Dek: A close approach becomes a safety decision through orbit estimates, covariance, new tracking and coordination. A changing probability is often the system learning—not the system contradicting itself.
ID: NU54-038 Section: Hot Slug: satellite-conjunction-risk-tracking Prepared: 2026-10-31 Evidence checked: 2026-10-31 Status: Editorial draft; not published
1. The warning is the beginning of the work
NASA's collision-avoidance handbook received an August 2026 revision, with its best-practice material and technical appendices separated into two volumes. The document describes a process rather than a single alarm: identify a possible close approach, assess its risk, then decide whether and how to reduce that risk. The revised first volume is publicly available.
That sequence is easily lost when a satellite encounter becomes a headline. A predicted separation appears, a probability is attached, and readers imagine two objects already committed to passing within a precisely known distance. Later the probability falls, and the original warning can look exaggerated. Alternatively it rises, making the first assessment look complacent.
Both interpretations can be wrong. Orbit predictions carry uncertainty. New observations alter the estimated paths and the uncertainty around them. Operators may also change a satellite's trajectory. A later assessment is not necessarily a correction of a careless earlier statement; it may be an assessment of a better-observed or deliberately altered situation.
The difficult task is to act before uncertainty has completely resolved, while avoiding unnecessary actions that consume resources or create other risks. Waiting has value because new tracking can clarify an encounter. Waiting also has a cost because manoeuvring opportunities disappear as the encounter approaches.
This article follows that tension through a fictional conjunction involving a spacecraft called Meridian-7. Its times, probabilities and operational choices are illustrative, not a reconstruction of a real incident or instructions for flying a satellite. The purpose is to explain what a warning contains, what it leaves uncertain, and why the quality of the decision depends on more than the most alarming number in a message.
2. A conjunction is not a collision prediction
Meridian-7's fictional operations team first receives notice that another tracked object may pass through a screening region around its predicted path. That region is deliberately much larger than the spacecraft. Its job is to identify encounters worth examining, not to define the physical boundary of a collision.
NASA's description of conjunction-event prediction distinguishes these proximity notifications from completed collision-risk assessments. A screening process compares predicted trajectories and identifies close approaches; the resulting message contains information for subsequent analysis. The agency explains that first stage separately from risk assessment.
The distinction resembles a search filter. A broad filter may return many candidates that turn out to be harmless after closer examination. Making the filter too narrow could omit encounters that deserve attention. Counting every candidate as an imminent collision confuses the deliberately cautious first stage with the conclusion of the entire process.
For Meridian-7, the first practical questions are therefore not “How soon will it crash?” but “Which objects are involved, when is closest approach expected, and what information supports the prediction?” The team needs to establish that the message refers to the right spacecraft and a genuine event, not an outdated trajectory or a duplicated notification.
Even the term primary object is operational rather than moral. It usually identifies the spacecraft being protected in a particular assessment. It does not establish that the other object is at fault or must manoeuvre. A debris fragment cannot respond to a request, and two active spacecraft may each appear as primary in their own operators' systems. The geometry needs analysis before the vocabulary becomes a story about blame.
3. Position is a state estimate, not a perfect coordinate
A satellite's orbit is inferred from observations and a model of its motion. The estimate includes position and velocity at a specified time. Those quantities together form a state. Predicting a future encounter requires propagating the state forward, accounting for forces and any known manoeuvres.
For a fictional example, imagine knowing Meridian-7's present position very accurately but its velocity slightly less accurately. Over a short interval, the velocity uncertainty may matter little. Over a longer interval, it can produce a much larger uncertainty in where the satellite will be. A small difference in speed accumulates into a difference in arrival time and position.
The observations themselves do not all provide identical information. A sensor's geometry, timing, measurement type and error characteristics affect how well different components of the orbit can be determined. More observations can help, but a hundred nearly redundant measurements are not automatically better than a smaller set that constrains a previously uncertain direction.
This is why a displayed coordinate with many decimal places should not be mistaken for exact knowledge. Numerical precision describes how a value is represented. Accuracy describes how closely the estimate corresponds to reality. A prediction can be printed to the centimetre while its uncertainty is far larger.
The state also has an epoch: the time to which it refers. Comparing coordinates associated with different epochs without proper propagation is meaningless for collision assessment. Two objects can pass through the same location safely at different times. The dangerous condition is overlap in space at the same time, making timing information part of the geometry rather than an administrative detail attached afterward.
