{"id":"orbital-evidence-status","level":"infrastructure","name":"Orbital computing evidence ladder","shortName":"Orbital computing evidence ladder","role":"Separates onboard processing, flight demonstrations, laboratory tests and proposed services.","description":"Computing in orbit spans different kinds of evidence. A payload can process its own sensor data, demonstrate an accelerator on a small mission or form part of a proposed distributed service. Starcloud reports an H100 flight and nanoGPT training demonstration, while Google’s Suncatcher study combines laboratory experiments and a proposed architecture. NASA’s TBIRD establishes an on-orbit communication result, not a compute-cluster result. Each supports a bounded statement. A scaled scenario must separately close energy, thermal, link, reliability and lifecycle budgets rather than treating one successful operation in orbit as proof of a continuously available commercial data center.","specs":[{"label":"Evidence categories","value":"Flight, bench test, simulation, plan and scenario","context":"Use the category associated with the specific result."},{"label":"Service boundary","value":"One demonstrated task versus ongoing accepted work","context":"Availability, failure recovery and access require additional evidence."}],"interfaces":[{"name":"Project result or proposal","direction":"in","signal":"Measured or reported task under dated conditions"},{"name":"Bounded conclusion","direction":"out","signal":"What was demonstrated and what remains unresolved"}],"tradeoff":"Small missions test assumptions affordably, but scaling changes heat rejection, communication, replacement and fleet-level failure exposure.","whyAI":"Onboard inference can reduce downlink demand for space-generated data; ground-user services have different ingress, latency and economic requirements.","related":["orbital-solar-storage","orbital-link-contacts","orbital-lifecycle-economics"],"sources":[{"title":"Starcloud-1","url":"https://www.starcloud.com/starcloud-1"},{"title":"Exploring a space-based, scalable AI infrastructure system design","url":"https://research.google/blog/exploring-a-space-based-scalable-ai-infrastructure-system-design/"},{"title":"NASA’s Record-Breaking Laser Demo Completes Mission","url":"https://www.nasa.gov/directorates/somd/space-communications-navigation-program/nasas-record-breaking-laser-demo-completes-mission/"}],"scale":"Spacecraft to proposed constellation; orbital geometry schematic","kind":"architecture","branch":"Frontier infrastructure","subbranch":"Orbital evidence and physical budgets","mechanism":["Name the payload, task, date and physical test environment.","Distinguish provider-reported outcomes from independently measured or published evidence.","List which subsystem constraints the demonstration actually exercised.","Keep proposed mission dates and constellation illustrations in planned or research states.","Evaluate the next scale using separate physical and operational budgets."],"misconception":"Works in orbit is not equivalent to qualified, economical, continuously available fleet-scale service.","check":{"question":"What does a ground optical bench test directly establish?","options":["A completed orbital network deployment","A universal orbital computing cost","Behavior of the tested ground link under its conditions"],"answer":2,"explanation":"The environment and system boundary constrain the conclusion; flight pointing, contact and reliability require additional evidence."},"science":[{"title":"Service readiness model","equation":"ServiceReady = Compute ∧ Energy ∧ Thermal ∧ Links ∧ Reliability ∧ Operations","explanation":"Each term is a supported capability for the same workload and operating envelope.","assumptions":"A conceptual gate model, not a certification or a prediction of commercial viability."}],"prerequisites":["serving","rack"],"learningObjective":"Classify orbital claims by evidence type and the capability actually tested.","workstream":"W12","visualFamily":"orbital-system","engineeringExample":"Synthetic evidence exercise: a ten-minute inference run verifies that workload on that payload for that interval. It does not establish one-year availability, a sustained ground-data path or fleet replacement cost.","evidenceNotes":"Starcloud’s page has inconsistent Gemini/Gemma wording, so only its H100 and nanoGPT statements are used. No universal first-AI-in-space claim is made.","labId":"library","scope":"Browse the numerical library; no current calculator directly models this lesson’s central mechanism.","labScope":"Browse the numerical library; no current calculator directly models this lesson’s central mechanism."}