1. The spiders and the teacher

Adrian Tchaikovsky’s Children of Time is about a species that learns across generations, guided and misguided by the tools its ancestors left behind. The lesson is that intelligence is not only in the design. It is in the ability to learn from experience over time. The desktop will not be perfect on launch day. It will be taught by telemetry, anomalies, operator decisions, and the slow accumulation of flight hours.

Entry 306 covered optionality and modularity. This entry focuses on adaptation and learning.

2. What the desktop can learn

A platform in orbit generates data about itself constantly. From that data it can learn:

  • Degradation patterns: which components age faster than expected.
  • Operational margins: how much power, thermal, or propulsion capacity is really available.
  • Failure signatures: the early signs of component problems.
  • Environmental conditions: the actual radiation, atomic oxygen, and debris environment.
  • Customer usage: which services are demanded and when.

This learning is valuable only if it is captured, analysed, and fed back into design and operations.

3. Human and machine learning

Learning can be human-driven, automated, or both:

  • Operators analyse telemetry and update procedures.
  • Engineers review flight data and revise designs for the next unit.
  • Software monitors thresholds and triggers safe modes or alarms.
  • Autonomous systems adjust pointing, power, or thermal control in real time.
  • Machine learning models detect anomalies or predict failures from patterns.

The desktop does not need full autonomy on day one. It needs a learning loop that improves over time.

4. Feedback from orbit to design

The most important learning is the loop from operations back to design. If a radiator coating degrades faster than expected, the next unit uses a different coating. If a software safe mode triggers too often, the logic is refined. If customers use compute more than storage, the next platform is rebalanced.

This loop requires discipline: data must be stored, analysed, and acted upon. Without it, the same mistakes fly again and again.

5. Adaptation as culture

Adaptation is not only technical. It is cultural. The team must be willing to:

  • Admit when a design assumption was wrong.
  • Change course based on flight data.
  • Deprecate attachments that do not perform.
  • Invest in improvements that only show value over years.

A programme that cannot learn from its own spacecraft is flying blind.

What this changes

  • The desktop is expected to learn from telemetry, anomalies, and operational experience.
  • Learning can be human-driven or automated, and both should be used.
  • The critical loop is from operations back to design.
  • Adaptation requires a culture that accepts error and acts on evidence.
  • The next entry will close the uncertainty and resilience arc with a contemplation.