4. Covariance describes the shape of uncertainty
An orbit estimate's uncertainty is not usually a neat sphere surrounding the satellite. It can be stretched along the direction of travel and narrower in other directions. Errors in position and velocity can also be related. Covariance is the mathematical structure used to describe these variances and relationships within the adopted uncertainty model.
A simplified picture helps. Imagine a cloud of plausible future positions for Meridian-7. If the cloud is elongated along its orbit, the main uncertainty may be how early or late the satellite reaches a location. If it is elongated across the orbit, uncertainty in another component matters more. Two clouds with the same overall volume can have different collision implications because their orientations differ.
The cloud is a representation of knowledge, not a swarm of physical copies of the spacecraft. There is one actual trajectory. The distribution expresses uncertainty about that trajectory under the assumptions of the estimation process. Treating its outer contour as a solid exclusion zone would misinterpret the probability model.
Covariance also does not guarantee that the assumed distribution is correct. An omitted manoeuvre, a poorly modelled force or a systematic sensor error can make a mathematically tidy covariance unrealistic. The question is whether the stated uncertainty matches observed prediction errors over relevant situations.
For Meridian-7's team, a covariance matrix is therefore not merely a field that must be present before software will run. It is an assertion about how much confidence to place in the state estimate. A collision probability calculated from unrealistic uncertainty can be misleading even when every numerical integration step is executed perfectly.
5. Miss distance alone cannot rank the danger
Suppose one hypothetical encounter has a predicted separation of 200 metres and another has a predicted separation of 500 metres. It is tempting to call the first more dangerous. But the comparison is incomplete without object sizes and the uncertainty in the relative trajectories.
If the first separation is known with very high confidence and the objects are small, the probability of physical overlap may be extremely low. If the second estimate has uncertainty that places substantial probability near overlap, it may deserve more attention. The nominal separation is a useful geometric summary, not a self-contained risk measure.
NASA's risk-assessment description explains collision probability as the likelihood that actual separation becomes smaller than the objects' combined size, using predicted trajectories and their uncertainties. It also distinguishes likelihood from consequence. Those are separate parts of the risk assessment.
For Meridian-7, the team therefore examines the relative state rather than comparing a distance with an arbitrary comfort threshold. The direction of the predicted miss matters, as does the orientation of uncertainty around it. A miss along the best-known direction can mean something different from an equal miss along the least-known direction.
This is also why a dramatic visual animation can be deceptive. To show the spacecraft clearly, graphics often enlarge them relative to their separation. The resulting image may look like a near impact even when it is only a schematic. A responsible illustration should identify that scale distortion and avoid letting a cinematic perspective substitute for the actual geometry and uncertainty.
6. The spacecraft has a size, and the model must represent it
Collision assessment cannot treat satellites as dimensionless dots. Solar arrays, antennas and other structures occupy space. A simplified calculation often represents the combined size of the two objects through a hard-body radius. NASA's FAQ explains this quantity as the sum of the radii used for the primary and secondary objects, alongside miss distance and covariance in the probability calculation. The agency's explanation identifies these inputs.
A radius is a modelling choice, not necessarily a claim that the spacecraft is spherical. Real vehicles can be long, flat or irregular, and their projected area depends on orientation. A tumbling debris object can make that orientation uncertain. Conservative geometric approximations may be appropriate, but their effect should be understood rather than hidden.
Imagine Meridian-7 has a compact central body and extended arrays. A model using only the central body would omit structures that could be struck. A model enclosing every possible orientation in a large sphere might overstate the effective collision cross-section in some geometries. The suitable representation depends on available information and the calculation's assumptions.
Object-size uncertainty can be particularly relevant for debris. Tracking an object well enough to estimate its orbit does not automatically reveal its exact dimensions, mass or attitude. These are different observational questions. An uncertainty in size should not be mistaken for uncertainty in position, even though both influence the assessment.
The practical lesson for a reader is that a collision probability has physical as well as statistical inputs. It is not obtained by feeding two coordinates into a universal formula. Geometry, uncertainty, encounter dynamics and the assumptions used to represent the objects all contribute to the result.
7. Probability can rise when knowledge improves
At the first fictional assessment, Meridian-7's encounter probability is modest. A later observation narrows the uncertainty around a path that passes closer to the other object. The estimated probability rises. Nothing physical has necessarily become more dangerous between those messages; the analysis has become more confident that the dangerous part of the original range is relevant.
