OrbitalHub

Where curiosity reaches escape velocity.

Domain is for sale. $50,000,000.00 USD. Direct any inquiries to contact@orbitalhub.com.

 

Multi-panel figure from the TOI-2535 b paper showing each transit shrinking and fading from 2019 to 2026, the transit chord sliding across the stellar disk, and the timing offsets growing to six hours

Figure 1 of Yee et al. (2026), reproduced under CC BY 4.0 from arXiv:2611.13343.

 

For seven years a planet called TOI-2535 b crossed its star every 8.24 days and dimmed it by a fraction of a percent. Then it stopped. The star is an old K dwarf about 1,000 light-years away in Corvus, catalogued as TIC 204698337. Nothing points to breakage or a fall inward. What the photometry shows is a planet whose orbital plane tilted until its path stopped crossing our line of sight. TOI-2535 b is the first confirmed planet orbiting a single star whose transits have been watched directly into disappearance (Yee et al. 2026, Astrophysical Journal Letters 1011, L60).

Every planet found by the transit method is a coincidence of geometry. A planet blocks its star only while its orbit is tilted within a few degrees of the sky plane. Astronomers call that tilt the impact parameter, written b: zero for a planet crossing the middle of the stellar disk, one for a planet clipping the edge. For nearly every world measured so far b is fixed by the orbit. TOI-2535 b’s b changed, and that change is the whole story.

The planet is what the field calls a hot Jupiter, though it weighs less than half of Jupiter. TOI-2535 b completes an orbit every 8.2358 days at 0.0993 au, carries about 0.47 Jupiter masses in a radius of 0.844 Jupiter radii, and sits inclined 90.8 degrees to our line of sight (Yee et al. 2026; exoplanet.eu catalogue, November 9, 2026). Hundreds of hot Jupiters are known, and their origin is still unsettled. Standard models build giants far from their stars, so a planet at 0.0993 au has to have migrated inward.

Its host is unremarkable: 0.864 solar masses, 0.877 solar radii, an effective temperature of 5256 K, and an age near 9.7 billion years (Yee et al. 2026, Table 1). Its space velocity of 114 km/s places it in the galactic thick disk. A planet around a star like this is not a freak of the exoplanet zoo.

The Warm gIaNts with tEss collaboration, a volunteer group that mines TESS light curves, flagged TOI-2535 as a community object of interest, and the MIT TESS team vetted it into an official one (Yee et al. 2026). TESS visited the field in five sectors between April 2019 and March 2026. The period listed in the catalogue, 16.4453 days, is double the truth, because one early transit fell into a data gap. Corrected, the planet crossed its star every 8.2358 days, every time.

Across that baseline the transit changed shape. The impact parameter measured 0.22 in the first sector, about 0.58 by sector 63 and 0.88 by sector 90, and the dip went from a deep, flat-bottomed notch to a shallow graze. Mid-transit times wandered too, running up to six hours early or late against a straight-line ephemeris, a 3 percent error on an 8.24-day year. Individual transit times were pinned to two or three minutes, so those excursions were signal. Ground follow-up filled the gaps between sectors, with fifteen attempts and six successes.

Velocities came from the NEID spectrograph on the WIYN 3.5-m telescope at Kitt Peak: 15 exposures of 600 seconds, a median precision of 11.2 m/s. The star’s reflex motion at the transit period came out near 108 m/s. Fit as though a single planet were responsible, that implies 0.99 Jupiter masses, more than double TOI-2535 b’s own mass; the paper calls such a fit illustrative only. The surplus is a clue. Speckle imaging on the SOAR 4.1-m ruled out an unresolved stellar companion, and a brown dwarf or star would have dragged the velocities by kilometres per second.

Then came the disappearance. TESS sector 101 closed the baseline in early 2026 and no transit appeared anywhere in it. Absence is harder to defend than presence, so the team built a detector for a planet that might not be there, S(b, δT_c) = χ²_flat - χ²_T(b)(δT_c), the fit improvement from adding a transit template at impact parameter b, shifted by δT_c. The largest value in the real data was 5.18. Running the same search on 20,000 synthetic twelve-hour blocks gave a 3-sigma threshold of 18.38. The planet was not merely overlooked. It was gone. A shallow transit at b = 0.991, clipping the limb, would have been recovered half the time.

The impact parameter follows b(t) = (a/R★)·(1-e²)/(1+e·sin ω)·cos i_sky(t). The orbit is nearly circular, so the prefactor barely moves and the drift comes almost entirely from the inclination. The inclination to the sky is the sum of two rotations, cos i_sky(t) = cos I·cos i_inv + sin I·sin i_inv·cos Ω_inv, where I is the angle between the sky and the system’s invariable plane. The dominant way for that inclination to swing is nodal precession, a rotation of the longitude of the node. The measured drift, ḃ = 0.16 ± 0.01 yr⁻¹, implies a precession rate faster than 0.5 degrees per year and a period of 30 to 50 years. The comparable systems TOI-4504 and TOI-216 take 250 and 150 years.

