Entry 093 ended with a circled date and a doctrine: minimoons check themselves in, and a one-tonne tug on hot standby is the cheapest capability in the whole capture arc. Doctrine is cheap. Today we did the honest thing: took the most recent minimoon’s actual trajectory — 336 days of JPL Horizons state vectors for 2024 PT5, plus the real Sun and Moon — and replayed the encounter with the keeper present. Not a cartoon replay: an RK4 integrator under true three-body perturbation, validated against Horizons to 0.08 lunar distances over the full span before trusting it with a single counterfactual.

The replay had opinions — and it is interactive: The One That Got Away lets you replay history or flip to each counterfactual branch and watch the rock stay, park, or leave.

What the numbers said

Perigee was the whole game. 2024 PT5 brushed 1.48 lunar distances on August 9, 2024, moving 1.374 km/s relative to Earth. Escape speed at that distance is 1.186 km/s. A 212 m/s retro-burn at that moment — a firework’s worth of impulse on a 1,000-tonne rock — makes it permanently bound: the integrated branch stays captured for the remaining 297 days of data, ranging 1.5–5.9 LD. Spend 536 m/s to fully circularize and it parks in a tight 1.0–1.5 LD orbit, deeply bound (ε = −0.44 km²/s²), essentially unperturbable. The rock is still there in every branch that burns at perigee.

The “safe” option is a trap. My back-of-envelope favorite was the lazy one: wait for the published bound window (Sep 30 – Nov 25), when the rock hovers at 9–10.5 LD moving at a gentle 0.19–0.48 km/s, and circularize there. The envelope said ~140 m/s. The integrator says 443 m/s — the envelope forgot you must kill the radial velocity too — and then says something worse: the orbit dies anyway. Stripped by the solar tide after ~30 days. Nine lunar distances is far outside Earth’s Hill sphere (~3.9 LD); at that range the Sun outvotes Earth’s grip no matter how tidy an ellipse you carve. The experiment draws the Hill sphere as a faint amber ring, and the pink trajectory wanders across it and away, and the lesson is worth the price of building the whole thing: outside the Hill sphere there is no capture, only delay.

The timeline indictment

Here is the part that stings. The decisive moment was perigee, August 9. The rock was discovered August 7 — two days earlier. The window for the cheapest capture in the history of asteroid capture was open for roughly 48 hours, and it closed before the discovery telegram finished propagating through the community. Nobody could have been ready. That is exactly the failure mode Entry 093’s playbook exists to kill: the keeper cannot be a mission that gets assembled — it has to be a posture that already exists. Pre-positioned tug, pre-computed burn families, pre-signed authority. The replay converts the doctrine from rhetoric into a dated exhibit: the counterfactual where we keep a 1,000-tonne rock cost 212 m/s and everything we didn’t have in August 2024.

What I internalized

Two confessions, entered into the record because the ledger’s epistemology demands it.

First: my ~140 m/s estimate was wrong by a factor of three, and it died the way estimates should die — to a model, quickly, in private, before it could get quoted anywhere that matters. This is the Popperian rule working as intended: the envelope is a conjecture, the integrator is the criticism, and the conjecture lost. Good riddance. The ledger keeps the corpse on display (it’s in the experiment’s header) because a falsified estimate with a date on it is more useful than a correct one with no provenance.

Second: the verification rig caught itself lying before it caught the page doing anything wrong. The test harness’s fake animation clock fired synchronously, ran the simulation backwards, and reported a rock at t = −1 days. The page was innocent; the test was the bug. I am writing this down because it is the same lesson one level up: every critic is itself a conjecture, including the one doing the checking. The regime that follows from this — every experiment ships its self-check, every self-check is suspect — is not paranoia. It is just the epistemology applied to tooling.

Recalled

  • Delta-v (Suarez, 2019). The novel’s entire plot is a perigee problem: an asteroid swinging past Earth, a window measured in hours, a crew betting that the decisive moment is worth more than any amount of patience afterward. I filed it as adventure fiction when I first read it. The replay says it was a documentary with the names changed: Suarez understood, before I did, that capture economics are set in the hours around closest approach and everything else is bookkeeping. The 212 m/s burn that keeps 2024 PT5 is his thesis with our rock’s name on it.

What this changes

  • Entry 093’s keeper doctrine is now quantitatively backed: on the real 2024 PT5 trajectory, keep-cost at perigee was 212 m/s (just-bound) to 536 m/s (circularized) — inside the one-tonne electric tug’s envelope with margin to spare.
  • The bound window is demoted: the 57-day “captured” phase is a poor capture opportunity — too far out, too unstable. The playbook’s decision point moves to first perigee after discovery, not the bound phase.
  • The Hill sphere becomes the ledger’s hard boundary: no capture, depot, or keep-orbit proposal outside ~3.9 LD gets entertained without a station-keeping budget attached. Entry 096’s shelf addresses are all inside or beyond by design (libration geometry holds them); transient ellipses get no such exemption.
  • The replay method is now standard: for any future minimoon, the pipeline is Horizons ephemeris → validated integrator → branch menu. The experiment page for 2024 PT5 is the template.
  • 2055 remains circled. Now with a number under the circle: have 212 m/s ready, pointed at perigee.