The reverse can happen. Additional tracking may establish that the objects are more clearly separated than the early estimate suggested. Probability then falls without any manoeuvre. Both patterns are compatible with a well-functioning estimation system.
An elementary mathematical example shows why more uncertainty is not always associated with a larger probability for a small target region. Consider a purely illustrative one-dimensional normal distribution centred ten arbitrary units from a target interval running from minus one to plus one. With extremely small uncertainty, almost no probability lies in that distant interval. With intermediate uncertainty, more of the distribution reaches it. With extremely broad uncertainty, probability spreads across such a large range that the fraction in the small interval falls again.
This is not an orbital calculation. It demonstrates a property of probability distributions that can make a small reported probability ambiguous when uncertainty is very large. A low value can reflect a well-resolved safe separation, or a diffuse estimate that places little probability in any small region.
Meridian-7's analysts must distinguish those situations. They cannot infer good knowledge from a reassuring number alone. The quality and evolution of the state estimates matter, as do diagnostic checks appropriate to the encounter. The number is an output of an uncertainty model, not a certificate that uncertainty has disappeared.
8. A message needs provenance as well as numbers
A conjunction data message, or CDM, packages information about an encounter so that different systems and operators can exchange it. Its value depends on more than a probability and a closest-approach time. Object identity, state information, covariance, reference frames and details of the underlying orbit determination help a recipient interpret the result.
The US Office of Space Commerce's TraCSS CDM specification describes the message's role and field structure, building on the international CCSDS format. It includes information intended to support assessment and, where needed, avoidance planning. Version 2.1 is available as an official technical document. That version is cited here as a concrete example, not asserted to be the latest implementation requirement for every service.
For fictional Meridian-7, provenance answers practical questions. Was the spacecraft trajectory supplied by its operator or estimated from external tracking? Does it include a planned burn? When were the last observations used? Is the covariance calculated from the same orbit solution as the state, or supplied through another process?
A message can be syntactically valid yet unsuitable for a particular decision. Units can be misunderstood; an identifier can refer to the wrong object; a timestamp can be interpreted in the wrong convention. Robust systems need validation at both the formatting and meaning levels.
The message's creation time also matters. A screenshot detached from its timestamp can circulate after a newer assessment has superseded it. The public then sees a precise-looking probability that no longer represents the operator's current information. In a rapidly updated process, provenance is part of the factual claim.
9. Two messages may describe one event
Meridian-7 receives several fictional messages over two days. They do not represent several different impending collisions. They may be successive assessments of the same encounter as observations and predictions change. Counting messages without grouping them into events can inflate the apparent number of threats.
A single event can also have more than one assessment based on different trajectory sources. SpaceX's space-safety documentation describes a design in which CDMs can draw on operator-supplied trajectories and observation-derived trajectories, with metadata identifying the source; its page still uses prospective wording about a spring 2026 change. It also describes how a message can become inactive when superseded or when a newer screening no longer finds the approach within the screening volume. The operator's own documentation explains these distinctions.
That is a useful example of why a dashboard's total needs a definition. Is it counting messages, unique object pairs, distinct closest-approach events or events above a selected threshold? Each metric can be legitimate. They cannot be substituted for one another.
For Meridian-7's team, keeping an event history is more informative than collecting disconnected alerts. The history shows whether the nominal miss is stabilizing, whether uncertainty is shrinking and whether different information sources converge. It also records when a planned manoeuvre entered the trajectory model.
A disappearing alert does not necessarily mean the physical objects disappeared or that the original message was fabricated. It may mean the latest predicted geometry no longer meets the service's screening criterion. Understanding that state change requires the event record. Without it, a public account can turn routine updating into an invented mystery about warnings being withdrawn.
10. Better tracking is an action, not just more waiting
Waiting for another message can sound passive. In practice, reducing uncertainty may involve obtaining additional observations, improving an orbit solution or resolving conflicting information. The important distinction is between waiting with a credible prospect of useful new evidence and simply allowing the decision window to close.
Imagine Meridian-7's own orbit is well constrained but the secondary object has a sparse recent tracking history. Improving the primary's already precise position may add little to the relative uncertainty. Observing the secondary could be much more valuable. The most useful measurement targets the uncertainty that dominates the decision.