Several other mechanisms were tested and dropped, and the rejections narrow what the hidden object can be. Apsidal precession alone would move b by the ratio (1+e)/(1-e); matching the observed range from 0.2 to 1.1 that way needs e > 0.692, and it would shorten the transit to about 43 percent of its circular length, which the durations contradict. The star’s own spin would need a quadrupole moment J₂ ≳ 10⁻², far beyond an old, slowly rotating K dwarf. A distant, eccentric perturber is ruled out by arithmetic: precession this fast needs M₂ > 8.7 Jupiter masses for a one-year orbit, and such a body would show kilometre-per-second velocity swings. Because the observable rate scales with Ω̇_inv·sin i_inv, a face-on perturber would project weakly, so the geometry also sets a floor on the mutual inclination near 5 degrees. Nothing in the search reproduced the data below that floor.

The dynamical search behind those exclusions was brute force. Three hundred thousand starting configurations were scanned across period ratios from 1/50 to 10, concentrating on the neighbourhood of mean-motion resonances, then optimized against all three data sets. Good solutions clustered on the 2:1 and 3:2 commensurabilities. The hundred best were then integrated for a million years with WHFast in REBOUND. Two-thirds of them ended in a close encounter and an instability, and nothing near the second-order resonances survived. What remained were eccentricity-type resonances, where the angle ψ = 2λ_out - λ_in - ϖ librates instead of circulating. Across the surviving models the perturber’s period runs from 4 to 17 days, its mass from 0.16 to 0.54 Jupiter masses, and its mutual inclination from 7 to 33 degrees. Those masses predict velocity semi-amplitudes of 30 to 120 m/s, and fitting two-planet models cuts the required velocity jitter from 55 to 30 m/s.

That composition is the strangest part of the result. Apart from WASP-148, hot Jupiters with nearby planets keep sub-Neptunes or super-Earths for company, the usual outcome of high-eccentricity migration. Two giants on short orbits do not fit that story. Disk migration fits it, and resonant capture is what disk torques deliver, but a disk also damps mutual inclination quickly. Samuel Yee, the paper’s corresponding author at UCLA, puts it plainly: “Because TOI-2535 b has a close neighbor, it probably migrated by slowly spiraling in instead of getting flung in” (UCLA, November 2026). Something then tilted the orbits out of their birth plane. With no transits left to time, the next step is a high-cadence velocity campaign, and the models put TOI-2535 b back in front of its star in 30 to 50 years. A transit is not a property of a planet. It is a relationship between an orbit, a star and a line of sight, and gravity can change the last term.

 

November 19, 2026

A Robot Built Its First Truss in Orbit

Posted by

 

The ASTRA truss builder in low Earth orbit, Earth's cloud-covered limb behind it and the gold Kapton surface and composite truss bays in the foreground

Image source: Caltech/Momentus.

 

Gold kapton, black composite tubing, and a crowded little mechanism: from a distance the ASTRA payload looks less like a satellite than a workbench that got photographed by accident. That is roughly what it was. On the morning of March 30, 2026, the hardware rode a SpaceX rideshare out of Vandenberg Space Force Base, and for six months it worked in low Earth orbit on a task no machine had performed there before: joining a structure together out of loose parts, with no astronaut and no spacewalk anywhere near it. The machine was Caltech’s Autonomous In-Space Tensioned Reflector Assembly, and the structure it was building was the perimeter truss of a 1.4-meter reflector antenna (Caltech, November 5, 2026).

Every large structure ever flown has been shaped by the inside of a rocket fairing. A mirror too big to launch has to be a mirror that folds. Astro deployable structures, the Shuttle-era solar wings, the space station’s photovoltaic arrays: each trades stiffness and precision for a package small enough to pass through a payload shroud barely five meters wide. The other answer is to fly the pieces up separately and let people put them together, which is how the Shuttle assembled the space station and how Hubble was upgraded. Both routes have a ceiling, and the ceiling is the rocket. DARPA set out to attack that gap in 2022 with a program whose name is a mouthful: Novel Orbital and Moon Manufacturing, Materials, and Mass-efficient Design, or NOM4D. Its stated ambition is to ferry raw materials from Earth for in-orbit manufacturing, so that structures are designed for the environment they will live in rather than for ground test and launch survival (DARPA program page).

Program manager Andrew Detor put the scale question bluntly in the agency’s November 5 announcement: “By ‘very large,’ I mean hundreds of meters or kilometers in scale. Nothing approaching that scale can be achieved today” (DARPA, November 5, 2026). Caltech’s own reference architecture for a robotically assembled observatory is a 100-meter segmented primary, carried on more than 300 deployable truss modules that robots would build in Earth orbit before the system flew out to the Sun-Earth L2 point (Lee et al., Journal of Astronomical Telescopes, Instruments, and Systems, 2016).

The demonstration hardware came from Sergio Pellegrino’s group at Caltech, which manages NASA’s Jet Propulsion Laboratory, and rode to orbit hosted by Momentus aboard the company’s Vigoride 7 vehicle, which carried ten payloads of its own on SpaceX’s Transporter-16, a rideshare that delivered 119 payloads to sun-synchronous orbit (Caltech, November 5, 2026; NASASpaceflight, March 30, 2026). The payload itself was roughly one cubic meter. Inside it were two truss bays already built and folded for launch, an arm-like manipulator, and motorized dispensers holding the struts and joints needed for everything that followed (Caltech, November 5, 2026).

