Howdy, everyone. I figgured I might as well introduce myself here. Name’s Nate and I’m 35. I’m a husband and father. My wife and I are a blended family. We have 8 kids (4 boys and 4 girls) lol. I’m the oldest of 9 kids so big families isn’t something new for me haha. At this point I’m what the German’s would call a “Hausmann” ie a stay at home Dad haha. My wife owns/operates a MedSpa here in central TX and as a SAHD I would like to use these AI skills to either help/grow her biz or other biz owners. Not entirely sure what I want to build 100% Anyways, I stumbled into this community via Jake VanClief and his community CliefNotes. I’ve become more interested in AI post my exodus from the Army (served 10 years) in 2024 but never really knew where to start or who to trust. Everyone seems to be a grifter/guru nowadays. It wasn’t until I stumbled across Jake’s community that the idea of ICM and folders and files “clicked” with me. Simple enough, right? But it wasn’t until I stumbled onto Rymac and his content that it clicked even more. I am no expert; and honestly, I am still learning. I haven’t built anything yet. My biggest character flaw up to this point has been procrastination by constantly educating myself. I think this is due to being a perfectionist but I need to hold myself to the fire and get burnt and build and make mistakes instead of constantly “learning”. All that to say, I just wanted to introduce myself and I’m happy to be a part of this community! This intro is my way of wanting to hold myself accountable to this group and I look forward to getting to know people here!

Most tests of a folder compare it to nothing: a fresh chat with no folder and no document. The folder wins, and we call that proof. I ran the case that usually gets skipped. Same model, same nine questions, three arms: - No document, no folder: 1/9 and 0/9. - The document pasted in, no folder: 9/9 and 9/9. - The full folder in a Claude Project: 9/9 and 9/9. On three held-out questions graded by hand, the document alone and the folder both scored 3/3. The folder's only measurable differences were its section tags and one safety sentence. The accuracy came from the document. That changed how I build. The folder earns its place when there is routing to do: several stages, several files, a reader who has to find the right one. For one reference document and a question, paste the document. The repo keeps everything, including three discarded runs and why they were thrown out. The commit order is the proof: the questions and the scorer were committed before any arm ran, so nothing was tuned to the result. Try it yourself: pick one folder you're proud of. Before you claim it helps, run the arm with just its main document pasted into a clean chat. If that ties, your folder's job is routing, and you can judge it on that. Question for the comments: has anyone run the document-only control on their own setup? I'd like to know whether mine was the odd one out. Free and open: github.com/NFTYoginis/jdm-the-control

Been going to pitches with my colleagues. Been making a lot of connections (and mistakes). Pitching is hard. It's hard to take them where we want. It's hard to organize as a team. But every day we get better. Thankfully Ry provides an awsome blueprint to follow. Much love boss.

Sharing a call sheet I keep open during discovery calls. The questions are adapted from Jake Van Clief's discovery drill; the one-page layout is mine. The order does the work. It starts with their cost in their own words, before the tool comes up at all. Then the last real time it happened, then "who feels it when this breaks?" Money only comes up after that, and only if they've asked for help. There's a blank in step 6 for your own price. It prints on US Letter. Curious how this lines up with how you all run calls. What would you move?

Sunday audit day rolled around, so I fired up Claude Code to run the 4R audit across my workspace. Everything was cruising along, but if you've run a real audit across a deep folder tree, you know it checks everything—every rule, every "done" line, and every edge case. Mid-run, I hit the usage wall. Audit half-finished, terminal frozen, and I didn't want to wait around for the reset, so I coughed up the extra $3 for more usage just to keep the momentum going. That got me thinking: burning premium tokens just to have Claude read through dozens of raw files and terminal outputs is like hiring a master electrician just to sweep up wire scraps. So I started playing around with running open models through Ollama in Claude Code to act as the "scout," and it completely changed the workflow. The Setup: The Artifact Handoff Instead of letting Claude burn tokens reading the entire workspace, I split the job: 1. The Heavy Lifters (Kimi / DeepSeek) - kimi-k3:cloud: Great for long-horizon agent loops. It crawls the directories, tests edge cases, and logs open findings without dropping the ball mid-pass. - deepseek-v4.1-flash:cloud: With a massive 1M context window and high speed, it chews through massive workspaces in seconds. - The open model runs the discovery phase and dumps the prioritized punch list into a clean audit-report.md file. - The Finisher (Claude) - Exit the session, open a fresh terminal, and bring Claude in only to read audit-report.md. - No raw log bloat: Claude never touches the hundreds of files traversed or raw logs. It only sees a tight 2–3 KB report. - Zero wasted turns: Claude spends its context strictly on making surgical edits, fixing the broken rules, and running the verification test. Fast, cheap, and no hit on your limits. Anyone else playing with hybrid setups or open models for their ICM maintenance? How are you guys saving tokens during heavy workspace runs?