Reasoning models need a clear architectural brief
A note on o3-mini and two proposed review exercises: separate the evidence, assumptions and questions that remain open.
OpenAI introduced o3-mini on 31 January 2025 as a reasoning model oriented toward science, mathematics and coding. Its release documentation describes text input without image understanding. That makes it a useful reference for thinking about what a written architectural task must contain.
Exercise one: inspect a planning question
Provide the applicable, dated rules with the project quantities. Ask the assistant to separate stated facts, assumptions and missing information, and to show the calculation steps with a source for each rule.
The responsible designer must check those sources and calculations. A coherent explanation without the applicable references cannot establish project compliance.
Exercise two: outline a specification
Ask for a structure that identifies which performance values need confirmation. Supply approved values where they exist; leave missing ones explicit. The model should help organize the draft without inventing project requirements.
A useful output gives the architect and consultants something specific to review: the requirement, its source and the decision still needed.
Keep the review reproducible
- Retain the supplied rules and project inputs with the draft.
- Distinguish a quoted requirement from the assistant’s suggestion.
- Check calculations independently before using the result.
The output we would assess in these two exercises is a draft whose sources can be followed. Another designer should be able to see what information the assistant used, what it assumed and which decisions remain open.