1. The harvester who ran on a schedule

Duncan Jones’s Moon returns as the recalled work because Sam Bell’s lunar outpost is an exercise in automated energy infrastructure. Helium-3 harvesters roam the surface, the base keeps its own lights on, and a single human maintenance presence is stretched across a three-year shift. The station does not ask permission every time it charges a battery or starts a compressor; it follows a regimen. The desktop is smaller and closer to Earth, but the same question applies: who decides when to store energy, when to spend it, and when to refuse a request?

This entry wonders how the desktop predicts its own energy balance and acts on that prediction.

2. State of charge is a guess dressed as a number

A battery does not have a fuel gauge like a car tank. State of charge is inferred from voltage, current, temperature, and a model of how the chemistry behaves under those conditions. The estimate drifts. The model ages. A cell that reads 80% at 20°C may read 70% at -10°C for the same stored energy. Coulomb counting helps but accumulates error. Impedance tracking is better but needs calibration cycles.

For autonomy, the key insight is that state of charge is not a measurement; it is a belief. The battery manager should maintain a probability distribution or at least a confidence interval, and the energy planner should make decisions robust to the lower end of that interval.

The desktop can improve its belief by exploiting known events. A full sunlight period after an eclipse is a chance to reconcile the model against measured recharge. A known constant-current discharge through a calibrated load is a chance to check capacity. The Resident does not need laboratory precision; it needs a conservative enough estimate that the platform does not quietly die.

3. Forecasting the energy ledger

The power reading arc introduced the energy budget as a static design artifact. In operation it becomes a rolling forecast. At any moment the autonomy should be able to project forward several orbits and estimate:

  • Generation: from the Sun vector, array geometry, predicted occlusion, and degradation model.
  • Demand: from scheduled payload operations, communications passes, attitude maneuvers, thermal heaters, and always-on housekeeping.
  • Storage: from the battery model, including temperature effects and cycle-dependent capacity fade.
  • Margin: a reserve kept for faults, unexpected eclipses, or contingency operations.

The forecast horizon matters. A five-minute horizon catches an imminent undervoltage. A ninety-minute horizon catches an eclipse. A multi-day horizon catches a string of bad geometry or a degraded cell. The planner should run at nested horizons: fast protection loops at the bottom, orbit-average planning in the middle, mission-level degradation tracking at the top.

4. Deciding when to say no

The most interesting battery decisions are not about charging; they are about refusing load. The desktop will receive overlapping requests: a payload wants to image, a radio wants to transmit, a heater wants to warm a battery, a user wants to run a job. Each has a value and an energy cost. The energy manager must choose a subset that fits the forecast.

One approach is a priority table. Housekeeping, thermal survival, and commandability are highest. Science and customer payload operations are next. Discretionary tasks are lowest. The table is simple but brittle; a hard priority can starve a medium task forever if higher tasks are greedy.

A better approach is value-aware scheduling. Each request advertises a deadline, a duration, an energy cost, and a value. The scheduler solves a knapsack-like problem over the forecast horizon. The result is a plan that may defer a high-value task by ten minutes to fit a better sun window, or shed a low-value heater cycle to preserve margin.

5. Depth of discharge as a strategic variable

Depth of discharge is not just a battery-sizing input; it is a knob the autonomy can turn. In an emergency, the desktop can draw deeper than the nominal limit, accepting accelerated aging in exchange for mission survival. In normal operation it can stay shallow, preserving cycle life. The decision should be explicit and logged, because a battery that died early should have a story attached to it.

6. What this changes

  • Battery state of charge is an estimate with uncertainty, not a ground truth.
  • Energy management is a forecasting and scheduling problem, not just a protection problem.
  • Load shedding should be value-aware, not merely priority-based.
  • Depth of discharge can be modulated by situation.
  • The next entry will wonder about autonomous fault protection and emergency power modes.