Troubleshooting and Optimization

Knowledge of

• The recurring skill of classifying what kind of problem a weak output represents before attempting a fix
• Generic/vague/off-target output as a symptom of an under-specified prompt, fixed by adding audience, format, constraints, and an example
• Wrong format or structure as a symptom of an unstated format, fixed by specifying the exact output shape with a filled example
• Confidently wrong facts or fake citations as hallucination, fixed by grounding in source material, requiring verifiable citations, and allowing 'I don't know'
• Quality that drops only late in a long chat as a symptom of a crowded context window, fixed by summarizing, restarting, or persisting key info in a Project
• Overkill cost/latency or output that is too shallow as a symptom of model mismatch, fixed by right-sizing the model to the task
• A missed part of a complex ask as a symptom of missing decomposition, fixed by breaking the task into ordered, checkable steps

Skills in

• Naming the specific symptom an underperforming output shows before choosing a fix
• Matching each symptom to its most likely cause using a structured mental checklist rather than guessing
• Applying the targeted fix for the diagnosed cause instead of regenerating the same prompt repeatedly
• Distinguishing a knowledge gap, a prompt gap, and a context problem so the response fits the actual failure

Knowledge of

• Troubleshooting as iterative and evidence-driven rather than a single corrective action
• Reading the actual result against the intended outcome to name the specific gap
• The one-variable-at-a-time principle: changing the instruction, an example, the context provided, the task breakdown, or the model, one at a time
• Re-running and comparing after a change to confirm it helped before keeping it
• Folding recurring lessons into standard prompts and Project instructions over time

Skills in

• Naming the specific gap between an actual output and the intended outcome instead of reacting generically
• Changing exactly one variable per iteration so the effect of that change is diagnosable
• Re-running after a change and comparing results to decide whether to keep or revert it
• Recognizing recurring patterns across tasks and standardizing the fix into a reusable prompt or Project instruction

Knowledge of

• Right-sizing the model per step -- a fast model for bulk simple work, a stronger one only where reasoning is hard
• Persisting reusable context in a Project so instructions and knowledge aren't rebuilt every time
• Standardizing proven prompts as templates or system-level instructions so quality doesn't depend on re-inventing the prompt
• Keeping human review where it counts and streamlining where it doesn't, so verification effort matches the stakes
• Decomposing long tasks into checkable steps so errors are caught early rather than compounding
• The cost/speed/quality balance as the frame for judging whether a change is a genuine optimization

Skills in

• Assigning bulk, simple, well-defined steps to a fast/cheap model and reserving a stronger model for genuinely hard reasoning steps
• Persisting instructions and reference knowledge in a Project instead of re-supplying them each session
• Turning a proven one-off prompt into a reusable template or standing instruction
• Calibrating how much human review a step gets to the stakes of that step, rather than uniformly maximizing or removing it
• Breaking a long task into ordered, checkable steps to catch errors early
• Evaluating a proposed workflow change against cost, speed, and quality together rather than optimizing one lever in isolation