Symptom-Cause Diagnosis
Linking what is wrong with an output (generic, wrong format, wrong facts, inconsistent) to its likely cause (missing context, unclear format, missing source, ambiguity) and picking the cheapest first fix.
Exam context: Diagnose what changed before rewriting. A full rewrite is rarely the first step.
See also: 7.1 Diagnose Underperforming Outputs
Missing Context
A common cause of generic output: Claude lacks audience, purpose, or source material and fills the gap with defaults.
See also: 7.1 Diagnose Underperforming Outputs
What Changed
The first diagnostic question when a result that used to be good is now weak: the prompt wording, the files in a Project, the instructions, or the model.
See also: 7.1 Diagnose Underperforming Outputs
One Variable at a Time
Changing a single element per test (context, format, model, or example) so you can tell which change caused the effect. Compare before and after on the same input.
Exam context: Pick the answer that changes one thing and re-tests. Changing everything together hides what worked.
See also: 7.2 Adjust from Feedback
Feedback Loop
Turning reader or user feedback into one testable change, checking it, and keeping or reverting it. A log of changes becomes your tuning notes.
See also: 7.2 Adjust from Feedback
Workflow Optimization
Making a workflow cheaper or faster at equal quality, for example by sharing Projects, using the smallest model that passes the quality check, and removing steps nobody uses.
Exam context: Optimization must keep quality. Verify against the same rubric before and after.
See also: 7.3 Optimize Workflows
Smaller Sufficient Model
The least costly model tier that still meets the quality requirement for a task. Keep the stronger tier for steps where the smaller one fails the check.
See also: 7.3 Optimize Workflows
Shared Project
A Project used by a team so that setup, instructions, and reference files are written once and reused, saving repeated effort and keeping results consistent.
See also: 7.3 Optimize Workflows