A column about curious stories from the worlds of science and technology. Today — a news item a couple of weeks old, but with a 58-year rehearsal behind it. In mid-October, a Russian military satellite launched in the spring under the remarkably informative name Kosmos-2612 altered its orbit, matched a foreign orbital plane, and sidled right up to the commercial radar spacecraft ICEYE-X36 — Finnish by blood, American by registration, Ukrainian by clientele. At closest approach, estimates put just a couple of hundred meters, or perhaps mere tens of meters, between them. Then the visitor backed off without breaking anything — and that, frankly, is the most chilling part of the story.
There have been no official explanations from either Moscow or ICEYE itself. The Pentagon spoke later, and more briefly: by October 23, Washington assessed the approach as “consistent with an on-orbit weapons operation.” But the event was played out from beginning to end in open data: orbital-tracking companies, one observer with a blog, and thousands of enthusiasts reading the dry lines of two-line elements. And they spotted it not on the night of the close approach — but back in May, when the Russian fuel budget voted for maneuvers whose price tag spoke louder than any press release. This is the story of what war looks like when nobody is shooting yet: its mechanics of approach, the arguments over numbers, and the 1968 forebears of this stunt, with nearly two and a half hundred fragments to their name.
Let us begin with the victim pretending to be an extra. ICEYE-X36 is a compact radar spacecraft placed in 2024 into an orbit 556 km high, with an inclination of 97.8°. Its owner is the American subsidiary of Finnish company ICEYE, while ICEYE itself is a constellation of dozens of identical radar satellites that image Earth through clouds and at night, when optics are blind. Its clients are both civilian and military: the US Army, several European defense ministries, and, since 2022, the Ukrainian military. Radar provides weather-independent reconnaissance: for SAR, the Kerch Bridge in fog is an open book, if the radar is looking at it.
Kosmos-2612 is its opposite in every respect. It was launched from Plesetsk on April 16, 2026 (23:17 UTC), aboard a Soyuz-2.1b rocket with a Volga upper stage — incidentally, the combination was flying for the first time — alongside six spacecraft: Kosmos-2609…2614. Moscow wrote almost nothing about what they do in orbit, and that is standard genre: “spacecraft for military purposes.” Observers have long suspected that such a family carries a specialized package for signals intelligence and inspections — but suspicion is documented only by behavior. And the group’s behavior was such that even hardened telemetry readers were whistling by May.
Here we need to explain one rule of orbital mechanics on which the whole detective story rests. A satellite flies in a plane like a train on rails: to turn off its path, it must rotate its velocity vector through an angle, and in LEO that is insanely expensive — delta-V in the neighborhood of hundreds of meters per second for every half-degree to degree of inclination. A normal satellite works in its own plane from launch to death, because nobody has fuel to change rails except machines built for exactly that purpose.
In May 2026, five spacecraft from April’s sextet raised their inclination by roughly 0.8° — to 97.8°, precisely into ICEYE-X36’s plane. Five times. Each of those five adjustments cost around 106 m/s — as Greg Gielinger of Integrity ISR calculated back in May. All told, more than half a kilometer per second of delta-V for the sole right to fly in somebody else’s plane. A handsome tactical touch: wait for the planes’ natural drift, then pay only for inclination — four times cheaper than a direct plane change. Budgets do not lie: these machines were designed to maneuver aggressively — and to maneuver right next to someone.
Then the cluster did what maneuver-trained machines are expected to do: in late May, it brought its trajectories close to ICEYE-X36. On May 26, Kosmos-2609 passed within 2.4–2.5 km — but that was an intersection of different planes, a conjunction, not a rendezvous (CSIS, using COMSPOC data, calculates 2.5 km). On May 29, Kosmos-2614 passed within 20 km four times in two and a half hours — twice in lighting favorable for “photographing.” After that the planes drifted apart, everyone exhaled, and for three months it looked as though it had been an excursion.
