital is rotating toward the propulsion, composites, and validation platforms that can actually close the gap. When your test range can't keep pace with your design cycle, innovation doesn't matter. This is the infrastructure play.

That's all for today. See you tomorrow.