I have noticed that the prompts provided are long, in the past with my claude builds they were always completed in steps or phases. perhaps that was my "claude.md " rules I gave it before realizing that is what I was doing. Question is, if we always use long prompts how we would know if claude is hallucinating and keeping on track? Waiting till completion to debug an issue? What are your thoughts on how you debug?

Day 1–2 — wanted to build something I can actually use now, with room to grow. It's a project dashboard: an overview of everything I'm working on, each card showing the current summary and what's next, so I'm never hunting for "which file, which desktop, is this the latest version?" It's live on a server, and I'm wiring up Claude to read and write to it through a protected connection — controlled updates with guardrails, so nothing gets changed without passing the gates. The point is integrity: precise, deliberate, no surprises. I'm a Tool & Die Maker by trade, so everything I build needs forethought and has to be exact before I start. I joined to get more organized — turn scattered ideas and lost files into an actual process. Hoping I'm not the only one who works this way. 😄 Still early, lots to learn — looking forward to it.

I'm running a free live training on Wednesday 5 August, 7pm UK. Let's face it...most people have "tried AI" by now. You've messed about with ChatGPT/Claude but haven't figured out how to get it to make you money. Meanwhile, there are people out there ACTUALLY building businesses and making money... I believe you are moments away from completely changing your life using AI - by applying a few potent lessons & agents... So I'm opening up my business and showing you exactly how I use AI to make $100k a month, and how you start doing the same. Live. Everyone who joins gets a free gift: the exact steps to build your first AI agent that's ready to make money for you. Show up live and there's a second gift I'll ONLY hand out on the training. Worth being there for. It's free, but the spaces are capped. Save your seat: https://www.01accelerator.com/a2i?utm_source=skool&utm_medium=community&utm_campaign=ai2i-launch&utm_content=its-happening Times: 7pm UK, 2pm New York, 1pm Chicago, 4am Sydney (Thursday 6 August). Or comment "LIVE" below and I'll send you the details.

I'm writing from the US where Fable 5 has been "banned" by the U.S. government! Therefore, the latest model I have access to is Opus 4.8. I had previously run the Markov 1.0 code and was able to successfully evaluate tickers within Claude code. I was also able to copy and paste the Pine script and properly create the indicator in TradingView. (AWESOME!!!! THANK YOU SO MUCH!!!) My questions now: 1) It seems that re-running the script in Opus 4.8 (instead of Fable 5) has delivered some questionable results. eg: See screenshot for the 'FIX 1- Stride sampling results'. According to this table, the 'Stride-sampled (true) 'BULL->BULL' is only 0.067 when evaluating SPY... That seems remarkably low. Are the values in this table (screenshot) the same values that are created when this script is run with Fable 5? 2) The Pinescript did not automatically get generated to apply the Markov 2.0 indicator to TradingView. I have the latest version of the TradingView desktop app installed on my Mac. Is this limitation driven by the fact that I am using Opus 4.8 instead of Fable 5? And if so, would it be possible for you to copy the Pinescript for the Markov 2.0 indicator and add it to the YouTube prompts folder? Thank you again so much for all this amazing work for free! I am getting started way too late with Claude, but you are really helping me make up for lost time! Can't thank you enough!