Operations began with releasing launch restraints, then pushing the two prebuilt bays out of the builder, then asking the machine to assemble a third bay from parts it had never touched. Commands came from the ground, sent by Charles Sommer, Alan Truong, and graduate student Tara Venkatadri, and the only way to know whether any of it had worked was to download photographs from the payload’s eleven cameras and compare them against what should have happened (Caltech, November 5, 2026). Anomalies were plentiful. The workaround was to re-create the failure on a flight-spare copy of ASTRA sitting in Pellegrino’s lab, solve it there, and send the revised recipe up. There were only about seven or eight contact windows a day, each roughly fifteen minutes long, so the machine spent twenty-two hours of every twenty-four on its own. Six months in, the antenna was not finished. The truss was designed for eleven bays, and the demonstration had added one.

Two design decisions explain why it worked. The first is the manipulator. DARPA describes it as a gantry robot able to move along three linear axes, and that constraint does more work than it first appears (DARPA, November 5, 2026). A free-flying arm in orbit has to hold its own pose against a moving vehicle, and place a part accurately while reacting against the very structure it is building. Bolted to a rail, the arm inherits its position from the machine carrying it, and a hard control problem collapses into repeatable Cartesian motion. The gantry is not clever. It is a decision to spend the complexity on the joints instead of on the arm.

The second decision is the surface itself. The reflective layer ASTRA deploys is a perforated polyimide film, Kapton, co-invented with a group at the University of Pennsylvania led by Igor Bargatin, and it functions as two cable nets prestressed against one another by springs (Caltech, November 5, 2026). That geometry matters because a tensioned net holds its shape through tension rather than through stiffness. For a cable spanning a distance L under a distributed load q, the out-of-plane sag runs roughly as δ = qL²/(8T), with T the tension per unit length. Halve the tension and the sag doubles; a structure needing less prestress can therefore be built with less material holding it up. Knitted metallic mesh, the standard in furlable reflectors, asks a lot of prestress. Perforated film asks less, so the truss behind it can be lighter. In a program whose premise is that a kilometer of structure must stay affordable to launch, shaving prestress is not a detail. It is the budget.

Scale the demonstration up and the numbers get interesting. Eleven bays span the perimeter of a 1.4-meter aperture, a circumference of roughly 4.4 meters, or about two and a half bays per meter of edge. Carry that density to a 100-meter aperture and you need some 800 meters of truss, a few hundred bays, which is the scale the RAMST reference design assumes (Lee et al., 2016). Put the aperture against the diffraction limit, θ ≈ 1.22λ/D, and a 100-meter mirror working at 500 nanometers resolves roughly 1.3 milliarcseconds, against about 50 milliarcseconds for Hubble. That is the prize, and it is why the concept splits the observatory into parts: the primary mirror, an optics and instrumentation unit a few hundred meters away, a control unit beyond that, and a spider-like six-armed robot, a hexbot, that assembles the truss and then walks across it to install the mirror segments.

What ASTRA established is narrower than the headline and more useful. One bay, built autonomously, in six months, with a human verification loop running about two hours a day, proves the process rather than the product. The complementary half of NOM4D runs in parallel: the other Phase 3 team, at the University of Illinois Urbana-Champaign, uses a station module to harden carbon tubes in orbit, making the material up there instead of merely putting it together (Caltech, November 5, 2026). ASTRA’s mission winds down in December, with the team already pointing at the next round of simplification: fewer failure modes, less ground support, shorter sequences.

The open question is whether the limit on building in orbit is dexterity or bandwidth. ASTRA suggests bandwidth. The robot could pick and place, but every fifteen-minute window had to carry diagnosis, redesign, and a retest, and better hardware does not erase that arithmetic. The fairing has always been the ceiling on what we can put in space. The next twenty years will test whether a gantry, a box of tubes, and a few thousand contact windows a year can lift it.

 

 

The 2025 astronaut candidate class poses with NASA's red letters in the Johnson Space Center atrium.

The ten chosen from the 2024 application pool, photographed at Johnson’s atrium in September 2025. Image source: NASA/JSC, jsc2025e075919_alt.

 

On Saturday, November 7, 2026, at a Kennedy Space Center air show partly about lunar exploration, NASA opened applications for its next class of astronauts. The window shuts at 11:59 p.m. Eastern on December 3; the class will be named in the fall of 2027 (NASA, November 2026). Administrator Jared Isaacman framed it in the language of current agency plans: the people selected could walk on the Moon, live at the Moon Base, and help build the systems that carry crews to Mars. Underneath that language sat an arithmetic problem NASA has never solved. The last time it opened this door, in 2024, more than 8,000 people walked through it and ten came out. NASA has selected 370 people to train as astronauts since the 1960s, and 32 are active today.

