Diagrams — Module 0.3: The Rubric

xychart-beta
    title "Where the model's context actually goes (typical production session)"
    x-axis ["Tool outputs", "History + rest", "Tool defs", "System prompt", "Model logic"]
    y-axis "Share of context (%)" 0 --> 70
    bar [67.6, 18.3, 10.7, 3.4, 1.6]

flowchart TD
    MODEL((Model<br/>~1.6%))
    MODEL --> M1[1 · Execution Loop]
    MODEL --> M2[2 · Tool Design]
    MODEL --> M3[3 · Context Mgmt]
    MODEL --> M4[4 · Memory]
    MODEL --> M5[5 · Sandboxing]
    MODEL --> M6[6 · Permission]
    MODEL --> M7[7 · Error Handling]
    MODEL --> M8[8 · State & Checkpoint]
    MODEL --> M9[9 · Prompt Assembly]
    MODEL --> M10[10 · Subagent Orch.]
    MODEL --> M11[11 · Verification]
    MODEL --> M12[12 · Observability]

    style MODEL fill:#2d2d00,stroke:#8a8a00,color:#d0d000
    style M1 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style M2 fill:#152d15,stroke:#208a20,color:#a0d0a0
    style M3 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style M4 fill:#152d15,stroke:#208a20,color:#a0d0a0
    style M5 fill:#2d1515,stroke:#8a2020,color:#d0a0a0
    style M6 fill:#2d1515,stroke:#8a2020,color:#d0a0a0
    style M7 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style M8 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style M9 fill:#152d15,stroke:#208a20,color:#a0d0a0
    style M10 fill:#152d15,stroke:#208a20,color:#a0d0a0
    style M11 fill:#2d1515,stroke:#8a2020,color:#d0a0a0
    style M12 fill:#2d1515,stroke:#8a2020,color:#d0a0a0

flowchart LR
    P1[Phase 1<br/>First Contact<br/>README · cloc · structure · system prompt]
    P2[Phase 2<br/>Architecture Map<br/>loop diagram · tools · trace · stop conditions]
    P3[Phase 3<br/>Decision Audit<br/>12 modules · what & why · 3 agree · 3 disagree]
    P4[Phase 4<br/>Security Audit<br/>credential flow · exec paths · injection · blast radius]
    P5[Phase 5<br/>Benchmark<br/>published scores · token profile · cold start]
    P6[Phase 6<br/>Score & Synthesize<br/>/60 · Verdict · MLSecOps · 3 best · 3 fix]

    P1 --> P2 --> P3 --> P4 --> P5 --> P6

    style P1 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style P2 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style P3 fill:#152d15,stroke:#208a20,color:#a0d0a0
    style P4 fill:#2d1515,stroke:#8a2020,color:#d0a0a0
    style P5 fill:#15152d,stroke:#20208a,color:#a0a0d0
    style P6 fill:#2d2d00,stroke:#8a8a00,color:#d0d000

flowchart TB
    subgraph WRONG["Feature checklist (anti-pattern)"]
        direction TB
        W1["'Does it have a sandbox?'"]
        W2["'Does it have compaction?'"]
        W3["'Does it have multi-session?'"]
        W4["→ Output: a ranking. Tells you nothing a README couldn't."]
    end

    subgraph RIGHT["Decision audit (the rubric)"]
        direction TB
        R1["'What pattern did they choose?'"]
        R2["'What did they GIVE UP?'"]
        R3["'Where in the code is the evidence?'"]
        R4["→ Output: understanding. Tells you why and at what cost."]
    end

    WRONG -.->|the tell: if your output could come from a README, you didn't audit a decision| RIGHT

    style WRONG fill:#2d1515,stroke:#8a2020,color:#d0a0a0
    style RIGHT fill:#152d15,stroke:#208a20,color:#a0d0a0