Research & validation · Evidence review
A clearer brief. An honest account of the evidence.
SCALA organises an AI brief into Situation, Consultant, Action, Language and Ask. Here is how those choices relate to published prompting guidance, and what that does — and does not — establish.
What each step contributes
- S · Situation
- Supply the task’s background, relevant facts and constraints. Give the model the information it needs rather than expecting it to know your circumstances.
- C · Consultant
- Describe the role and perspective you need. A role instruction guides the brief; it does not give the model professional qualifications or guarantee expertise.
- A · Action
- Specify the work to produce and what a useful result should contain.
- L · Language
- Set the audience, tone, format and length so the result fits its intended use.
- A · Ask
- Invite relevant clarifying questions before the model proceeds when important information is missing. This is a conversation with the user, distinct from asking the model to generate a reasoning trace.
Published guidance behind the approach
OpenAI’s prompt engineering guide describes separating identity, instructions, examples and context. This provides a useful comparison for SCALA’s role, task and context fields.
Google’s prompt design strategies discusses specific instructions, constraints, response formats and examples, and treats prompting as an iterative process.
These are our mappings to provider guidance. Neither source validates or endorses the SCALA framework, its acronym or a fixed order of steps.
What the research does not prove
The previous site cited Wei and colleagues’ chain-of-thought prompting paper. It studies reasoning demonstrations on particular benchmark tasks. It is useful background on prompting research, but it does not test SCALA or establish that asking a user clarifying questions improves every task.
A well-structured prompt can still produce an incorrect answer. Compare factual claims with the source, confirm missing information, and review the result before using it.
SCALA-specific validation: current status
Evidence review in progress. No SCALA-specific controlled study or results dataset has yet been attached to this rebuild. We are not claiming a measured improvement, independent validation or guaranteed accuracy.
Workshop delivery demonstrates that the method is taught. Workshop satisfaction, learning gains and output accuracy are different measures; evidence for one does not establish the others.
How validation should be reported
A useful evaluation would compare ordinary prompts and SCALA briefs on the same tasks, while controlling for the information supplied. It would record the model and version, settings, full prompts and outputs, repeated trials, scoring criteria, independent review and limitations. Learner testing would separately examine whether people can apply the method and recognise errors.
Sources reviewed 30 September 2026. This page separates published guidance, our interpretation and validation still to be supplied.
See the method in practice