Submission Workflow

%%{init: {"themeVariables": {"fontSize": "18px"}, "flowchart": {"nodeSpacing": 24, "rankSpacing": 32, "padding": 10}}}%%
flowchart TB
    subgraph setup ["Get set up"]
        direction LR
        R[1 Get access] --> I[2 Install] --> P[3 Plan curve]
    end
    subgraph measure ["Measure and describe"]
        direction LR
        X[4 Run points] --> D[5 Author disclosures]
    end
    subgraph submit ["Register, check and submit"]
        direction LR
        G[6 Register runs] --> V[7 Validate] --> S[8 Submit] --> A[9 After you submit]
        V -.->|"errors: fix and register again"| G
    end
    setup --> measure --> submit

Step	What it produces
Before you begin	A decision to proceed, or to stop before it costs anything
1. Get access and an API key	A PRISM API key in mlc_… format
2. Install the tools	Reference client, submission CLI, checker
3. Plan your Pareto curve	The concurrency levels you'll run, and your plan for the Offline point
4. Run the measurement points	One run folder per point, with accuracy results at the required points
5. Author the disclosure files	system_desc.json and point.yaml per point, system_power.json per system
6. Register your runs	One run ID per run folder, uploaded to MLCommons
7. Validate locally	A clean checker report
8. Submit	An uploaded bundle and a review PR
9. After you submit	A finalized, published result