Thirty-two is a thin roster to run a program that includes a lunar outpost, a commercial station economy in low Earth orbit, and a road to Mars. It is also a number no press conference can raise, which makes the machinery underneath worth reading. Applicants must be US citizens and hold a master’s degree in engineering, biological science, physical science, computer science or mathematics, or meet that requirement another way: two years of work toward a doctorate, a medical degree, or enrollment in a nationally recognized test pilot school finished by July 2027. Three years of related professional experience follow. Pilots may substitute flight time, and the menu is oddly exact: 750 hours as pilot in command in a high-performance jet, 1,500 total fixed-wing hours including 750 as pilot in command, or 1,000 rotary-wing hours including 300 as pilot in command.

The schedule after that is fixed. Applications are read from December 2026 through April 2027, a first group comes to Johnson Space Center in spring 2027 for interviews and assessments, a second group comes in summer 2027 for final interviews, and the class is announced in the fall of 2027 (NASA astronaut selection timeline). The winners then spend two years learning to spacewalk, operate the station and fly T-38 jet trainers.

Everything so far is a paper filter. The one that bites is the long-duration astronaut flight physical, and NASA’s medical technical brief explains why it is built the way it is. During selection, applicants are screened for a list of disqualifying health conditions defined in OCHMO-STD-100.1 and are eliminated with no waiver considered at all. Waivers only enter the picture after selection (OCHMO-MTB-002, October 2026). That arrangement looks arbitrary until you see where it came from. NASA wrote its first minimal medical standards for astronaut selection in 1977, borrowed from Air Force, Navy, Department of Defense and FAA rules. Across the shuttle era, screening grew from a one-page medical sheet read by one flight surgeon into a 30-plus-page history backed by test data and presented to a selection board. The disqualifying list is revised continuously, because conditions that once ended careers have become treatable or waivable.

The clearest case of the standard being rewritten by data is eyesight. Until 2007 NASA astronauts could not have laser eye surgery, and before that rule changed, roughly half of all rejected candidates had been turned away for needing glasses or contact lenses (OCHMO-MTB-002). Soft lenses and wavefront-guided LASIK and PRK widened the acceptable parameters, and the requirement for uncorrected distant acuity has been relaxed ever since. Vision matters more than that history suggests, because of a hazard nobody had identified when those rules were written.

Spaceflight-associated neuro-ocular syndrome was first described in 2011, from seven astronauts who had spent months aboard the ISS. Five had developed a hyperopic shift of +0.50 to +1.75 diopters, and six showed optic disc edema, choroidal folds, flattening of the back of the eye or thickened retinal nerve fiber layers on optical coherence tomography (Eye, 2026). The same review now reports at least one feature of the syndrome in 81 percent of NASA astronauts, and in a questionnaire of three hundred astronauts, 60 percent of long-duration crew reported subjective vision changes. Among fourteen astronauts and cosmonauts, acuity fell by an average of 0.75 lines on the eye chart a day after return. NASA specialists published a consensus guide on diagnosing and managing it this year.

The name hides how ordinary the mechanism is. On Earth, intracranial pressure sits near 11 mmHg lying flat and turns slightly negative standing up, because posture shifts the pressure in the dural sinuses. In microgravity that hydrostatic gradient disappears. Venous pressure stops depending on position, so intracranial pressure in orbit should land above the upright value on Earth and close to the supine value instead (npj Microgravity, 2026). Hold a crew at that pressure for months and the optic nerve elongates, pressing on the rear of the eye and producing globe flattening, choroidal folds and disc edema. A 2021 follow-up on three of the original seven astronauts found the hyperopic shift, the flattening and the change in axial length still present up to twelve years later.

Measuring it in orbit is hard. Lumbar puncture is not feasible in space and implanted sensors need neurosurgery, so the least invasive option is a lumbar catheter wired to a telemetric device. A bench test of six Miethke M.scio sensors with a Medtronic catheter set found that system reading about 2.62 mmHg low against a reference column (npj Microgravity, 2026).

The gap between what a standard can clear and what an instrument can resolve is where the waiver process operates. Once someone is an astronaut, a condition that would have ended an application can be waived at recertification, on the recommendation of NASA’s Aerospace Medical Board. The Chief Health and Medical Officer signs the final disposition. The board weighs one diagnosis against the physics of the job, and NASA’s list of factors is specific: fluid shifts and bone loss in microgravity, decompression sickness and barotrauma under pressure change, extra loads on the spine at launch and landing, and time far from medical care (OCHMO-MTB-002). Waivers on the list include a recovered hip fracture, mild sleep apnea and a history of elevated intracranial pressure. For atrial fibrillation, NASA bypasses the FAA’s certificate process, treats with ablation and clears a crewmember for flight six months later.

What happens next is fairly predictable. The class of 2027 reports in the winter of 2027/28 and trains for two years, so the crewed lunar landing NASA has been discussing for 2028 will be crewed from the 2024 pool, not the 2026 one (The Guardian, November 2026). The class being hired now would fly in the 2030s, and a class every three years is a trickle for a program meant to keep people on the Moon. The limiting factor is the supply of people who clear the medical standard, and how quickly that standard absorbs what the crews bring back. SANS was an obscure finding in 2011 and a consensus-guideline subject by 2026, because enough astronauts flew long enough, and lived long enough afterward, to make the data legible. The same runs in reverse: every year at a lunar outpost generates measurements nobody has yet, and the document that decides who flies gets edited to fit them. The people who get to go will have been chosen by a standard that began as a 1977 Air Force checklist and has been rewritten every time the ship, the mission or the human body outgrew it.

