Divide the synthetic task by function
The fixed workshop example moves from one source note to a content brief, separate review, corrected final brief, and checkpoint export.
Each stage asks a different question:
- Which confirmed details belong in the task context?
- What brief structure serves independent consultants?
- Does the first brief add an unsupported schedule?
- Which correction resolves that finding?
- Which unknowns belong in the checkpoint?
The demonstration supports no claim about relative model quality.
Keep fixed transformations deterministic
Copying selected source excerpts, checking required fields, naming output files, and assembling an export follow declared rules. These steps do not need open-ended judgment.
A fixed process makes their outcomes easier to inspect. It also prevents creative variation from changing the record format.
Use reasoning where the task requires it
Brief structure requires synthesis. The system must connect the audience, decision, evidence, boundaries, and section order.
That stage deserves a path suited to ambiguity and consequence. A short formatting task and a source-sensitive editorial decision should not receive identical treatment.
Separate creation from review
The draft claims Saturday sessions are available. A separate review checks the source note, finds no schedule, and requests removal.
The value comes from a distinct question, not a provider name. Creation asks what brief the confirmed notes should produce. Review asks where that brief exceeds them.
Stop at the buyer's judgment
The corrected brief can resolve source and scope problems. It cannot decide whether its editorial angle matches the accountable person's intent.
Model choice supports the job. It does not inherit the final verdict.
Fit and poor fit
Task-based selection fits recurring work that mixes routine handling with synthesis or review. It adds little when one fixed script can produce the complete result.