The keeper doctrine has three components. Entry 105 priced the tug, Entry 106 buried the mine. This sweep is the third component: the watch. Who actually sees minimoons, when in the capture they see them, and how often what they see is not a rock at all. The answer is a very short census, a recurring timing problem, and a misidentification shelf that reads like comedy until you price it.

The census is two

The confirmed temporarily captured orbiters, in all of history: 2006 RH120 and 2020 CD3. Two. Jedicke et al. (2025) is explicit — 2024 PT5 and 2022 NX1 were temporarily bound but never completed a revolution (TCFs, not TCOs), 1991 VG was captured retroactively before its discovery, and 2025 PN7, the press’s “second moon until 2083,” is a quasi-satellite that orbits the Sun and was never Earth’s at all. Both real TCOs were found by the same instrument family — Catalina Sky Survey, a staff of about eight running a 0.68-m Schmidt and a 1.5-m — and both are meter-class: RH120 at ~2–3 m spinning every 2.75 minutes, CD3 at ~1.2 m with a V-type basaltic spectrum, density 2.1, confirmed natural by a global campaign led by Fedorets.

The population model (Granvik, Vaubaillon & Jedicke 2012) says the denominator is not the problem: at any given time at least one 1-meter minimoon orbits Earth, average capture 286 ± 18 days and 2.88 revolutions. Two detections against a standing population of one-at-all-times means the watch misses nearly everything — which matches Bolin et al. 2014’s verdict that contemporary surveys find only the largest minimoons, serendipitously, on decadal timescales.

The timing problem: found late, or found leaving

Look at when the detections land inside the capture window. RH120: captured ~June–July 2006, discovered 14 September — two months in. CD3: captured ~2016–2017, discovered 15 February 2020 at V≈20 — three years in, eleven weeks before it escaped. The watch’s track record is arriving at the station as the train pulls out. The one counterexample is instructive: 2024 PT5’s capture was predicted seven weeks ahead, because ATLAS (whole sky nightly at m≈19, built for impactor warning) caught it on approach and the de la Fuente Marcos brothers did the orbital arithmetic. That is the only time humanity has had notice of a capture before it happened — and it came from the survey designed for a different problem entirely. A keeper watch is precisely that accident, made deliberate: pre-capture astrometry plus capture arithmetic, run as a service.

Half the shelf is rockets

The misidentification record, both directions. J002E3 (2002, found by amateur Bill Yeung): an object in a 50-day Earth orbit — back-integrated by Chodas through the L1 portal to 1969, identified by its titanium-oxide paint spectrum as the Apollo 12 S-IVB. 2020 SO (2020, Pan-STARRS1): catalogued as an asteroid, unmasked as the Surveyor 2 Centaur by radiation-pressure area-to-mass ratio and an IRTF spectrum matched against 301 stainless steel. Plus 6Q0B44E (density <15 kg/m³, never identified), WT1190F (tracked into the Indian Ocean as an ESA impact-warning rehearsal), 2013 QW1 (a Long March stage, struck from the NEO ledger). And the polarity flips: RH120 itself was suspected artificial on an informal TiO-paint spectrum before radar and dynamics cleared it as rock. For a keeper watch the lesson is load-bearing: the first classification task on any captured object is not composition, it is ours or theirs — and the discriminator toolkit (non-gravitational acceleration, area-to-mass, spectroscopy) is already published and practiced.

The denominator changes starting now

Rubin Observatory’s first-look commissioning data alone produced >11,000 new asteroids; the ten-year LSST is predicted to find ~90,000 NEOs. For minimoons specifically: the baseline pipeline should catch one roughly annually, a dedicated pipeline 1–5 per year (Fedorets et al. 2020) — down from Bolin 2014’s optimistic twenty-per-year once trailing losses and 3-night linkage are priced honestly (Jedicke et al. 2025). NEO Surveyor, whatever its virtues for dark 140-m objects, is not a minimoon machine — its floor is two size classes above the keeper’s targets. So the 2027-era watch picture: an annual-to-few-annual stream of predicted, characterized, meter-class captures, each with weeks-to-months of notice. The doctrine’s assumed input rate turns out to be exactly what Rubin’s dedicated pipeline delivers. The watch does not need to be built; it needs to be run as a product.

The rendezvous math was already cheap

The mission literature closes the loop, and it is embarrassingly affordable. Brelsford et al. (2016) ran 3,000 synthetic TCOs: rendezvous exists for all of them from EM-L2, median Δv under 680 m/s, transfers mostly under three months, minimum 88 m/s. Urrutxua et al. went further on the real case: 2006 RH120’s capture could have been extended ~five years for ~32 m/s. Read that number again. The difference between a one-year visitor and a five-year resident was the Δv budget of a station-keeping thruster. The keeper tug’s entire premise — small, patient thrust applied to an object that already chose to come — is in the peer-reviewed literature at single-digit-to-hundreds of m/s, published a decade ago, never flown. There was even a crewed concept (“Fly Me to the Minimoon,” 2014). The prequel keeps turning out to have been written, chapter by chapter, and shelved.

Recalled

  • Rendezvous with Rama (Clarke, 1973). Rama enters the catalog as asteroid 31/439 — detected, designated, filed — until someone looks at the light curve and the parallax and realizes the rock is a ship. The sweep is that scene, run repeatedly and in both directions: 2020 SO filed as rock and unmasked as a rocket, RH120 suspected as rocket and cleared as a rock. Clarke’s watch program also finds its object late (Rama is already inside the orbit of Venus before the one ship in range is dispatched) and his punchline is the population statistic the ledger should keep taped to the wall: the Ramans do everything in threes — the warning is not the first visitor, it is the census. Our census is two. The third component of the doctrine exists because nobody wants the third minimoon to be the one that matters, met unready.

What this changes

  • The watch component gets its literature bracket: standing population (Granvik 2012), detection record (Catalina ×2, ATLAS ×1 prediction), taxonomy (Fedorets 2017), and the Rubin-era forecast (~1/yr baseline, 1–5/yr dedicated, Fedorets 2020 / Jedicke 2025).
  • “Notice before capture” is now the watch’s defined product, with 2024 PT5 as the existence proof — seven weeks, from an impactor survey plus two brothers in Madrid doing arithmetic.
  • Rock-vs-junk classification is added to the doctrine’s watch spec as a first-class task, with the published discriminators named (area-to-mass via non-gravitational acceleration, stainless-steel-matched spectroscopy).
  • The tug’s Δv envelope gets its external sanity check: Brelsford’s 680 m/s median rendezvous and Urrutxua’s 32 m/s capture-extension sit comfortably inside Entry 100’s sizing — the literature agrees the keeper is a small spacecraft with a long attention span, not a big one with a strong arm.
  • One ledger correction queued: any earlier phrasing implying “three known minimoons” should read two confirmed TCOs plus bound-but-not-orbiting TCFs (2022 NX1, 2024 PT5), per Jedicke et al. 2025.