Sol, Terra, Luna — pick the right tier

Model selection is an economic decision under uncertainty. Use Sol when errors are expensive or synthesis is genuinely hard, Terra for most professional drafting and analysis, and Luna for high-volume tasks with simple checks.

A reliable routing pattern is Luna for preparation, Terra for the main draft, and Sol only for disputed sections or final critique. Do not pay flagship rates to normalize filenames; do not use the cheapest tier to settle a consequential legal interpretation.

Workshop: Sol, Terra, Luna — pick the right tier

The practical objective of this chapter is routing work by uncertainty and failure cost. Before opening a chat or Work, write down how the job is performed today, who owns it, and what counts as an acceptable result. Choose one example you can personally inspect. This baseline prevents the novelty of the tool from being mistaken for real improvement and gives you a fair comparison for elapsed time, accuracy, corrections, and usefulness.

Worked example: a finance team uses Luna to normalize categories, Terra to draft variance explanations, and Sol to challenge two disputed assumptions. The team first narrows the outcome and assembles this pack: a fixed sample, expected outputs, correction severity, elapsed time, and token cost for each tier. It then runs the agent with an explicit stopping condition, checks every material claim, and records corrections. The example succeeds only when a named owner can explain why the result was accepted, which parts remained human work, and exactly what should change on the next run.

The most common misapplication is choosing the flagship for every task or optimizing price before establishing acceptable quality. The correction is not automatically a stronger model. First reduce scope, rank sources, reveal hidden constraints, and add an approval checkpoint. If the problem remains after those changes, test a higher tier or effort setting on the same sample. This allows you to compare cause and effect instead of merely comparing two different-looking outputs.

• Describe the current human process in five sentences.
• Prepare an evidence pack that includes a fixed sample, expected outputs, correction severity, elapsed time, and token cost for each tier.
• Set one measurable acceptance rule and one prohibited action.
• Ask the reviewer to classify errors by severity, not only style.
• Save the brief, settings, output, and corrections as one run record.