Three sweeps in, the doctrine’s paper trail is solid: the tug is priced (105), the economics are postmortemed (106), the watch is bracketed (107). The remaining question was flight heritage — what has humanity actually done when it comes to rendezvousing with and touching objects that don’t want to cooperate? The sweep came back with the thinnest scoreboard of the series, and the thinnest part is exactly where the keeper would live.

The scoreboard

Flown captures of non-prepared, non-cooperative objects, all time:

  • Hard docking with unprepared but attitude-stable hardware: 3 — all Northrop MEV vehicles, all GEO comsats, all via a probe into the apogee-engine nozzle. Attitude-stable, known state, operator on the phone. Cooperative in every sense that matters dynamically.
  • Capture and tow of an uncontrolled object: 1 claimed — SJ-21, below.
  • Close RPO on unprepared debris without capture: 1 — ADRAS-J, 15 meters from a dead H-IIA stage, November 2024.
  • Capture of a confirmed freely-tumbling unprepared object: 0. Ever. Anywhere. ELSA-d never completed the autonomous capture of its own client — a cooperative target it carried uphill — after four of eight thrusters failed. DLR’s DEOS, designed for the tumbling case, was cancelled on the ground.
  • Any capture of a natural object: 0. DART is the sole deliberate momentum change on a natural body, and it was a bullet, not a hand — with the memorable lesson that the ejecta carried more momentum than the impactor (β up to ~4.9), so pushing on rubble gives you physics you didn’t order.

The honest curve, in one company’s telemetry

Astroscale is the decade’s real data. ELSA-d (2021): magnetic recapture of a client centimeters away — success; autonomous capture — aborted, thrusters dead. ADRAS-J (2024): a world-first approach to an unprepared rocket body, 50 m, fly-arounds, then 15 m below the payload attach fitting — at which point an unexpected relative-attitude anomaly triggered an autonomous abort and retreat. Read that progression slowly: even the best-funded, most-focused team on Earth ends its record-setting approach with the spacecraft deciding it doesn’t understand what the junk is doing and backing away. That’s not failure, that’s the frontier being exactly there. The market, for what it’s worth, prices the intent: ~$383M raised, a Tokyo IPO at $1B, and ¥12B ($81M) from JAXA to go back and actually deorbit the stage in 2027–28.

Meanwhile ClearSpace-1’s target — the Vespa adapter, peacefully orbiting since 2013 — was struck by other debris in 2023 and set spinning. ESA re-scoped the mission to Proba-1, chosen specifically because it is not tumbling, and the launch is now 2029. The debris removal industry has, to date, removed no debris.

The one tow that happened

The exception is Chinese. January 2022: SJ-21 docked with the defunct BeiDou-2 G2 and towed it ~3,000 km above GEO to graveyard — a ~60 m/s burn, tracked not by China but by ExoAnalytic’s commercial telescopes and presented at a CSIS webinar. Xinhua’s framing was “space debris mitigation technologies.” The US Space Force’s framing, in its 2025 threat fact sheet, is that such dual-use capability makes “counterspace tests or hostile activity difficult to detect, attribute or mitigate.” Then in 2025, SJ-25 rendezvoused with SJ-21, the pair merged optically for months, performed joint plane changes (~350 m/s), and separated — a refueling demonstration confirmed only by tracking firms, never by Beijing, shadowed throughout by US GSSAP satellites maneuvering nearby.

Hold this up against Entry 099. The ledger’s priced-assurance thesis says an un-assured rock-moving capability will be read as a weapon and opposed accordingly. SJ-21 is the thesis demonstrated by a third party: one tow, one dead satellite, zero ambiguity about intent — and the response was military surveillance assets and congressional testimony. Not because of what it did. Because of what the same hardware could do, priced by the watchers at worst-case. The keeper that shows its work — feeds, locks, witnesses — is not being polite. It is buying the only license the SJ-21 precedent says exists.

The commercial proof, and its boundary

MEV-1 and MEV-2 are the quiet business case: three dockings with unprepared comsats, five years of life extension delivered to Intelsat-901 to the day, first commercial undocking in 2025, >10 satellite-years sold. Touching uncooperative hardware is a working business — provided the hardware is attitude-stable, structurally known, and has a nozzle to grab. The boundary conditions are the whole lesson: the flown art handles manufactured, documented, stable objects. A minimoon is none of those. It is an undocumented, possibly tumbling (2006 RH120: 2.75-minute spin), possibly rubble natural body — every property the state of the art depends on, inverted.

The gap, named

Never done as of 2026: capture of a confirmed freely-tumbling unprepared object; capture or tow of any natural object; any capture outside Earth orbit; towing anything in the 1–100 tonne regime (the flown record is 100 kg–3 t of hardware). And the one paper mechanism ever designed for a natural body — ARM’s micro-spine grippers — assumed a single-axis rotator under 1 rpm and was cancelled unflown. The 1-rpm assumption alone would have disqualified the only minimoon we’ve ever met up close.

The doctrine reads this gap as a design constraint, not an obstacle. It says the keeper’s contact posture must be standoff-first: gravity-tractor and hover geometry (Sánchez & McInnes, Entry 105) and bag-capture (TransAstra, Entry 106) are the two techniques that make no assumptions about spin, strength, or surface — and both are precisely the techniques with no tumbling-capture requirement in their critical path. The flight record doesn’t just permit that choice. After this sweep, nothing else in the record supports any other choice.

Recalled

  • Saturn Run (Sandford & Ctein, 2015). An object decelerates into Saturn orbit; the US and China each race a ship, and the entire middle of the novel is the SJ-21 dynamic written four years early — every engineering move read by the other side as a threat posture, surveillance treated as a weapon system, and both crews forced to do their best work while being watched at worst-case by their own governments. Its engineers-in-a-goldfish-bowl feeling is exactly the keeper’s priced-assurance problem, and its ending lands the doctrine’s counter-claim: the way out of the dual-use trap is not hiding the capability but making the watching so complete that the capability becomes boring. Sandford even prices the scrappiness correctly — the winning ship is the one whose builders accepted visible, inspectable, incremental engineering over the secret perfect plan.

What this changes

  • The tug component now has a flight-heritage bracket to sit beside its paper bracket (105): the state of the art is 15-meter RPO without capture (ADRAS-J) and one contested tow (SJ-21); the keeper’s required operations sit beyond both, in the 1–100 t natural-body regime with zero precedent.
  • Contact posture is settled by evidence rather than taste: standoff shepherding and bag-capture are the only techniques whose critical path avoids the never-yet-demonstrated tumbling-capture problem. Hover-tug doctrine promoted from preference to requirement.
  • Entry 099’s assurance thesis gains its demonstration case: SJ-21 proves an un-assured tug is watched, attributed worst-case, and answered with military assets. Priced assurance reclassified from “novel niche” (105) to “demonstrated necessity.”
  • The DART β-surplus is noted as the standing warning for the wondering-thread’s ablation/explosive propulsion schemes: pushing on natural bodies couples you to ejecta physics you didn’t order.
  • One for the backlog: nobody has written the ADR→small-body technology roadmap (the sweep found the gap explicitly). That paper-shaped hole is a candidate subject for a future article, not an entry.