📑My Read on the weekly comps: what's crucial, what changes, and where the context.md fits

⭐ I am waiting on some feedback on this, I could be off with my read, I am willing to be wrong in public so we can all be right when we are building in private!

The rules feel like they contradict each other week to week. They don't.

📍 WEEK 6 AT A GLANCE — THE RESEARCHER

https://www.skool.com/cliefnotes/weekly-comp-6-the-researcher?p=fa674d82

🔬 Build: a single folder-based AI researcher for ONE specific domain (you pick). Five files, single folder — so no handoff.md and no context.md this week.

🚫 The angle: a researcher is NOT a summarizer. It asks what's missing, questions the framing, weighs sources by credibility, and asks clarifying questions before it produces. That lives in rules.md

⚖️ New wrinkle this week: judging adds "does it weigh sources / flag uncertainty?" — bake that into rules.md.

📨 Submit: public GitHub repo + 2-3 sentences on what your researcher covers and what work it's best for.

🎟️ Prize: a free Lyceum seat (pick cohort — Technical / Business / Creator).

📅 Due: Sunday, May 31, 12:00 PM EST (shifted for Memorial Day). Winner Mon June 1. Premium + VIP only.

👇—My Take Below!

How the rules have read to me since my first entry and getting a bit of clarity🔮

💡There's a spine that never moves and a set of dials that change every Sunday. Once you can tell those apart, the whole thing gets a lot less stressful.

⚠️This is the post I wish I'd had before I submitted, hopefully it brings clarity to an awesome community opportunity! These competitions are like mini boot camps, the tools you take away compound.

🧩 One part to focus on is the use of: context.md

📑The lessons teach ICM with a context.md in basically every directory. Then the comp spec lists five files identity.md, rules.md, example.md, reference/, README.md, and context.md is nowhere.

🤔 So, you sit there wondering if the spec is wrong, or if you're supposed to "know better" and add it.

⚠️ Here's how I am seeing it fits. context.md is a routing file. Its whole job is navigation: in a multi-directory system, it tells the AI what lives in a given directory and where to find it. It earns its place when there's enough material that the AI would otherwise waste effort hunting.

💡The single-folder comps: Specialist (3), Coach (5), Researcher (6) have nothing to route.

One folder, five files. The README already orients both the human and the AI. A context.md there is a signpost pointing at a room you're already standing in.

And notice what happens the one-week routing actually matters. Week 4, the Agency, is multi-folder — and it asks for handoff.md, not context.md. That's not an accident, and it's the thing that finally made it click for me.

🔄 & 🔀 They're different jobs:

🔄 handoff.md is peer-to-peer. "Here's what this folder receives, and here's what it passes to the next one." Workflow, not navigation.

🔀 context.md is router-to-directory. "Here's what's in here and where to look."

In Week 4, the 00_orchestrator/ folder is the router. That role lives in its identity and rules. So even the routing-heavy comp doesn't use context.md the way the lessons do.

🤷‍♂️Bottom line the way I see it right now, build exactly the files the comp names. If you want to honor the ICM instinct, drop a lean context.md into your reference/ folder only if it grows subfolders worth orienting. That signals you understand why the file exists instead of cargo-cutting it from the lessons.

But it's not what wins or loses you the comp.

🦴 The spine (this doesn't move, Week 3 onward)

Everything from the Specialist comp on is the same skeleton:

📁 The five-file folder. identify.md, rules.md, example.md, reference/, README.com. Named exactly like that. (Week 4 is the only exception multi-folder, and it swaps handoff.md in.)

🐙 A public GitHub repo. Required from Week 4 on. Week 3 allowed Drive but told you GitHub was coming. The repo is the deliverable.

📝 A short writeup alongside the link.

🎯 The Bar: a stranger uses your folder cold and gets value with no explanation from you. If you have to hop on a call to explain it, it's not done.

⚖️ The four judging criteria, which barely change: (1) does it actually work / do its job, (2) is the domain specific enough, (3) is the methodology clean each file does one job, (4) is the README good enough for a stranger.

🔥 The one principal underneath all of it: the thing has to act, not just inform. A coach isn't a knowledge base. A researcher isn't a summarizer. That distinction lives in your rules.md, and it's where most submissions — mine included — leave points on the table.

If you nail the spine, you're already ahead of most of the field.

🎛️ The dials (these change every week — check them every time)

🧑‍💼 The role. Specialist → Agency → Coach → Researcher. The structure is constant; the job isn't.

🗂️ Single vs. multi-folder. Only Week 4 has gone multi-folder so far. Don't assume.

🚫 The anti-pattern callout. Each week names the failure mode to avoid (knowledge base, summarizer, etc.). Read it — it's a direct hint at what the judges are scoring.

✍️ The writeup format. This one bite people. Week 2 and Week 4 want ~100 words with specific fields. Weeks 3, 5, 6 want just 2-3 sentences. Count before you submit.

📅 The deadline day. Sometimes Saturday, sometimes Sunday — always 12:00 PM EST. Week 6 shifted for Memorial Day. Don't assume last week's day.

🏆 Judges and prize. Roster and prize (cash vs. a Lyceum seat) change. Doesn't affect your build but know what you're playing for.

📋 pre-submit checklist: Run this before you drop your link. It's the list I'll be using next time.

✅ I built exactly the files this week names — no extras bolted on, nothing required left out

✅ Repo is public and the link actually opens in an incognito window

✅ A stranger could use it cold from the README alone — no call, no DM

✅ Each file does one job (identity ≠ rules ≠ example — no overlap)

✅ My rules.md makes the thing act, not inform — it handles vagueness, pushback, and "I'm stuck," not just Q&A

✅ The domain is narrow ("salary negotiation for Series A tech," not "career coach")

✅ example.md shows 2-3 real interactions — show, don't tell

✅ The writeup matches this week's format (right word count, right fields)

✅ I can name a real design decision I made and why

✅ Submitted before the deadline — I checked the day (Sat vs. Sun) and its noon EST

💬 Last thing

What I have found in my research of our competitions is this: if you didn't place, it was never because your heart wasn't in it, the people who are entering these competitions clearly care and clearly know how to work with AI and ICM. Working with AI is a learning process, and we are all learning.

🤓In my experience (1 round 💪😅) I missed more than one of these dials while pouring everything into the build. That's a fixable problem, and honestly an easier one than it feels like in the moment. 💪

🛠️This list may be missing something, or be off in some way, if you see something say something, lists don't have feelings they have purpose! if it is, we should help it grow or fix it.

These competitions are a lot of fun, and they are getting really good, let's make the tools that help them get stronger!

Read the spec twice. Build the spine. Check the dials. Then go make the thing your heart is so passionate about! 💪❤️‍🔥🏆

@David Vogel

@Jake Van Clief

@Matthew Creamer

👍👎

9 comments