Project submissions are graded by Claude on five dimensions: tool selection, workflow logic, output quality, documentation, and edge-case handling. The average submission scores around 74%. Distinction (90%+) requires doing something specific on each dimension that most candidates skip.
The five rubric dimensions explained
Tool Selection (20%) — This isn't about listing the most popular tools. It's about explaining *why* you chose each one and how they fit together. A submission that says "I'd use Claude for AI processing and n8n for orchestration" scores lower than one that says "I chose n8n over Zapier because the project requires branching logic with conditional paths, and n8n's native code node handles the data transformation without an extra API call."
Workflow Logic (25%) — This is the highest-weighted dimension. Your workflow needs to be end-to-end — from data input to final output — with no unexplained gaps. Draw a mental map before you write: what triggers the workflow? What happens at each step? Where does data flow? What format does it leave in?
Output Quality (20%) — The grader evaluates whether your outputs (reports, notifications, enriched data) are actually useful in a business context. A Slack message that says "Lead scored: Hot" scores lower than one with a 3-sentence AI-generated summary of *why* the lead is hot and what the sales rep should say first.
Documentation (20%) — Most candidates write a description but skip documentation. Distinction-tier submissions include: the tools used and why, the assumptions made, how a non-technical manager could monitor the workflow, and how to handle the most common failure mode.
Edge Cases (15%) — Pick one realistic edge case and explain it specifically. "What if the API is down?" is too generic. "What if the lead's email domain is a free provider (gmail, hotmail) — how does the enrichment step handle the missing company data?" shows you've actually thought about real-world operation.
The structure that works
After reading hundreds of submissions, the highest-scoring ones follow this structure:
- **Overview** (2–3 sentences): What the workflow does, for whom, and what problem it solves.
- **Step-by-step breakdown**: Each step numbered, with the tool used and the specific action it takes.
- **Tool rationale**: A short paragraph explaining why this stack, not another.
- **Sample output**: Show exactly what a notification, report, or enriched record looks like.
- **Edge case**: One specific scenario + how your workflow handles it.
- **Scale consideration**: One sentence on what changes if volume doubles.
What drops you from Merit to Pass
- Using vague language ("the AI will process the data") instead of specific actions ("Claude will extract the company name, industry, and headcount from the LinkedIn URL using a structured JSON prompt")
- Skipping the sample output entirely
- Listing tools without connecting them ("I'd use Make, Claude, and Airtable")
- No edge case section at all
The mindset shift
Distinction candidates don't write a description of what they would build. They write as if they are handing the workflow to a colleague to implement on Monday. That level of specificity is what separates 91% from 74%.