 

 

NASA artist rendering of a Tracking and Data Relay Satellite with twin mesh reflectors and solar arrays passing above the blue limb of Earth

 

On November 4, 2026, NASA handed three companies a total of up to $1.07 billion and an odd-sounding problem: build a relay service in the Ka microwave band that spacecraft already in orbit can use without changing anything on board (NASA, Nov. 2026). None of the money buys a telescope or a rover. It buys plumbing. Almost none of the data coming home from space comes straight down. Most of it hops first through a satellite parked in geostationary orbit, 35,786 kilometers above the equator, and for more than forty years that hop belonged to NASA’s Tracking and Data Relay Satellites, TDRS. The agency has already told the fleet’s users it is leaving. Astranis Space Technologies, Global Enterprise Solutions and Viasat have been asked to build what replaces it, and to make sure the radio already flying still works.

The relay architecture was NASA’s answer to a geometry problem. A ground antenna sees a low Earth orbit spacecraft for only a few minutes at a time, so most missions talk to Earth through a bent-pipe chain of ground stations and orbiting relays. TDRS-1 launched in 1983, twelve have flown since, and seven were still active in 2025 across three generations, each with more automation and, in the third, working Ka-band equipment (NASA, July 2025). More than 25 missions have leaned on the system, from the International Space Station to the Hubble Space Telescope. As NASA Near Space Network chief architect Dave Israel put it, every astronaut conversation and every Hubble image “has flowed through TDRS” (NASA, Nov. 2024).

The network around those relays is larger than the satellites themselves. The Near Space Network pairs the space segment with more than 40 government and commercial ground antennas, including the sites at White Sands, New Mexico and Guam that TDRS uses, and moves an average of nearly 15 terabytes of data a day for NASA missions (NASA Near Space Network services brochure, Rev. 2.2). In November 2024 NASA stopped assigning new missions to TDRS, set the fleet toward retirement, and said existing users, the station and Hubble among them, would ride it into the mid-2030s (NASA, Nov. 2024). The economics are plain: the first TDRS was built for ten years of life and lasted twenty-six, and seven aging spacecraft are expensive to keep, patch and insure.

The November 4 award is the awkward middle of that transition. It sits under Next Space Technologies for Exploration Partnerships-3, Appendix E, a mechanism built for public-private partnerships rather than ordinary procurement, and its project name is NEXUS, for Network Extension for User Continuity and Sustainability. NASA wants an end-to-end Ka-band relay service, space segment, ground, launch, integration and operations included, backward compatible with legacy TDRS users for at least fifteen years. The solicitation states the reason without much decoration: some on-orbit missions cannot feasibly modify their flight hardware or move to incompatible commercial services, and continuity risk grows in the 2029 to 2031 window as the government fleet declines (NASA, Nov. 2026). The three firms, Astranis at $264.3 million, Global Enterprise Solutions at $340.4 million and Viasat at $460.9 million at maximum option value, start with a firm-fixed-price period through May 3, 2027, with priced options extending the work to 2029 and the option of an in-orbit flight demonstration.

Three winners instead of one is deliberate. NASA wants a market rather than a vendor, and the three approach the problem from different directions. Global Enterprise Solutions is the U.S. government arm of SES, which bought it from Leonardo DRS for $450 million in 2022 (SES, Aug. 2022), and SES already flies O3b mPOWER, a Ka-band network in medium Earth orbit that has been relaying satellite data in flight demonstrations with Planet Labs. Viasat took over Inmarsat in 2023 and brings a geostationary fleet plus a demonstrated L-band launch tracking service used on Blue Origin’s New Glenn. Astranis is the small-satellite case, planning clusters of Omega spacecraft about 90 degrees apart in geostationary orbit (NASA SCaN, 2025). NASA handed $278.5 million to six partners in 2022 to demonstrate commercial relay services, and the demonstrations now under way range from SES radio links to optical links over Amazon Leo, SpaceX Starlink and the coming Telesat Lightspeed network (NASA, Dec. 2025).

Ka-band is where this gets physically interesting. It covers roughly 26.5 to 40 gigahertz, high enough to carry orders of magnitude more data than the S-band and X-band plumbing most legacy spacecraft were built around, and high enough that the physics punishes you for it. Free-space loss follows FSPL(dB) = 20 log10(d) + 20 log10(f) + 32.45, with distance in kilometers and frequency in megahertz. Moving a link from 8.4 GHz to 32 GHz costs about 12 dB of path loss over the same geometry, so the receiving antenna has to work harder for the same gain: G = η (π D / λ)^2 means that at four times the frequency a dish must be one quarter the diameter to hold the same beam. NASA has run Ka-band at 32 GHz on the 70-meter dishes of its Deep Space Network for decades, because out there the extra margin buys more than the extra bandwidth does (NASA NTRS, 1987). Rain does the rest of the damage. Water absorbs and scatters far more strongly at these wavelengths, so Ka links are designed with extra margin, adaptive coding and usually a lower-frequency fallback.

