Space operations run on models that predict density, dose, charging. Good ones. But a model can’t tell you what it doesn’t know. That has to be measured from the outside. From LEO to the Mars surface, this atlas draws the frontier by calculating certainty where it can be earned from measurement, and recording gaps where it can’t. With five decades of model heritage, LEO is the most mature environment, but even there, storms locally drive the standard model's error to ten times its everyday miss. We flag these errors about three days ahead, at 92% precision, where drag residual is available from live satellites. The gaps are even bigger in GEO, cislunar, and the Mars corridor.
Latitude × altitude × local time. Every measured cell was tested on storms the model never saw. Solid cells are certainty earned from Swarm and GRACE-FO drag observations; dashed cells are waiting on a partner satellite’s drag signal.
Three environments, one standard. Measured means we made a prediction and scored the residual outcome. Modeled means there is a registered stake with a named measurement that would score it. Hover on any cell.
Here is how the atlas reads against what’s actually on the manifest. Which numbers you can schedule around, which are stakes awaiting a score, and which are gaps.
The big deployments of the next two years — Kuiper’s shells at 590–630 km, Starlink’s continuing build-out, and everything that rides with them — fly almost entirely in the 550–850 km rows of the grid above; The rows where nobody is measuring and the standard density models disagree with themselves by 4–13%. Storm-time drag error lands directly on conjunction screening and reentry prediction, and the declining phase of Solar Cycle 25 is historically rich in the recurrent high-speed streams that drive it.
What the atlas already offers that fleet: below 550 km — where the first commercial stations and most crewed LEO operations fly — storm response is certified cell by cell, and when the standard model is about to fail, we flag it in real time at 92% precision. The rest is closable. A fleet that shares the drag residual its GPS already computes lights up its own corridor. More participating satellites lights up more cells.
The declining phase of the solar cycle is when high-speed streams recur every solar rotation and GEO’s relativistic-electron episodes — the spacecraft-charging environment — are historically at their strongest. What was earned: faster solar wind pumps electron flux about two days later, in all four local-time sectors, strongest on the midnight side. That is a usable two-day heads-up for anomaly attribution and scheduling charging-sensitive operations.
What we couldn't earn (yet): the storm-time response is modeled, not certified — one longitude sees each sector only ~6 storm-hours a day — so storm charging margins should stay conservative until additional GEO longitude feeds close the cell.
As crews return to cislunar space — the Artemis flights, Gateway assembly, a steady lander cadence through 2027 — the atlas holds 17 years of measured lunar-orbit dose response to solar-particle floods, with the storm-vs-quiet structure intact across the whole archive. Two caveats matter operationally. Extreme floods are so rare that nobody can certify shelter margins for them from data. And, the magnetotail’s effect on dose runs opposite to textbook expectation; the Moon spends days of every month inside it, so for a long-stay base that is a real open question about the monthly dose rhythm, published in the gap register.
For the surface itself: the instrument exists (LND), the archive is blocked, and one data agreement converts base-siting dose assumptions from inference to measurement. It should happen before anyone pours regolith.
Every crewed Mars architecture spends six to nine months each way in the transfer corridor where four one-way snapshots in 25 years is the entire measurement record. Our current stake: ~0.28–0.29 mGy/day of free-space galactic-cosmic-ray dose (silicon), which compounds to roughly 50–80 mGy per leg before a single solar storm — if the stake is right. That “if” is the point. The late-2026 Mars departure window is the nearest chance to close the biggest hole on this map. Any rideshare hosting a small dosimeter, reporting hourly for one solar rotation, scores the stake. The analysis is pre-committed on our side and waiting for the signal.
At Mars, the measured results already carry weight for a base: ~20 g/cm² of atmosphere erases the electron-driven dose signal seen in orbit — a flight-data anchor for habitat shielding mass — and solar-particle events suppress surface dose rather than spike it, the kind of fact a storm-shelter strategy can be sited on. One outstanding mystery in the gap register: a seasonal surface-dose signal that has survived every explanation we’ve tested.
Every existing gap has a named target that is a flight opportunity, a data agreement, or a feed that already exists in someone’s flight-dynamics logs.
| ref | gap | what closes it | detail |
|---|---|---|---|
| TRANSIT | Earth→Mars corridor radiation | Flight opportunity | Crew transit dose has never been continuously measured. Any Mars-bound rideshare hosting a small dosimeter closes the biggest hole on this map. The analysis is pre-committed on our side. |
| LUNAR | Lunar surface dose | Data agreement | The measurements exist (LND instrument); access is blocked. One archive agreement converts this from gap to coverage, and it now carries a second justification. 17 years of CRaTER hold only 46 magnetotail-flood days, so a farside stream is the only way to close that cell. |
| LEO-FLEET | Upper-LEO density (550–850 km) | Live drag signals | Every dashed cell in the LEO grid is a satellite-shaped hole. The signal that fills it is the GPS drag residual every operating satellite already computes. We don't need the raw telemetry or proprietary state. Just that one signal lights up certainty where current models disagree and frequently miss at storm time. |
| LEO-STORM | Low-altitude storm seasons | Scheduled data | Severe-storm hours are rare; storm certainty grows each solar-cycle season. Next data window identified (declining phase). |
| GEO-STORM | GEO storm-time electrons | More longitudes | One satellite = 6 storm-hours/day/sector. Additional GEO longitude feeds certify the storm side. |
| SAA-POLAR | Polar inner-belt boundary | Different orbit | Never sampled from our current source orbit; a hosted sensor or alternate-orbit dataset closes it. |
| STORM-TYPES | Fast storm-train events | Advisory | A small class of storm sequences (notably 2023 storm trains) degrades our model; flagged for conservative margins until characterized. |
Nervous Machine learns the causal drivers of local environments — whether they live inside a machine like a satellite, a drone, a robot, or a quantum computer, or in an open environment like space. We believe the intersection where internal and external drivers influence each other is where some of the most unexpected edge cases arise. Assurance there has to be earned, not assumed.
This atlas is that conviction applied to the largest open environment there is. Certainty earned from measurement, models stamped and validated, and every gap named with the measurement that closes it.
Nervous Machine