Part of Temporal Constructs Appendix, which contains details behind the research: dates, quotations, descriptions of instruments, and sources.
General definition
A design in which a system remains usable across a range of capabilities, with no single threshold below which it fails.
Relation to the practice
Wessel and Wright (2002) treat intimate control across a range of skill as the central problem of instrument design. Russell and Norvig call the same requirement graceful degradation. The term used here names that requirement.
Because the command vocabulary is bounded, AI models of different sizes can operate the system. With a small model the artist makes more decisions. With a larger model the artist delegates more of them. The artist’s contribution scales inversely with the model’s capability. Access does not depend on expensive infrastructure.
First-year findings
Working with the agent through 2026 tested the principle and produced a result the plan had not expected. The research ran smaller models for most of its work, for cost, and because running locally is the aim. The findings did not form a fixed ranking by model size.
Among the models it tried, a smaller one was sometimes the more reliable. A larger one failed on problems that a smaller one later solved. The cost of the whole task decided the choice. That finding is scoped to those models and those months. The models available change without notice, so the arrangement is one of continuous rechoice, with no settled setting. That is itself an argument for a system that models of different sizes can operate.
Unforeseen
Here the unforeseen situations, sometimes difficult, sometimes unexpectedly interesting, will be described as the project runs. Many of these unforeseen situations have already occurred and have been recorded. They will be added in the time to come.