Then, on September 30, the second act began — this time solo. Over two weeks, Kosmos-2612 ran a campaign scripted like a textbook: September 30 — an orbital raise of ~46.5 km with a ~25 m/s burn (Integrity ISR reconstruction; KeepTrack counted two burns totaling ~31 m/s and all 47 km); then the spacecraft traveled in a phasing orbit and twice — on October 5 and just after midnight on the 12th — passed directly over its target from ~55 kilometers higher; October 7 brought a pinpoint inclination adjustment of ~4 m/s; on October 12 at 02:17 UTC — a ~28 m/s lowering burn that brought the spacecraft to 555 km versus the target’s 556, into its plane, into trailing echelon nearly 475 km behind. The entire October preparation came to roughly 57 m/s — economical, disciplined, like a dancer stepping out for one figure. Commercial orbital-monitoring firms supporting the Joint Commercial Operations Cell raised a flag for the possible start of RPO — rendezvous and proximity operations. After that, events unfolded according to the observers’ timetable.
The average follower of orbital news saw the following feed in mid-October. From October 15, Kosmos-2612 moved from philosophy to practice: it closed to 2–3 km and began “hovering” at that range, shifting from a position ahead of the target to one behind it. LeoLabs would later describe this in the deliberately dull language of a report: “this was not a routine conjunction or a flyby” — a controlled approach, changes in relative position, lingering in the proximity zone, further small maneuvers, a controlled departure. Translated from insurance-speak into human: that is not how ballistics fly. That is how a pilot flies.
Around 18:15 UTC on October 17, Marco Langbroek — a Dutch astronomer and veteran amateur satellite tracker — published a post titled “a worrisome cat-and-mouse game,” with a calculation: the pair could close to within a few meters in the coming hours. Not long before, Langbroek himself had considered the matching planes a coincidence: after all, close approaches had been rare. The September–October maneuvers convinced him — synchronization of altitude, eccentricity, and period. That familiar formula: we do not see intent, but we see the delta-V.
The night of October 18 was the peak. According to fresh elements, the close approach occurred around 01:35 UTC — to within several hundred meters, or perhaps considerably closer. LeoLabs puts it more bluntly: “within 500 meters and likely much closer” (Adam Marsh, the company’s vice president for technical solutions). On the morning of October 18, the two spacecraft passed over South America, as they say, in a single frame: a timelapse by s2a systems shows two dots traveling in formation — a sight that makes an astronomer’s heart clench. ICEYE-X36, according to open elements, remained motionless — and this was not negligence but a matter of design: a compact machine with unknown remaining fuel against a visitor that had planned the meeting for half a year. There was simply nothing with which to get away.
On Sunday afternoon and evening, the stalker left: between 12:25 and 23:34 UTC, it raised its orbit by 2.3 kilometers (a ~1.3 m/s burn). A satellite two and a half kilometers above its target falls behind by roughly 350 kilometers per day — and the separation began: by Monday morning, according to elements from October 19, it was nearly 300 kilometers behind. Incident closed? Not quite. Late on October 19, Kosmos-2612 returned once again in altitude to the target’s orbit, slowing the diverging drift. In an update, Langbroek dryly noted that it appeared they were not done with ICEYE-X36 yet. Adam Marsh of LeoLabs drew a line under the tension in a comment to The New York Times: “Usually these stories don’t end with one.”
Two separate counters immediately mark the episode as historic. LeoLabs: the first known case of an unauthorized rendezvous in low orbit (in Marsh’s words: “We’ve seen this for the first time without consent”). Jonathan McDowell — the human database of every spacecraft launched since 1957 — offers the ultimate formulation: as far as is known, this is the closest known intentional and uninvited interaction between two active spacecraft in LEO. The qualification matters more than the formula itself: in GEO, the USSR and United States spent decades making each other’s lives miserable through inspections by geostationary machines, and Russian approaches to secret American satellites had been recorded before. But this was low orbit, a commercial spacecraft, days in the proximity zone, zero permissions.
One part of this story deserves separate examination — the numbers. Readers of news digests were presented with: “they came within 60 meters.” The deeper you go down the rabbit hole of retellings, the more precise the number looks — and the less foundation it has. In our own space digest on October 31, the number looked respectably scientific; pull at it, and I found the classic food chain of a fake, only backward — from a fragment to a “fact.”
