Simulating Mission Control Instead of Another Dashboard
Most dashboards report. Mission control systems evaluate. That difference shaped this project. I was less interested in displaying exoplanet data than building an interface where ranking, uncertainty, simulation, and analysis felt like one operating system.
That became AURA Mission Control v2. A full stack simulation environment for assessing exoplanet colonization viability through a Composite Planetary Index, live telemetry, and AI generated mission reasoning. It was deliberately overbuilt in some places. That was part of the point.
The Core Problem Was Not the UI
The hardest problem was deciding what the system was actually scoring. A ranking system based on planetary viability can turn into arbitrary weighted math very quickly. The CPI model forced structure. Gravity, temperature, atmosphere, magnetic field, and resources all contributed, but weights were adjustable so assumptions could be challenged.
That decision made simulation interactive instead of static. Change the mission priorities and the rankings shift. That was much more interesting than hardcoding a best planet.
- A backend simulation engine computed CPI rankings and confidence related status metrics
- Weight tuning triggered live reranking rather than cosmetic UI updates
- Time series telemetry with synthetic noise made uncertainty part of the interface
The system started feeling less like a data viewer and more like a model you could interrogate.
Combining Visualization Layers
A lot of the project was about mixing representations without making them feel disconnected. Recharts handled planetary factor breakdowns. React Three Fiber handled spatial presence. Custom telemetry overlays showed observed versus predicted data. The reasoning log made AI output feel procedural rather than magical.
One non obvious challenge was getting those pieces to feel like one system. A pretty 3D planet beside unrelated charts would have been decoration. The visualizations had to support the same analysis loop.
The reconstruction transition component came out of that thinking. It tried to visualize inference itself, not just results.
What Did Not Work Perfectly
Some ambition exceeded what the data justified. The observational streams look realistic, but they are simulated. That matters. Mock data can support a convincing interface while hiding whether the underlying model deserves that confidence.
The AI integration also has tradeoffs. Structured OpenRouter output works well, but relying on an external model introduces fragility and latency. I spent more time than expected handling malformed responses and failure states.
I would also simplify parts of the interface. There is a lot happening. Some panels compete for attention. If I rebuilt it, I would reduce complexity before adding more realism.
What I Learned
The project taught me that simulation software lives at the boundary between modeling and interface design. Good engineering there is not just backend correctness or frontend polish. It is deciding how uncertainty should appear.
I also learned restraint matters. Some of the best parts were not the flashy pieces. They were the weight system, the confidence metrics, and the architectural separation that kept the system coherent.
Mission control is mostly disciplined reasoning under uncertainty. That was the real project.