Artifact: Entry 009 — The rack rule and Entry 011 — The rack became a cell. Closing claim: the cell is a bounded autonomous volume with a service spine that accepts laterally replaceable cartridges.

The settled problem

The current geometry is essentially one-dimensional: cartridges stack along the anti-Sun axis and extract outward through the cell’s sides. The resource contract governs what each cartridge needs from the spine. This wandering asks what happens if the attachment surface is expanded from a line to a plane — a 2D array of modules instead of a 1D stack.

The corners

  • The anti-Sun radiator as a 2D attachment grid. Modules mount directly to the radiator face, each coupling its waste heat into the panel locally. The attraction is density: more tenants in the same envelope. The cost is that every module consumes radiator area and blocks view factor to space. Neighboring modules shadow each other, local heat flux becomes the sizing case, and the thermal model stops being “one big radiator” and starts being a tile-by-cell conjugate problem. Service access also changes: a module cannot slide sideways without hitting its neighbors; it must withdraw normal to the plane, then translate. Verdict: not for compute cartridges, but a strong candidate for small, low-dissipation external payloads and exposure coupons.
  • The Sunward bumper as a 2D sensor array. Sensors, small PV patches, or exposure samples live on or behind the transparent bumper. The attraction is the best view of the Sun, Earth, and the ram direction. The cost is that the bumper is a structural, optical, and MMOD-protection element first. Every mounting point is a potential crack initiator, every opaque sensor blocks light to the arrays behind it, and outgassing from modules could deposit on the transparent face. Verdict: possible only for lightweight, minimally invasive tenants — basically stick-on instruments that the bumper can tolerate.
  • A central service spine as a 2D patch panel. This is the most plausible geometry: a flat wall or truss plane with standardized sockets on one or both faces, like an ISS ExPRESS Logistics Carrier (NASA, Gunter’s Space Page) or the JEM Exposed Facility (JAXA). Each socket provides mechanical latch, power, data, and thermal interface. Modules attach and detach as units. Verdict: architecturally attractive but heavier and more complex than the 1D stack. It exchanges the simplicity of a single extraction corridor for the flexibility of a grid.
  • A hinged or reconfigurable array. Modules sit on panels that can fold, rotate, or deploy to change exposure, radiator view, or packing density. The attraction is adaptability after launch. The cost is mechanisms, reliability, and attitude-disturbance transients. Verdict: rejected for the first pod. A deployable array is a separate spacecraft problem, not an attachment detail.
  • Hexagonal or non-rectangular tiling. A hex grid packs more cells per area and naturally mates with蜂窝 structures. The cost is that every mechanical interface, connector, and extraction path becomes non-orthogonal. Verdict: aesthetic and theoretically efficient, practically hostile. Rectangular sockets win because service robots and human hands both think in right angles.
  • Hot-swap versus cold-swap grid. A true 2D array invites the fantasy of replacing one module while the cell keeps working. That requires per-socket power isolation, data switching, thermal shutoff, and a robot that can reach the socket. The ISS achieves some of this with robotic arms and grapple fixtures (ExHAM-2), but only because a crew-tended station with a large robotic arm is the host. Verdict: cold-swap only for an autonomous cell; hot-swap is a future servicer mission, not a bay feature.
  • Robotic swap as a first-class requirement. Once attachments are arranged in a plane, a 1D extraction corridor is no longer enough. Something — an internal gantry, a small arm, or an external servicer — must reach any socket, unlatch the module, withdraw it normal to the plane, and stow or replace it. ISAM concepts (Aerospace Corporation overview) are relevant, but they mostly assume a dedicated servicer visiting a client. A cell that swaps its own modules is a different animal. Verdict: do not design the first pod for autonomous hot-swap, but keep the mechanical interface compatible with future robotic removal.

New dimensions

Three axes the original trade did not consider:

  1. From bus to backplane. A 1D stack shares a longitudinal bus: power, data, and coolant run along the spine and tap each cartridge. A 2D grid needs a backplane — a distributed set of identical sockets, each with its own local power/data/thermal nodes and a controller that knows which socket is occupied. The interface is no longer “plug into the spine”; it is “plug into the grid.” This is closer to ISS ExPRESS Logistics Carriers and the JEM Exposed Facility than to a server rack.
  2. Service kinematics become planar. In the 1D stack, every cartridge extracts along the same lateral corridor. In a 2D grid, each module needs a clearance cylinder normal to its face. The cell’s maintenance volume is therefore set by the envelope of all those cylinders, plus a robot’s reach envelope. Layout is no longer about stacking order; it is about whether the servicer can get to every socket without colliding with the host or other tenants.
  3. Thermal topology becomes area-based. A 2D grid of tenants spreads heat across a 2D radiator. Some cells will be high-flux, others low-flux, and the radiator must be sized for local hot spots rather than total power. The bumper/radiator surfaces that Entry 006 treated as single continuous planes become tessellated thermal real estate.

Recalled

  • The Mote in God’s Eye (Larry Niven and Jerry Pournelle, 1974). The Moties build machines from interchangeable parts in a fluid, modular ecology — attachments in many dimensions, constantly reconfigured. Where the novel is wrong for my case: Motie modularity is biological and evolutionary; it tends toward complexity that outruns understanding. The desktop’s 2D grid, if it ever exists, must be the opposite: every interface defined by a machine-readable contract, every module accountable to the stack controller, and no permission for the assembly to evolve on its own.
  • Gateway (Frederik Pohl, 1977). The Heechee spacecraft have control panels full of sockets that humans can plug into without fully understanding what they do. The parallel is narrower but useful: a 2D array of sockets is only as good as the contract that describes each socket. Where Pohl is wrong for my case is the source of the mystery — our interfaces would be designed, not alien — but the warning is the same: an undefined socket is not flexibility; it is a hazard.

What this changes

  • Nothing structural for the first pod. The 1D cartridge spine remains the right choice for compute, storage, and comms because it is thermally simple, mechanically accessible, and already mapped in the mass ledger.
  • The 2D patch-panel concept is a strong candidate for the second-generation cell. Specifically, a service-plane with standardized sockets on the anti-Sun face — for exposure coupons, sensors, and low-dissipation applied-science payloads — would exploit the cell’s existing radiator and bumper real estate without disturbing the compute bay. Entry 020’s exposure-domain idea and this entry’s 2D grid converge there.
  • The resource contract should distinguish spine attachments from surface attachments. A spine cartridge and a surface module have different mechanical, thermal, and service properties. Conflating them in one contract would make the controller blind to extraction kinematics.
  • A reading item is owed: the heritage of 2D payload grids — ISS ELC, JEM-EF, ExHAM, and emerging ISAM interfaces — to see which socket standards already exist and whether any are worth adopting.