What the primary sources actually contain. LeoLabs: “within 500 meters and likely significantly closer.” Scientific American’s science editor: “no more than a few hundred meters, with some estimates as low as 30.” Langbroek: with two-line elements, any exact meter count is self-deception; a spread by a factor of several is built into the method. KeepTrack reconstructed the pass from the catalog and found a corridor of 150–650 meters, stressing that its own element-set covariance for 2612 is around 1.9 km along-track; the public catalog cannot permit anything tighter. space.com wrote “at some point — perhaps just meters.” And the “25 meters” circulating online is not a measurement at all, but an extrapolation of element sets through maneuvers: one wandering STK repost placed a 25 m pass at 10:00 UTC on October 18 — a time when, according to post-maneuver elements, 2612 was already drifting kilometers behind; the automated Orbital Radar service quoted by Militarnyi “predicted” 26 meters for October 19 — when the close approach had long been over. So the honest summary goes like this: less than half a kilometer — certain; less than a hundred — likely; tens of meters — nobody knows. Each successive publication cut away uncertainty like inconvenient wrapping paper — and after two weeks of the cycle, a neat “60 meters” was born, source-less as urban legends are passport-less.
The moral of this paragraph extends beyond orbital mechanics. A figure with one decimal place in a field that is difficult to verify is not information but its makeup. We will most likely never know how many meters there were: radar measured, telescopes filmed, TLE readers calculated — but exact telemetry could be read only by whoever possesses it. The rest of us are offered an honest remainder: the order of magnitude is hundreds of meters, compression into tens cannot be ruled out, precision is a comforting phantom.
This audacity has a long pedigree, and it begins not in the 2020s but in the 1960s. Six years and one month after Sputnik, the Kremlin understood that American photo-reconnaissance satellites were passing over the country on an assembly line, while intercepting them from the ground was a dream. The project began with machines bearing the crystal-honest name Polyot: on November 1, 1963, Polyot-1 left Baikonur and immediately demonstrated maneuvers beyond the reach of ordinary satellites. Its telemetry at 19.945 MHz was caught and recorded in the West — the Sohio laboratory in Cleveland tracked the spacecraft on orbits 3, 4, 5, 11, 12, and 13. This detail matters: a Soviet military satellite of the sixties was visible live to radio amateurs and students. Back then it was stray radiation. Now it is a genre.
The program’s apogee came in October 1968, test number one in an aggressive series. On October 19, Kosmos-248 enters a 476–543 km Earth orbit inclined at 62.25° — the target, carefully placed into an orbit with a stabilized track: the track repeats day after day, making things easier for the interceptor. The next day, Kosmos-249 launches. By its second orbit, it is already on an intercept trajectory with an apogee above 2100 km: a classic attack from ahead and above, to build up monstrous closing speed. According to Russian sources, the interceptor “repeatedly approached” the target, after which it was detonated — a test of its destruction charge: Western sensors counted 109 fragments. Twelve days later — day for day, on November 1 — Kosmos-252 did the same thing against the same target, leaving 139 fragments. An amusing detail arrived later, from the 1970s: radio amateurs around the world began catching broadband buzzing in their own two-meter band, 144–146 MHz — telemetry from Soviet targets, probably with a miss-distance indicator. Engineers who hid everything else had plugged their telemetry into the most densely populated amateur section of the airwaves; Sven Grahn, the Swedish radio engineer who compiled the definitive chronicle of those tests, still wonders: accident or showcase.
The “fly up and explode nearby” scheme survived for decades as a calling card — the grid of tests in the 1970s and 1980s followed the model of 249 and 252. It is crude, it is indiscriminate, it showers orbit with debris. Modern evolution reads like a change of tools in surgery: in 2020, Kosmos-2543 released a projectile accelerating relative to its carrier to 186 m/s — an incident after which the United States and United Kingdom exchanged official condemnations; in 2026, US Space Command chief Stephen Whiting observed that Russia is “operationalizing” co-orbital weapons. The fragmentation grenade has been replaced by the scalpel: now there is no need to explode — one simply needs to be able to sit, very precisely, a hundred meters from somebody else’s spacecraft for as long as necessary. Which is what Kosmos-2612 demonstrated. It did not shoot. It stood there. That was the blank shot: a demonstration for customers at home and recipients abroad.
There is another archaeological layer — the calendar. 63 years ago, on November 1, 1963, Polyot-1 left Baikonur; 58 years ago, on that same date, Kosmos-252 went out to intercept its target. These lines are being written on November 1, 2026. A calendar coincidence, but engineers’ superstitions tend to be precise: military space programs have a strange love of autumn windows.
