Classifying Weak Output Before Fixing It
CoreDiagnose underperforming prompts and outputs · Difficulty 1/5
Explanation
Diagnose, Then Fix
When Claude's output is weak, the productive first move is not to try again -- it is to classify *what kind* of problem the output shows, then apply the fix that matches that diagnosis. Troubleshooting draws on every earlier domain in the exam: a bad output is usually a prompting, context, model-fit, or verification problem in disguise, not a mysterious failure.
Symptom-to-Fix Table
| Symptom | Likely cause | Targeted fix |
|---|---|---|
| Generic, vague, off-target | Under-specified prompt | Add audience, format, constraints, and an example |
| Wrong format / structure | Format not stated | Specify the exact output shape; provide a filled example |
| Confidently wrong facts / fake citations | Hallucination | Ground in source, ask for verifiable citations, allow "I don't know" |
| Quality drops late in a long chat | Crowded context window | Summarize, restart, or persist key info in a Project |
| Overkill cost/latency or too shallow | Model mismatch | Right-size the model -- Haiku/Sonnet/Opus to the task |
| Missed part of a complex ask | No decomposition | Break the task into ordered, checkable steps |
Each row pairs a distinct, recognizable symptom with a specific fix. Recognizing which kind of "unexpected" an output is -- a knowledge gap, a prompt gap, or a context problem -- lets a user respond with the right remedy instead of guessing.
Common exam traps
- Regenerating repeatedly without changing anything. Random re-rolls are not troubleshooting; the same under-specified prompt tends to produce the same class of weak output on every retry.
- Reaching for a bigger, more expensive model when the real problem is a vague prompt or a crowded context window. Model mismatch is only one row in the table -- upgrading the model does not fix a missing format spec, a hallucination, or context rot.
Key Takeaways
- Classify the specific symptom of weak output before choosing a fix -- do not just retry
- Generic/off-target output means under-specification; the fix is audience, format, constraints, and an example
- Confidently wrong facts or fake citations are hallucination; the fix is grounding and verifiable citations, not a shorter prompt or a 'more confident' instruction
- Late-conversation quality drops point to a crowded context window, fixed by summarizing, restarting, or persisting context in a Project
- Switching to a bigger/pricier model is a common wrong first move when the actual cause is a prompt or context problem
Glossary Terms
The practice of directing requests to different Claude model tiers based on assessed complexity and requirements. A common pattern uses a fast, cheap model (Haiku) to classify task complexity, then routes to Sonnet or Opus accordingly.
The process of breaking a complex task into smaller, independently executable subtasks that can be assigned to specialized subagents or processed sequentially. Good decomposition creates subtasks with clear boundaries, independent execution, and verifiable outputs.
Related Concepts
Two High-Stakes Symptoms: Hallucination and Context Crowding
Hallucination (confidently wrong facts or invented citations) is fixed by grounding in source and requiring verifiable citations, not by tone or temperature changes
The One-Variable-at-a-Time Feedback Loop
The feedback loop is: read the gap, change one variable, re-run and compare