The trade shows up in NASA’s own service sheet. The Near Space Network’s return path tops out at 3.5 gigabits per second over Ka-band, the figure NASA lists for NISAR’s dual-polarization science data, while the forward command path over Ka runs 40 megabits per second across 40 megahertz of bandwidth. The receiving system figure of merit sits between 38 and 47.5 dB/K, 41.3 typical, and maximum return bandwidth on the service is 1,500 megahertz (NASA NSN services brochure, Rev. 2.2). Those are the numbers a commercial operator now has to promise for spacecraft whose radios were frozen into the hardware decades ago.

Backward compatibility is the constraint that shapes everything else. A compatible service has to accept the modulation, symbol rates and coding that legacy terminals already speak, deliver data with the timing and routing those missions expect, and keep doing it for fifteen years while its customer base shrinks. NASA is not commissioning a spacecraft to fly. It is asking for a business that can raise its own satellites, ground stations and launch contracts, then sell the service to NASA and anyone else who needs it. That is why the awards include launch services and operations rather than only hardware, and why NASA insists it will not be the only customer. Building a government-owned Ka-band fleet, the alternative, is precisely what the agency is trying to stop owning.

The next two years show which of the three can actually build it. NASA will down-select on demonstrated performance, technical credibility and commercial viability, and the decision has to land before the fleet’s continuity gap opens in 2029. New missions have already been pointed elsewhere, toward optical links and commercial constellations where the relay sits far closer to Earth and the light-time penalty falls from about a tenth of a second each way to milliseconds. The stubborn residue is the old hardware: a spacecraft built around a radio NASA no longer wants to own, still in orbit, still returning data. Somehow the replacement has to speak its language for longer than the machines that will use it.

 

 

NASA's generated concept rendering of the X-70 Fractal, a dark experimental jet with canted twin tails emerging from a bank of vapor

 

On the morning of November 4, 2026, at the UP.Summit event at Kennedy Space Center in Florida, NASA revealed a shape hiding in a bank of vapor: a dark single-seat aircraft with canted twin tails, a bubble canopy, and wings that droop a few degrees below horizontal. It is the X-70 Fractal, the agency’s newest experimental aircraft, and the picture NASA released with it is labelled as generated content (NASA, Nov. 4, 2026). That label measures how much of this airplane is design and how much is promise.

The numbers that came with it are few and specific. Takeoff weight of 10,000 pounds, cruise at Mach 0.75, a ceiling of 45,000 feet, range a little past 1,500 nautical miles, room for up to 2,000 pounds of experimental payload mounted inside the airframe or hung off it, and a first flight under NASA ownership expected within two years. The crew is one person (Wikipedia). Missing are the engine count, the wing geometry, the contractor’s name and the list of control laws the aircraft is meant to test.

An X-plane answers a question that wind tunnels and computers cannot. The lineage runs from the Bell X-1 in 1947 through the X-15 at Mach 6.7 in 1963 to NASA’s X-59, first flown on October 28, 2025 to make supersonic flight tolerable over land (NASA). Crewed X-planes stopped for decades, and Administrator Jared Isaacman has made reviving them a priority. At the All-In Summit in September 2026 he showed a slide called “Rebuilding our X-plane portfolio” with an aircraft emerging from clouds; asked whether it was a Blackbird, he replied, “You sure that’s a Blackbird?” (The Aviationist, Sept. 28, 2026). Days later, on NASASpaceflight’s podcast, he said NASA was “just cleaning it, man,” a joke read as a promise about SR-71 844, quietly removed from its display pad at Edwards. The X-70 turned up instead.

The X-59 gives the program a live example to point at. One year after its first flight the quiet supersonic demonstrator has flown 40 additional sorties, climbed above 56,000 feet and reached Mach 1.53, and it sits at Kennedy Space Center this week for NASA’s Max Power air show, which runs November 6 through 8 (NASA Quesst blog, Oct. 28, 2026).

The interesting word in NASA’s announcement is not “Fractal” but “reconfigurable”. The release describes rapidly reconfigurable flight-control technology intended to let the agency explore maneuverability in ways never before demonstrated in a crewed flight (NASA, Nov. 4, 2026). Two thousand pounds of swappable payload, inside and outside, is the same idea in hardware: change the equipment, fly, log data, change it back. Laurie Grindle, who runs NASA’s Aeronautics Division, put it plainly: “For decades, NASA Aeronautics has used X-planes to develop technologies that advance aviation” (NASA).

What was announced is a research platform with a specification sheet rather than a design, and the gaps are revealing. The aircraft will be managed by NASA’s Armstrong Flight Research Center in Edwards, and Isaacman said it will fly “under the NASA banner” within two years, implying the airframe currently sits somewhere other than NASA with a contractor the agency has not named. Aviation Week made the same point: beyond weight, cruise speed, altitude and range, “further details were not disclosed, including the contractor” (Nov. 5, 2026). Off the rendering, the trade press could read a slight negative dihedral, canted tails and a bubble canopy. Whether the intakes sit under the wing or beside the fuselage, nobody outside the agency is saying.

