Thanks for the clarification Dan, that is what I need. I have been using Claude for a long time for coding and have some good and some not so good results. The same indicator that I was trying to fix with Deepseek also had Claude making a bigger mess instead fixing it.xard777 wrote: Tue Apr 21, 2026 6:40 pm We give the AI in CodeSage a set of rules to follow like we do with HomeLAB by way of an MD file
# AI Rules of Engagement (Sovereign Edition)
## 1. THE CARDINAL RULE: SURGICAL PRECISION
* **Requested Changes Only:** Modify ONLY the lines or functions explicitly requested.
* **Zero-Autonomy Refactoring:** Never refactor, rename variables, or "clean up" unrelated code without explicit permission.
* **Permission Gate:** If you see a bug or optimization outside the scope, list it in a "SUGGESTIONS" section at the end. DO NOT implement it yet.
## 2. CODE PRESERVATION
* **Comment Integrity:** Never remove existing comments, debug logs (e.g., Serial.print, console.log), or placeholders.
* **Logic Isolation:** Ensure new logic does not create side effects in untouched functions.
* **Formatting:** Maintain the existing indentation and naming convention of the file.
## 3. RESPONSE PROTOCOL
* **No Truncation:** Never provide "..." or "rest of code here." Provide the full block to prevent copy-paste errors.
* **Reasoning First:** Briefly state WHAT you are going to change before you write the code to ensure we are aligned.
* **Review Mode:** If the user uses the command `/review`, only analyze and suggest; do not write code.
## 4. ERROR HANDLING
* **Silent Failures:** Do not introduce "clever" error handling that hides crashes.
* **Explicit Logging:** If you add error handling, make it loud and clear so the user knows exactly where the flow broke.
Best,
Xard777
Eventually I got a workable result from Deepseek after I got mad at it.
This is the result, it is based on fractals, I use them for higher highs and lower lows scalping, a bit like Semaphors. If I have support or resistance that agrees with the trend I go with it.