The proof layer.
Constanellis designs, builds, and operates cislunar infrastructure. This page is the engineering face of that work: the measurement chain we are building, the reference data it produces, the public standards our interfaces are designed toward, the autonomy stack meant to run at the edge, and the closed-loop surface research sitting furthest out. Each section states its own maturity. Nothing described here has flown.
Every number on this page is published elsewhere on this site and repeated verbatim. Where a fact is not confirmed, the copy says so.
Three convictions the engineering follows from.
Own the proof.
Hardware gets built either way. What decides whether it flies is whether anyone can say, with evidence, what it does in the environment it will actually see. The measurement, the uncertainty budget, and the traceable record behind them are the scarce goods, and they are the position we are building toward.
Think in closed loops.
An outpost is resupplied. A settlement makes itself. Every architecture decision we make is judged on which loops it closes, power, thermal, materials, and data, and on what the open ones cost forever. An open loop is a bill that arrives on every mission after this one.
Publish what we find.
Reference data belongs in the open, and methods belong next to the data that produced them. We intend to publish both, and to bring measurements rather than opinions to the interoperability conversation, because a standard gets written by whoever arrives with evidence. We publish results when results exist.
An optic that is right on the bench can be wrong in vacuum.
Glass does not care what the drawing says. It responds to temperature and to whatever is holding it. Take an optical assembly that closes to specification on a bench at room temperature, pump the chamber down, and walk it out to the cold of a shadowed lunar crater. The mount contracts against the substrate. The bond line shrinks. Coating stress changes. The alignment you certified in the clean room is not the alignment the mission gets, and nothing failed: the part was simply measured in the wrong place.
The commercial answer to this today is screening, which proves the article survives the environment. Survival is not performance. A retroreflector that comes out of a thermal cycle intact tells you nothing about where its beam is pointing while it is cold, and pointing is the entire function of a retroreflector. The measurement that answers the real question is harder to stand up, because the instrument has to see the article through the chamber, and the calibration has to survive the trip.
That chain is what FORGE is being built to operate. Every link in it is ordinary engineering. The value is that the links sit in one building under one quality system with an unbroken calibration path to national standards, so the number on the certificate is a number a customer can put straight into a model without discounting it first.
Values repeated verbatim from the FORGE capability targets, where they are published under the same qualifier. Chamber, cleanroom class, and tolerance-band specifics are staged for publication as commissioning completes.
The measurement chain, link by link
The question, written down first
Nothing is measured until the question exists on paper: which parameter, at which condition, to which tolerance, and what disposition follows from the answer. A test run without a written acceptance criterion produces a number, not a result.
Vacuum and temperature, held
The environment is the point of the whole exercise. Pressure is driven to the regime the hardware will see and the article is walked across the temperature range it will see, with the fixture treated as part of the article, because the fixture moves too.
Measured in place, not after
Wavefront, alignment, and radiometric response are captured while the article is still in the environment. A measurement taken after the chamber returns to ambient describes the recovery, not the condition anyone asked about.
Traceability to national standards
Every instrument in the setup sits on a calibration schedule with an unbroken path to national standards. Traceability is not paperwork about the measurement. It is the reason the measurement means anything outside the building that produced it.
A value without a budget is an opinion
Each result carries the uncertainty budget that produced it: instrument, fixture, environment, operator, and the terms that could not be separated. Stating the budget is how a customer decides whether our number is good enough for their margin.
Two outputs, one measurement
The customer receives the data package. The corpus retains the calibrated record. One is the delivery. The other is the thing that compounds, and it is why the lab is worth operating between missions.
Every measurement we sell leaves a copy behind.
A qualification job has two outputs. The customer gets the data package they paid for. We keep the calibrated record of how a real article behaved under real conditions: the drift, the fixture effects, the population statistics that handbooks approximate and nobody publishes. That record is worth more in year five than in year one, because each campaign narrows the uncertainty on the next one and lets a test be designed against observed behavior instead of a table.
The counters below are the live ones, read from the same source of truth that feeds the corpus page. They read zero. We publish them at zero because a company selling measurement has no business rounding its own telemetry up, and because a counter that starts honest is the only kind that stays honest. When the floor qualifies its first article, the number moves and the review date under it moves with it. Until then, zero is the reading, and showing it is the point.