One could imagine this story with a military target and be three times less surprised. The point is the addressee. The satellite belongs to an American subsidiary of a Finnish company: the money comes from civilian markets, the imagery goes to armies. And ICEYE US’s first response (“we are aware, we are monitoring, the constellation architecture is built for continuity and resilience”) is the language of an insurance policy, not a military manual. The resilience logic is ironclad: a constellation of dozens of identical machines absorbs the loss of one node much as a router absorbs one disabled port. Kill one — inconsequential. Kill them all — there is no way to do it, and, judging by 2612’s demonstration, it was not in the plans yet.
But military arithmetic has its own balance sheet. ICEYE imagery since 2022 has been part of Ukraine’s intelligence network; its clients include the Pentagon and European defense ministries; as early as 2022, Russia publicly stated that commercial satellites working for Kyiv were considered legitimate targets. And now close-orbit behavior has been tried for the first time on a commercial node — as if a ceremonial formation chose not a military convoy but a civilian truck for its demonstrative drive: cheaper, quieter, but the message still gets through. Retired Lieutenant General John Shaw, former deputy commander of US Space Command, drew a direct analogy: Russian aircraft have been cutting corners near foreign borders for years, and in 2023 one such “close pass” ended with an American drone being shot down over the Black Sea. “If the Russians were willing to do that to an uncrewed military platform, their threshold for attacking an ‘uncrewed’ commercial space platform is probably not much different,” he believes. The entire story looks like a steady lowering of the threshold in the gray zone: cables, GPS jamming, drones over airports — and now one more item added to the list from above.
For the industry, this is also the quiet end of an age of innocence. Commercial space built its business model on the assumption that orbit was neutral territory, like the high seas. Historically, the “high seas” always end the same way: as soon as strategic cargo begins moving through them, people appear who learn to sail alongside it. The engineering conclusion of the past two weeks: strategic value always converts into strategic risk, and a cheap small satellite is an attractive target precisely because it is cheap and cannot go anywhere.
Finally, the most underappreciated layer of the plot. Fifty years ago, the events described in this text would have been known to three people at the CIA and, nine years later, to readers of declassified documents. Today, the full incident-observation pipeline looks like this: commercial radar company LeoLabs measures distances; commercial space-domain-awareness providers working with the Joint Commercial Operations Cell raise a flag; Integrity ISR analysts read delta-V like a financial statement — as early as May 22, five months before the headlines, they published an analysis titled “Is Russia maneuvering to threaten the ICEYE satellite?”; observer Langbroek publishes a calculation on the evening of October 17 predicting a close approach within hours; enthusiasts at s2a systems film the pair through a telescope as it passes overhead; McDowell turns the milestone into a single tweet. Within a day, everyone who wanted to know did. Within two days — everyone who did not.
The only thing not released into the open is intent. Observable things — fuel, geometry, lighting — can be measured with almost absurd precision; observable intent cannot be measured at all. That is an honest limitation of the discipline, and analysts hold to it firmly: the approach is compatible with spacecraft reconnaissance, with a demonstration of “we can,” with a rehearsal, with an attack begun and called off — the data do not distinguish among these possibilities, and any single definitive version is speculation in uniform. The perfect closing note for a column about curiosities: for the first time in history, humanity has access in real time to every movement of a hostile satellite — except the main thing for which those movements are made.
As this text was being written, the event ledger looked like this: Kosmos-2612 had withdrawn — roughly 2.5 km above the target’s orbit and nearly half a thousand kilometers behind along-track, with the separation increasing; no damage was evident in open data; its brothers remained in orbit — Kosmos-2611 and 2613 have geometry bringing them toward neighboring ICEYE-X37 in early and mid-November. Observers are watching one dry parameter for signs of a “second season”: 2612’s semimajor axis. If it starts descending, it means it is building another phasing pattern, and the lesson is not over. It has the budget for it: since launch, the satellite has spent around 175 m/s — roughly 60 m/s more than any of its four brothers. The dull counter of semimajor axes is now an indicator of a war that is not shooting yet.
And the calendar, as is customary in this column, gets the final line. Exactly 58 years ago, on November 1, 1968, another Soviet interceptor went out toward its target and showered orbit with nearly a hundred and fifty fragments — the world tracked it by radar, while radio amateurs were still catching their telemetry over the air a decade later. Today, nobody needs to wait for declassification: the stalker is visible live, with disputed meter counts and timed maneuvers. We have learned to watch. What remains is learning to answer — but that, as they say, is another column.