NASA has not said what will be flown. The Aviationist points to the two areas where flight-control research is concentrated: shape adaptation, in which a wing changes form in flight, and active flow control, which replaces moving surfaces with jets of air that push on the airframe and change how it moves (The Aviationist, Nov. 5, 2026). Active flow control has an aircraft of its own coming: DARPA’s X-65 CRANE, built by Aurora Flight Sciences to maneuver by blowing bursts of compressed air and moving no exterior surfaces at all, was mated with its wings in Virginia in June 2026 and is now working toward a first flight in 2027 (Aurora Flight Sciences; The Aviationist, June 24, 2026).

Control authority is a physics problem before it is a software problem. At Mach 0.75 and 45,000 feet the speed of sound is about 328 metres per second, so the aircraft moves at roughly 246 metres per second through air with a density near 0.0024 kilograms per cubic metre. The dynamic pressure comes out around 73 pascals. A 10,000 pound airplane carries about 44,500 newtons, so at a cruise lift coefficient of 1.0 the wing has to cover roughly 600 square metres to hold it up. That is a large wing, and it has to be large because the air up there is thin. Wing loading sets how big the control surfaces must be, so a testbed built to carry a 2,000 pound payload on a big, slow wing is a machine designed to be flown to its edges.

The physics gets interesting at high angles of attack, where the tail stops helping. Past roughly 15 to 20 degrees the horizontal stabilizer sits in the wing’s wake, its effectiveness collapses, and the airplane that stalls is the airplane that cannot be recovered. That is the gap thrust vectoring was built to close. The Rockwell-MBB X-31, the first international X-plane, steered its single GE F404 exhaust through three paddles and held controlled flight at a 70 degree angle of attack on November 6, 1992, then used a post-stall Herbst maneuver to pull a minimum-radius 180 degree turn in April 1993 (Wikipedia).

The other route to control power is to let the structure move. NASA’s X-53 Active Aeroelastic Wing took a modified F/A-18 and used airflow over the wing to twist it, reviving the Wright brothers’ wing-warping idea. Its second flight phase covered 18 test points between Mach 0.85 and Mach 1.3 and produced roll rates within 15 to 20 percent of a production F/A-18’s using the flexible wing alone (NASA, X-53 project page). The payoff was structural: wings thin enough to twist without failing, and therefore 10 to 20 percent lighter, for about 45 million dollars over eight years. An aircraft that can move between aerodynamic, thrust-vectored and structural control lets those control laws be compared in the same air, in the same week, against the same pilot.

What happens over the next 24 months will say whether this is a research program or a publicity exercise. Max Power runs this weekend at the Shuttle Landing Facility, where to watch for the real shape of the airplane, if it appears at all. Then watch for a named contractor, a propulsion choice and a range. Watch the wording too: “first flight under the NASA banner” is a lower bar than first flight, and the gap between those phrases is where the interesting questions live.

The X-plane idea endures because control laws, unlike airframes, can be revised between sorties. An X-plane lets an agency fly a control system hundreds of times, log how it behaves in a gust, at a high angle of attack, with a failed sensor, and then write a better version. That loop is what turned X-1 data into supersonic airliners and X-15 data into spacecraft return stages. The X-70 does not need to be a beautiful airplane. It needs to be a fast, honest instrument for measuring how an airplane should be controlled, and NASA is betting that the next twenty years of flight control will be written on it.

 

 

The Matroshka AstroRad Radiation Experiment phantoms Helga and Zohar are installed in the passenger seats of the Artemis I Orion crew module at NASA's Kennedy Space Center, each torso packed with radiation detectors

 

Two mannequin torsos were strapped into the passenger seats of an Orion spacecraft at Kennedy Space Center in Florida in August 2022, and for 25 days they flew out past the Moon and back. They had no skin, no blood and no nerves. What they carried was the densest survey ever flown of what ionizing radiation does to a human body outside low Earth orbit: thousands of detectors buried in slices of tissue-equivalent material built to mimic bone, organ and skin, plus more than ten thousand small crystals logging every particle that passed through them (IFJ PAN, October 2026). The flight ended in December 2022; the readout is still running. A paper in Radiation Measurements published this past September put the smallest of those crystals, chips of lithium fluoride a few millimetres across, to work on a question that sounds simple and is not: how much radiation does a person actually absorb near the Moon, compared with a person working a few hundred kilometres up over the equator? The Moon is the worse place, and by a smaller margin than intuition insists.

Three particle populations decide what a crew receives. Galactic cosmic rays are the steady background, mostly protons with a thin tail of heavy nuclei at energies past 10^20 eV. Solar energetic particles arrive in bursts after a flare, mostly protons in the tens of megahelectronvolt range. The Van Allen belts hold geomagnetically trapped protons that dominate the radiation budget in low Earth orbit, then vanish as soon as a spacecraft climbs out of them. The three behave very differently against shielding. A few centimetres of water knocks down most of a solar energetic particle and does nothing useful against a galactic cosmic ray, which carries enough energy to cross metres of metal, plastic and water while showering secondary particles.