Counters are real and start at zero. They move when the floor does, not when the roadmap does. Last reviewed 2026-08-01. The measurement corpus
The intent is to give the reference data away: methods published alongside the datasets, uncertainty stated rather than implied, and a stable identifier on anything we ask another engineer to cite. The publication license is being set before the first dataset is released, so that anyone building on this work knows their rights the day it appears.
Positions we intend to hold, stated as intent.
A network only its owner can use is a private radio, not infrastructure. The public interoperability work for lunar operations is already underway in the open, and the architecture is being designed against it so that a customer terminal can roam across compliant networks instead of being rebuilt for each one.
What follows is engineering direction, not status. We are not describing systems that interoperate today, because none of ours are operating.
Design the EVERLIGHT user segment toward LunaNet
The LunaNet Interoperability Specification (LNIS) is the public description of the service interfaces a lunar user terminal is expected to speak. We intend to design against it, to test against it, and to publish where we conform and where we do not, in that order.
Speak the common language at the link and file layers
CCSDS recommendations are the shared vocabulary of space data systems. We intend to adopt them rather than invent private formats, because a private format is a bill the customer pays later, every time they integrate anything else.
Assume the link breaks, because it will
Delay-tolerant networking and the DTN Bundle Protocol are built on the assumption that connectivity is intermittent, which in cislunar space is the normal case rather than the fault case: occultation, range, and contention all guarantee it. We intend to build store-and-forward behavior in from the first node instead of adding it after the first outage.
Publish the interface before asking for a signature
Interface control documents, conformance statements, and test methods are intended for publication so an integrator can design against the network before committing to it. Each document is released as it clears export-control review; integrators who need one ahead of publication can request it under NDA.
Bring measurements to the open questions
Where a standard is still being written, we intend to contribute data: characterization results, test methods, and the uncertainty behind them. Evidence is the only currency in that conversation that does not depreciate.
Downlink the decision, not the frame.
Ground-in-the-loop operations rest on two assumptions that stop holding past geostationary orbit: that the link is there when you need it, and that the round trip is short compared to the event. Across the cislunar volume neither is reliable. A close approach that develops faster than a contact window cannot be adjudicated from a console on Earth, and an observation that must be downlinked before it can be evaluated is an observation that arrives after it mattered.
The NEXUS edge stack is the answer we are building to that: put the processing where the collection happens, hold custody onboard, and spend the link on conclusions instead of pixels. It is in development. It is not operating, and none of it has flown.
Onboard detection and tracking
Sensor frames reduced to candidate detections and tracks on the spacecraft, so the link carries tracks rather than raw imagery. The link budget stops being the limit on how much of the volume can be watched.
Custody through maneuver
Correlation logic designed to keep an object identified when it changes its orbit. Losing an object and re-acquiring it later is not the same as never losing it, and only one of those is custody.
Autonomy under disruption
Edge behavior designed for intermittent connectivity rather than degraded by it: queue the work, act inside delegated authority, and reconcile with the ground when the next contact opens.
Assurance and the operator
Every automated conclusion is designed to arrive with the evidence that produced it and a version identifier for the model that produced it. An analyst who cannot audit an answer cannot defend it, and an answer nobody will defend does not get used.
An outpost is resupplied. A settlement makes itself.
This is the furthest out work on the page and the least developed. It is a concept track: a set of questions we are studying, not a program. No hardware exists, no article has been built, and nothing described here carries a schedule.
The reasoning is arithmetic. Every kilogram of oxygen, water, structure, and spare mass that leaves Earth is paid for twice, once to buy and once to lift, and an outpost pays that bill on every flight for as long as it operates. A settlement is the state where the loop closes locally and the mass is made where it is used. Between those two states sits a long list of unglamorous problems, and the material ones are the ones we find most interesting.
What we are studying: regolith treated as feedstock rather than as a nuisance, oxygen recovered as part of a materials process rather than from a dedicated plant, energy and thermal storage that carries a surface system through a lunar night of about 354 hours, and the measurement problem underneath all of it. That last one is the thread back to the top of this page. A part made on the surface is worth nothing until somebody can characterize it, and the discipline that characterizes an optic in vacuum is the same discipline that will eventually characterize a locally made structural part.
Bring us the measurement problem.
Component and assembly work quotes through FORGE. Research collaboration, interface questions, and technical briefings go through the same door, and the specifics are discussed under NDA rather than advertised.