Observation opportunities are constrained. A sensor cannot see every object at every moment; geometry, weather for optical systems, equipment availability and competing tasks can matter. A plan based on “one more observation” must consider whether that observation is realistically obtainable before the manoeuvre deadline.
The value of information also depends on the available actions. If a spacecraft can only perform a suitable burn during one upcoming contact window, data arriving afterward may clarify the encounter but no longer enable the same response. An accurate late answer is not operationally equivalent to an accurate timely answer.
For readers, this explains why two missions can handle apparently similar probabilities differently without one being irrational. Their observation prospects, communication schedules and manoeuvring capabilities can differ. A safety decision combines the estimated physical risk with the timing and feasibility of mitigation. The probability is central, but it does not contain the spacecraft's entire operational situation within its decimal places.
11. Forecasting forces becomes part of forecasting separation
A predicted orbit is not obtained by extending a straight line indefinitely. Gravity dominates orbital motion, but other forces and effects influence the path. In low Earth orbit, atmospheric drag can be particularly important to prediction. The atmosphere at orbital altitudes varies, and a spacecraft's effective area and orientation influence its response.
For a hypothetical comparison, two objects initially have well-determined states. One is compact and well characterized; the other is irregular and tumbling. Their future trajectories may accumulate uncertainty differently even if their current positions are known equally well. The uncertainty model must represent the relevant prediction errors rather than merely repeat the accuracy of the last observation.
Space weather can complicate drag prediction by changing upper-atmospheric conditions. That does not mean every solar event produces an immediate collision emergency. It means the relationship between present tracking and future position can become harder to predict in particular circumstances, which affects how confidence should evolve with forecast horizon.
Known manoeuvres create a different issue. If Meridian-7's operator plans a burn but an external trajectory assumes no burn, the two predictions describe different futures. Neither should be compared as though it were simply a more or less accurate estimate of the same plan. The manoeuvre assumption must be reconciled first.
This is one reason operational data sharing matters. The operator may possess information about planned actions that no amount of external observation can reveal before those actions occur. Sharing a predicted trajectory with its assumptions allows other assessments to evaluate the intended future rather than extrapolating a past that the operator already plans to change.
12. Early action can save fuel but waste opportunities
A small change made sufficiently early can accumulate into a useful separation by the encounter time. Later action may require a different or larger manoeuvre to achieve comparable protection. But acting early can mean spending resources on an encounter that subsequent tracking would have resolved as low risk.
NASA's mitigation explanation identifies this tension between earlier, potentially smaller manoeuvres and waiting for uncertainty to decrease. It describes evaluating combinations of manoeuvre timing and size, then screening the proposed trajectory for other high-risk encounters. The agency presents mitigation as a trade space rather than a single reflex.
In Meridian-7's fictional case, an early burn could interrupt a planned observation campaign. A later burn could preserve that campaign if the risk falls, but leave less margin if the risk remains. The team is not choosing between safety and an obviously frivolous inconvenience. It is choosing among uncertain futures with different resource and mission consequences.
Fuel is one cost, but not the only one. A manoeuvre may require attitude changes, communication opportunities, instrument shutdowns or later orbit restoration. Some spacecraft have limited propulsion or constraints on when they can fire. Others may use forms of trajectory modification that operate more slowly.
No universal countdown captures all those differences. A public explanation should therefore avoid turning one agency's example timeline into a rule for every spacecraft. The relevant question is whether the operator had a documented decision process suited to its vehicle, including a realistic latest commitment point and a plan for what to do if the expected new information did not arrive.
13. A manoeuvre changes more than the original encounter
Suppose Meridian-7 plans a burn that increases separation from the original secondary object. That is necessary evidence, but not enough to establish that the new trajectory is acceptable. The changed orbit can create different encounters with other objects. Avoiding one conjunction by introducing another would not be a complete risk reduction.
The candidate trajectory therefore needs to be assessed against the broader catalogue or relevant set of trajectories. This is a systems problem rather than a two-body escape puzzle. The spacecraft occupies a shared orbital environment before, during and after the initial encounter.
Execution uncertainty also matters. A planned burn has a duration, direction and magnitude, but the actual result may differ within known limits. A robust plan considers those differences rather than assuming the intended trajectory will be followed perfectly. The uncertainty after a manoeuvre can temporarily differ from the uncertainty before it.
