Decision intelligence for
space operations

CelestCore helps satellite operators turn collision risk data into defensible maneuver decisions, including knowing when the data itself should not be trusted.

Scroll

More satellites. More alerts.
Same decision bottleneck.

Satellite operators receive hundreds of collision alerts every week. Each one must be evaluated because the consequences of getting it wrong are permanent. An unnecessary maneuver wastes fuel and shortens mission life. A missed threat can destroy an asset.

Existing systems deliver tracking data and risk metrics, but they stop at the operator's screen. Tracking, risk analysis and maneuver planning address different parts of the conjunction problem. CelestCore connects them at the decision layer, where evidence quality, timing, fleet constraints and independent verification determine whether the operator should act, monitor or defer.

The bottleneck is no longer collecting orbital data.
The bottleneck is making the decision.

Operators often monitor alerts while tracking improves because many resolve naturally. The decisions that matter are the ones where waiting no longer helps. CelestCore determines when continued monitoring remains useful, when the maneuver window requires action and when the available evidence is too weak to support a recommendation. The May 2024 Gannon solar storm showed what happens when the data stops reflecting reality. Changes in the upper atmosphere altered satellite paths faster than forecasts could keep up, while thousands of satellites maneuvered unexpectedly. Reliable collision warnings became nearly impossible for three days.

~16,000
Functioning satellites in Earth orbit
~600,000
Conjunction Data Messages generated daily by 19 SDS
~56
Spacecraft supported by NASA CARA
53/53
NASA CARA reference cases reproduced
Sources: ESA Space Debris Office, updated July 31, 2026; Lewis and Skelton, 2024, reporting approximately 600,000 CDMs per day in 2023; NASA Scientific Visualization Studio, February 2024. CDMs are messages and multiple updates may describe the same conjunction event. The 53/53 result is from CelestCore's internal benchmark.

From risk data to defensible decisions

NASA missions supported by CARA are backed by dedicated analysts who review close approaches, compare multiple methods and advise on decisions. CelestCore brings that analytical rigor into software for commercial operators that do not have a dedicated analyst team behind every conjunction.

The system ingests conjunction data from existing providers, assesses data quality, optimizes maneuver options across fleet constraints and delivers confidence-scored recommendations with full traceability.

Assess data quality

Analyzes how uncertainty signals evolve across successive data updates. Detects when risk data is physically inconsistent. Tracking gaps, estimation resets, atmospheric model failures. All flagged before any decision is made.

Optimize fleet decisions

Evaluates maneuver options across an entire constellation simultaneously. Fuel budgets, mission schedules, overlapping alerts. All weighted by how much the system trusts each conjunction's data.

Verify independently

Every recommendation passes through a separate safety verification path that cannot be bypassed or influenced by the advisory system. When confidence is insufficient, the system formally declines to recommend rather than producing an answer the data cannot support.

What the operator could see

A view into the decision workflow an analyst could use day to day. The fleet view below shows a hypothetical workday for an operator running 80 satellites.

CelestCore does not execute maneuvers. Every recommendation is advisory and every override is logged. The operator is always the one in command.

C
CelestCore / Fleet
80 active satellites, 40 events in last 24h
Fleet posture
CVaR 95, next 72h
Nominal
No fleet-level risk concentration. Three events need judgment.
Needs judgment
3
Withheld
0
Auto-resolved
37
Fuel reserved (72h)
0.1% of budget
Needs human judgment
3 events
Primary vs secondary
Posture
Confidence
Reason
TCA
Sat-04 vs SL-16 R/B
Act
High
Pc 2.8e-4. Maneuver at 0.6 m/s clears 55% of risk.
22h
Sat-17 vs Cosmos 2251 Deb
Defer
Low
Covariance contracted 96%, faster than OD allows.
2d 14h
Sat-42 vs Starlink-2891
Monitor
Medium
Operator coordination pending, no action needed yet.
3d 6h

The analyst works through three events while the system resolves the other thirty-seven through the configured workflow. The same pattern holds from 80 satellites to several thousand.

Tested against NASA's benchmark

53/53 cases matched to at least 14 significant digits

Validated against NASA CARA's published benchmark. Real conjunction events involving Hubble, TERRA, AQUA, ICESat-2 and other high-value assets.

Hubble (HST) TERRA AQUA ICESat-2 WorldView CALIPSO NOAA-19 GPM
A collision risk that looked like zero was not safe to trust
Standard calculation
~0
Risk appears negligible
CelestCore
DEFER
Evidence unreliable
NASA reference
ATTENTION
Risk warrants attention
Case Study Details

Across three updates, the standard 2D calculation returned collision probabilities from 3.86 × 10-168 to 4.51 × 10-81. CelestCore recorded physically unrealistic covariance behavior, insufficient evidence of consistent covariance growth, low segment confidence and an unstable covariance structure. It assigned the combined event a DEFER posture rather than issuing an unsupported recommendation.

CARA's higher-fidelity SDMC reference placed the probability between 1.30 × 10-6 and 1.17 × 10-5. The entire range exceeded NASA's 1 × 10-7 level for operational attention. It remained below the default 7 × 10-5 threshold for evaluating maneuver options and the 1 × 10-4 mitigation threshold.

Sources: CelestCore reproducible audit artifact; NASA CARA Close Approach Risk Assessment; NASA Differential-Drag Decision Aid.

37/53
NASA cases where deeper analysis found a higher risk estimate
13,154
Real satellite close-approach histories analyzed
Dual-path
Every recommendation independently verified
End-to-end
From collision alert to operator-ready decision
Benchmark Details

In 37 of NASA CARA's 53 reference cases, finite covariance-scale profiling found a higher interior candidate collision-probability estimate than the endpoint value. This candidate is not a certified continuous bound. The operational-data evaluation covered 162,634 CDM-derived rows grouped into 13,154 ESA conjunction-event histories.

Founder-led. Mission-driven.

Hajymyrat (Ata) Geldimuradov
Founder, CelestCore

My background spans machine learning systems, orbital data analysis and decision system architecture.

CelestCore was born from a pattern that I often noticed. Many maneuver decisions were being made on data that looked clean but was not. The tools operators had were not built to tell them the difference. What they wanted was not more alerts or prettier dashboards. They wanted to know when to trust the number in front of them; that question was not being answered anywhere in the stack. As such, I built CelestCore as an answer and at a cost and commitment level that fits operators of diverse workflows, missions and fleet sizes.

Focus
LEO constellation operators
Location
United States