The useful question is not how much radiation exists but how much a body in a specific place collects. Inside the space station, the earlier Matroshka experiment found the dose piled into shallow tissue near the surface, where low-energy trapped protons dumped their energy early (IFJ PAN, 2026). NASA caps an astronaut’s career effective dose from spaceflight at 600 mSv and selects crews so their lifetime risk of radiation-induced death stays low (NASA-STD-3001, Volume 1). Every millisievert on that ledger is an engineering problem paid for in kilograms.

Artemis I was the first chance to measure any of it with a body in the room. Orion launched on 16 November 2022, flew past the Moon on 21 November at 130 kilometres and again on 5 December at 128 kilometres, and splashed down off San Diego on 11 December (Nature, December 2024). The Matroshka AstroRad Radiation Experiment, led by the German Aerospace Center with NASA and the Israel Space Agency, carried two headless female phantoms built from 38 slices of tissue-equivalent material. Helga flew bare, Zohar wore a prototype vest called AstroRad, and both torsos held more than 12,000 detectors between them (ESA, August 2026).

The first results, published in Nature in 2024, showed how much the spacecraft shapes the environment: dose rates inside Orion varied by a factor of four between the most and least shielded positions, and rotating the capsule during the inner-belt pass halved the dose rate as the thick axis swung into the proton plane (Nature, December 2024). Total absorbed dose across the mission reached 13.47 mGy at the least shielded location, and the two belt crossings still contributed up to 23 percent of that. No solar particle event occurred during the flight, and the only other change was the Moon itself, whose bulk blocked about a third of the cosmic ray flux during the close passes.

Two papers this year fill in the rest. A Science Advances study in August modelled the August 1972 solar storm, which fell between the Apollo 16 and Apollo 17 missions, and found the AstroRad vest could cut effective dose by 58.2 percent, worth up to 193 days of deep space exposure (ESA, 2026). Then in September, Pawel Bilski and colleagues at the Institute of Nuclear Physics of the Polish Academy of Sciences reported the first quantitative use of fluorescent lithium fluoride track detectors from the lunar flyby (Bilski et al., 2026). The crystals are millimetre-scale chips grown by the Czochralski method, pulled slowly from a melt of purified lithium fluoride. Ionizing radiation leaves defects in the lattice that act as color centers, and under blue light those centers re-emit red fluorescence. Under a microscope, a heavy particle appears as a scratch in the crystal marking where energy was deposited.

Reading the pictures was easy. Extracting numbers was not, because cosmic rays arrive from every direction, so tracks wander out of the focal plane and blur into a background glow the camera collects with the sharp segment. Highly ionizing particles cut continuous lines, weakly ionizing ones leave dotted chains, and both change the fluorescence intensity that carries the ionization density. After corrections to subtract that background, the lithium fluoride results land within 15 percent of conventional etched track detectors across the overlapping energy range, and they pick up tracks below those detectors’ threshold (Bilski et al., 2026). The Krakow group also supplied nearly 600 thermoluminescent detectors, chips that store absorbed energy in the lattice and release it as light when heated.

The result underneath the instrument work is the headline: the dose inside the phantoms near the Moon came out essentially uniform, where the station data had been lumpy (Bilski et al., 2026). Trapped protons in low Earth orbit stop at different depths in a body, so they pile dose into the outer tissue. Galactic cosmic rays carry enough energy to cross a whole torso and deposit only a fraction of it along the way, so their dose spreads out almost evenly. The Van Allen contribution disappears near the Moon and the galactic contribution replacing it is larger, so the total goes up; the two effects partly cancel, and the margin is not the jump the geometry of the belts suggests. Dose equivalent still has to be computed with radiation weighting factors, H = sum_R w_R D_T,R, where protons count as one and alpha particles and heavy nuclei count as twenty, because heavy nuclei deliver a disproportionate share of biological damage per unit of absorbed dose (NASA-STD-3001, Volume 1). By that measure a crew on a lunar mission would spend a small fraction of a career’s allowance.

What the uniform dose map changes is design. If the dose is flat, shielding the skin buys little and every kilogram has to go somewhere with real mass: a storm shelter, a water-walled bunk or a berm of regolith on the surface. If the dose is lumpy, as it is in low Earth orbit, the same shelter can be tens of centimetres of structure parked behind the seats. The Moon is also a useful if temporary shield for the rest of the sky: during the Artemis I flybys it cut the cosmic ray dose by about a third simply by being in the way, one more argument for keeping habitats and crewed vehicles apart.

None of this replaces a crewed measurement. Artemis II carried four astronauts around the Moon in April 2026, the first people to leave low Earth orbit since Apollo 17 (NASA), and the missions behind it will wear the dosimeters these crystals have just calibrated. The value of the two papers this year is that they turn a vague dread into a number with an uncertainty attached, the only form in which radiation risk can be engineered against.