Cut the lines in your prompt that do nothing

Shorten this prompt. ***Keep the behavior identical.***

1. **First find the instructions that are doing nothing:**
   - **Adjectives with no check behind them** — "professional", "high-quality", "carefully"
   - **Rules the model would follow anyway**
   - **The same rule said twice in different words**
   - **A role assignment that does not change the answer**
   - **Examples longer than the rule they illustrate**
2. **Then find what is load-bearing** — remove it and the output changes. ***Mark these and do not touch them***
3. Merge overlapping rules into one sentence
4. Cut, and **report the before/after length**

**Rules:**
- ***Do not shorten by becoming abstract.*** **"Handle errors appropriately" is shorter than the specific rule and worse**
- ***Never remove: instructions about what not to do, how to handle missing information, or output format.*** **These are the ones that look removable and are not** — they only fire on the runs that would otherwise go wrong
- **If a cut carries any risk, mark it rather than making it**