For the fictional team, the planning record should distinguish the commanded action, the expected result and the range of credible execution outcomes. That distinction helps assess whether the manoeuvre meaningfully reduces risk even when it does not perform exactly as intended. It also guides what observations are needed afterward.
The same reasoning applies to a return to the original operational orbit. Restoring a desired ground track or altitude is another trajectory change, not an automatic reset button exempt from assessment. A complete account of an avoidance action includes its operational aftermath. The dramatic burn is only one moment in a sequence that begins with prediction and ends with confirmation of the spacecraft's new state.
14. Two active operators must avoid incompatible plans
If the secondary object is debris, Meridian-7's team cannot negotiate with it. If it is another operational spacecraft, coordination becomes possible—and necessary. Each operator may be considering a manoeuvre based on assumptions about what the other will do.
A simple fictional example shows the risk. Operator A evaluates a plan assuming B will maintain its trajectory. Operator B evaluates a different plan assuming A will maintain its trajectory. Each plan may appear acceptable alone. The combination is a third scenario that neither assessment has necessarily evaluated.
This does not mean two manoeuvres always increase danger. They may improve separation. The point is that simultaneous uncoordinated action changes the assumptions. Sharing intentions lets both teams assess the same prospective situation rather than two incompatible ones.
Coordination requires more than a contact address. Messages need to reach the responsible team in time, identify the event unambiguously and state whether a manoeuvre is tentative, committed, executed or cancelled. A preliminary option should not be mistaken for a confirmed action. A cancelled manoeuvre should not remain embedded in another operator's prediction.
Language and time conventions matter here as much as orbital mechanics. Clear timestamps, identifiers and acknowledgement procedures reduce avoidable ambiguity. Automation can help deliver and validate information, but it cannot compensate for a message whose operational meaning is undefined.
For public discussion, this is a more useful frame than assuming there must be a simple right-of-way rule analogous to road traffic. The available actions depend on vehicle capability, geometry and timing. Whatever coordination arrangements apply, the safety objective is compatible, well-understood trajectories—not a contest over which operator can insist that the other move first.
15. A threshold is a policy choice informed by engineering
A numerical threshold can help determine when an encounter receives attention or requires mitigation under a particular programme's rules. It makes a response more consistent, but it does not convert a probability model into a law of nature. Different organisations may use different thresholds and supplementary criteria.
Consider two fictional missions. One carries a unique scientific instrument with limited manoeuvring fuel. Another operates many replaceable spacecraft with frequent automated trajectory adjustments. Their acceptable operational tradeoffs may differ, while both still have responsibilities to other users of the orbital environment. A threshold should be interpreted within that broader decision framework.
The probability itself must also be comparable before thresholds can be compared. Different object-size assumptions, uncertainty treatments or encounter methods can produce different values. Claiming that one operator is safer merely because its published threshold contains more zeros may ignore those underlying differences.
A threshold crossing is not necessarily an instantaneous command to fire. A process may include data-quality checks, review of available actions and a commitment decision at a defined time. Conversely, a value below a usual threshold may still warrant attention if the inputs are unreliable or the circumstances unusual.
For Meridian-7, the useful question is whether the team's decision rule was established in advance and applied to credible inputs, with an escalation path for cases the ordinary rule does not handle. Retrospectively selecting a threshold to justify whatever happened would be weak evidence of good practice. A defensible process explains both why an action was taken and why apparently similar cases might legitimately lead to a different action.
16. Collision consequence is not captured by probability alone
Two encounters can have the same estimated probability and very different possible consequences. A collision can end a mission, damage a spacecraft or create debris that threatens other objects. The masses, relative velocity and physical configuration influence what happens after contact.
A basic physics illustration makes one part clear. Kinetic energy scales with mass and the square of speed in the relevant frame. Doubling an illustrative impact speed, while holding mass constant, multiplies that kinetic-energy term by four. That arithmetic is not a debris-generation model, but it explains why relative velocity matters beyond the simple question of whether two shapes overlap.
The broader environmental consequence can extend beyond the two operators. Fragments may remain in orbit and generate future hazards. Thus an operator's private mission-loss calculation does not encompass every cost imposed by a collision. A spacecraft that is inexpensive to replace can still contribute to an expensive shared problem.
