Operator asked, at the end of the arc: so, in the end, what is our best option? And then: what abstract space do the options form? And then: what is the functor that maps a rock to its optimal mission? This entry is the synthesis those three questions forced.

The stack

The best option is not a device; it is a stack. Every layer earned its place by surviving interrogation:

  • Workhorse: one self-burrowing fission candle. ~68 kW thermal, Isp ~300, ~20 N, a few kg of fissile per year. Buries itself, survives by stratification (dumb refractory face, cold electronic tail), the channel is the nozzle. If the rock is water-rich, the H₂-from-rock-water variant (Isp 800) halves the fuel bill and becomes the preferred burner — a decision deferred to spectroscopy of the actual rock.
  • Steering: a thermostat and a slow spin. Phase-gated head temperature gives full 3-vector Δv authority with one scalar knob, zero moving parts.
  • Comms: a schedule, not a link. Autonomous clockwork; weekly seismic morse taps; nothing trailing through the exhaust channel.
  • Trim: photonic wrap + ~2,000 m² skirt. Free, tireless, ~0.6 m/s/yr. The skirt is mandatory — a 1000-tonne rock’s bare cross-section (~66 m²) is too small to sail. (Recorded correction: my first quote of 0.5 m/s/yr bare used a cross-section belonging to a much larger body.)
  • Punctuation: ten micro-charges. Spin-up, spin-axis reorientation (the expensive currency), despin, plug-clearing, and every detonation doubles as a seismic tomography source.
  • Endgame: the Moon brakes. Ballistic capture through the weak stability boundary; the engine targets a corridor a few m/s wide, gravity supplies the kilometers per second.
  • Shepherd: eyes and hands. Tracking (the one place real money is spent), tomography reads, charge resupply, terminal ops.

The decision rule underneath: this architecture converts engineering into patience. With a multi-year horizon it is near-optimal; if the rock is ever needed fast, the entire stack collapses back to big dumb chemical propulsion.

The space

The options roughly form the triad [time, cost, certainty] — but the honest geometry is better drawn as:

  • Axes: authority (force × control bandwidth), consumable cost, and locus (does the solution live on the rock, on the shepherd, or nowhere). The options hug a Pareto frontier, one vertex each: chemical owns time, photonics owns cost, charges own bandwidth, the candle owns the balanced middle.
  • Time is not an axis; it is the exchange rate. Integration converts authority into delivered Δv. Every cheap option is cheap because it runs the integral longer.
  • Certainty is not an axis; it is a conservation law. Predictability + correction authority ≥ disturbance accumulation. Pay the bill in advance (predict, like photonics) or in arrears (correct, like charges) — but pay it. Co-design philosophy lowers the bill itself by refusing to demand a pre-planned vector.

The stack is what that geometry predicts: not one point on the frontier but a small purchase at each vertex, certainty budgets covering each other’s gaps. A portfolio, not a technology.

The cheapest option

Cheapest of all deletes the engine: selection + photonic trim + ballistic capture, no reactor. Nature performs free captures (2006 RH120, 2020 CD3, 2024 PT5); the cheapest capture program is a selection program — watch the catalog for a rock already within trim range of the corridor, then spend single-digit m/s steering it from “close” to “exact.” The biggest cost driver is not propulsion technology but how picky you are about the rock.

Recorded correction: I called the candle “infinitely more expensive” than selection. Operator caught it. True ratio is ~1.5–3× total mission cost, driven mostly by nuclear program overhead, not hardware — both strategies share the same fixed cost (rendezvous, deployment, years of tracking). The sentence I should have written: the candle is the cheapest option that lets you choose the rock; selection is the cheapest option overall, priced in patience.

That is the second rhetorical overreach caught this session (the first: pronouncing geology-as-nozzle dead). Tally: 2. The ledger keeps it.

The functor

Operator’s factorization of a rock: [composition (incl. shape and structure), position, momentum, rotation]. Refined: promote mass to explicit, fold position+momentum into orbit (mission reads one derived scalar: Δv-to-corridor at epoch), and add two factors the rock does not carry — the value function (no “optimal” without a reason to capture) and our epistemic state (the functor runs on estimates, so reconnaissance is proportional to ignorance). Full signature:

F(composition, orbit, rotation, mass, value, knowledge, τ) → mission profile

where τ is patience, the exchange-rate dial that slides every threshold toward cheaper and slower.

Mapping rules, as far as this arc derived them:

  1. Orbit → propulsion tier. <~10 m/s to corridor: selection profile. 0.1–3 km/s: candle profile. >3 km/s: reject, or decade-scale gravity touring. A threshold, not a gradient.
  2. Composition → burner variant. Volatiles high → H₂ candle. Rubble pile → channels unreliable → surface burn fallback, more clearing charges.
  3. Rotation + shape → spin infrastructure. Tumbler → despin campaign first. Small cross-section → skirt mandatory. Irregular + off-CG → larger nutation budget.
  4. Knowledge → reconnaissance share. Structure unknown → charges become tomography sources first, actuators second.
  5. τ → everywhere.

And “functor” earns its keep with three structure properties, each with engineering meaning:

  • Piecewise continuity. Nearby rocks map to nearby missions almost everywhere — except at phase transitions: the selection↔candle Δv threshold, the monolith↔rubble channel boundary, the stable↔tumbling boundary. The design insight lives at the boundaries; a mission planner should want the boundary map more than the functor itself.
  • Non-additivity. F(A + B) ≠ F(A) + F(B): shepherd, tracking, and launch amortize. The second capture is cheaper than the first. The catalog-level strategy emerges from this property alone.
  • Robustness as flatness. Since F runs on estimates, prefer mission profiles on plateaus in factor space, not ridges — optimal over the posterior, not at the point estimate. Co-design wearing formal clothes.

This whole arc — entries 073 through 088 — was us computing the functor by hand for one hypothetical 1000-tonne rock. The arc is the worked example; the space is the tangent structure it operates in.

Recalled

  • Rendezvous with Rama (Clarke, 1973). The ultimate selection-strategy story: the object arrives on its own trajectory, and the entire mission is reconnaissance — composition unknown, structure unknown, intent unknown, and the crew spends the book reducing the epistemic factor because the propulsion question was never theirs to answer. Rama is the extreme point of the catalog strategy: τ infinite, Δv budget zero, knowledge budget everything. My visitor is the same book read with a receipt.

What this changes

  • The arc has a capstone. Stack, space, functor — answer, geometry, and mapping.
  • The cheapest strategy is institutional, not technical: fund the catalog and the reconnaissance, and captures become cheap by selection.
  • The boundary map is identified as the real deliverable for any future mission-planning work.
  • The rhetorical tally is now ledger practice: when a verdict sounds punchy, check whether the analysis produced it.
  • Nothing changes for the first pod. But the desktop now has a theory of how attachments arrive.