For Meridian-7's fictional assessment, the team should therefore distinguish likelihood of contact from the potential severity of contact. Those are not competing metrics. They are different dimensions of risk, and each has uncertainty. An estimated debris count should not be reported as a guaranteed outcome any more than a collision probability should be reported as a guaranteed collision.
This distinction also discourages a misleading form of reassurance: “The probability is tiny, so the event is unimportant.” Small probabilities can deserve attention when consequences are substantial, especially across repeated operations. The appropriate response still depends on feasible mitigation, but consequence belongs in the discussion rather than being silently replaced by the probability alone.
17. Many small risks do not become a simple sum of headlines
A constellation encounters many potential conjunctions over time. Understanding cumulative risk requires more than quoting the largest single probability or counting every alert. Events can be related, messages can be duplicates, and manoeuvres can change later trajectories.
For a purely mathematical illustration, if ten independent events each had a one-percent probability of an outcome, the probability of at least one outcome would be one minus 0.99 raised to the tenth power, about 9.6 percent. Simply adding ten percentages gives ten percent, a close approximation here but not an exact general rule. Independence is an explicit assumption in the example, not a claim about real satellite encounters.
Real operational histories introduce further complications. Several messages may describe the same event. Two encounters may share uncertain trajectory information. An avoidance manoeuvre can alter the probability of later encounters. Adding a series of updated probabilities for one conjunction would count the same prospective collision repeatedly.
A meaningful aggregate measure therefore needs a defined unit of analysis and a stated time interval. Is it estimating expected collisions, probability of at least one collision, number of mitigated events or operational workload? Those are different quantities with different uses.
For Meridian-7's operator, an annual review should preserve those distinctions. A large alert count might indicate a busy environment, a broad screening policy or better detection. A large manoeuvre count might reflect conservative mitigation or changing operations. Neither statistic alone establishes how many collisions would otherwise have occurred. The counterfactual—the trajectory history without intervention—requires analysis rather than a rhetorical conversion of every action into a saved spacecraft.
18. Automation needs an intelligible exception path
Frequent conjunction assessment creates a strong incentive to automate repetitive tasks. Software can ingest messages, group events, compare trajectories and evaluate candidate actions faster than a team manually inspecting every record. But speed does not remove the need for clearly defined authority and exception handling.
Imagine Meridian-7's fictional system receives a malformed covariance matrix, a missing object-size estimate or two conflicting planned trajectories. An automated process needs a specified response to each condition. Quietly substituting a default may keep the pipeline running while changing the meaning of its output. Halting without notifying anyone can be equally dangerous near a deadline.
The important design question is not whether a human touches every alert. It is whether the system's assumptions, permissions and failure modes are understood. Which actions may it take autonomously? When must it escalate? What happens if communication with the ground is unavailable? How are stale inputs detected?
An audit trail should connect the input message, software version, decision rule and resulting command or recommendation. Otherwise an operator may know that a manoeuvre occurred without being able to reconstruct why. Reproducibility is valuable after an anomaly and during routine validation, when the objective is to discover weaknesses before they affect a real encounter.
Automation also changes workload rather than simply eliminating it. Engineers must maintain data interfaces, test new versions, investigate unusual cases and monitor performance. The best public account would therefore describe the quality of the automated process, not merely celebrate the absence of human intervention. A fast decision is useful only if the system knows when its ordinary decision logic no longer fits the evidence.
19. Passing safely does not prove the warning was wrong
Meridian-7 and the secondary object pass without collision in the fictional case. That outcome is welcome, but it does not retrospectively establish that the estimated probability was zero. A low-probability event will usually not occur even when the probability is well calibrated.
If the team manoeuvred, the observed safe passage occurred on a changed trajectory. It cannot directly reveal what would have happened without the manoeuvre. Calling the action unnecessary solely because no collision followed confuses the intervention's outcome with its counterfactual.
Even without a manoeuvre, one safe encounter provides limited information about probability calibration. Evaluating a system requires many predictions and careful attention to which cases were altered by intervention. Rare outcomes, changing observation quality and selection effects make that evaluation difficult.
A useful fictional post-event review instead asks whether the prediction errors were consistent with the stated uncertainty, whether the decision used the best information available at the time, and whether communication and execution worked as intended. It can identify improvements without pretending that hindsight supplies the exact probability that should have been known earlier.
The same principle applies to a warning whose probability declined before closest approach. A well-designed system is supposed to identify potentially important events early and then refine them. Some will cease to require action. That is not evidence that the first-stage screening should be abolished.
For readers, the appropriate standard is prospective quality: was the assessment reasonable given the information and options then available? Outcome alone is too crude. A lucky result can follow a poor process, and a sound process cannot guarantee that every uncertain event ends safely.
20. The catalogue is not the whole orbital environment
Conjunction screening operates on objects and trajectories known well enough to include in the relevant system. It cannot issue a specific predicted encounter for every untracked fragment. Avoidance is therefore one part of spaceflight risk management, not a comprehensive shield against all debris.
The European Space Agency's September 14, 2026 overview of its annual Space Environment Report describes increasing complexity in the orbital environment and stresses monitoring, coordination and debris mitigation. It states that the report is based on data through the end of 2025. The publication and its data cutoff are explicit. Its 2026 title should not be mistaken for a complete census of events throughout 2026.
For Meridian-7, better avoidance of tracked encounters would not eliminate the need to consider small-debris exposure, spacecraft reliability and end-of-life disposal. Nor would more precise tracking remove existing derelict objects. Measurement and prevention address related but distinct parts of the problem.
This matters when a successful avoidance system is offered as proof that orbital growth has no environmental limit. A system can become better at managing known encounters while the underlying population of hazardous objects continues to grow. Operational competence and long-term sustainability are not synonyms.
Conversely, the existence of untracked debris does not make tracking pointless. Reducing one class of risk remains useful even when another persists. The correct comparison is not between perfect safety and uselessness. It is between a documented reduction in a defined hazard and the residual risks that require other measures, including preventing new debris and limiting how long failed spacecraft remain in crowded regions.
21. What a public collision-risk report should preserve
A useful public report on a specific encounter would identify the assessment time, predicted closest-approach time, information source and whether the trajectory includes a planned manoeuvre. It would distinguish a proximity notification from a completed risk assessment and avoid presenting a nominal miss distance as an exact future measurement.
If a probability is reported, its context should travel with it. Was it an early estimate based on sparse tracking or a later estimate supported by improved observations? Did it change because the orbit estimate changed, because uncertainty changed, or because an operator acted? Sometimes the available information cannot separate those explanations, and the report should say so.
A sequence of dated assessments is often more revealing than a single peak value. It shows the development of knowledge and the timing of decisions. Selecting only the highest probability can make a routine refinement process look like an unmitigated emergency. Selecting only the final low value can erase the uncertainty operators actually faced.
Counts need equally clear labels. Messages are not events; events are not manoeuvres; manoeuvres are not confirmed collisions averted. A report that preserves those distinctions can still communicate a growing workload or a concerning environment without inflating its evidence.
Finally, the account should distinguish operational claims made by an operator from independently established outcomes. An operator is the primary source for its procedures and reported actions, but its self-description is not automatically an external audit. The most credible reporting makes the evidentiary chain visible: who estimated, who acted, what changed and what remains unknown after the objects passed.
22. The changing number is part of the safety system
The fictional Meridian-7 encounter has no dramatic final certainty to offer. Its lesson is that a collision assessment is a time-dependent inference connected to a time-dependent set of choices. The state estimate changes. Its uncertainty changes. Available manoeuvres change. Other operators may change their plans too.
A probability stripped from that sequence is easy to misunderstand. It can look like a prediction that must either come true or be exposed as false. In reality it expresses a conditional assessment under particular information and modelling assumptions. Its usefulness depends on those assumptions being credible and on the decision process responding appropriately as they evolve.
The technical work is demanding because several kinds of precision must align. Time and reference frames must be consistent. Covariance must represent uncertainty realistically. Physical dimensions must be modelled sensibly. Candidate trajectories must be checked beyond the original encounter. Coordination must convert tentative plans into shared, current information.
None of this makes the probability meaningless. It makes the number interpretable. A changing assessment can be evidence that the system is incorporating new observations, resolving uncertainty or evaluating a safer trajectory. The important question is why it changed and whether the change arrived in time to support a better decision.
The night sky's traffic problem will not be solved by a dashboard of alarming decimals alone. It requires measurements, disciplined inference, reliable spacecraft and operators able to coordinate before their options close. A warning begins that work. A responsible account follows the warning all the way through the decision, rather than mistaking the first number—or the last—for the entire story.