prototype

/prototype — Build something testable

You are building a proof-of-concept for the current Grainulator sprint. Read CLAUDE.md for sprint context and claims.json for existing research claims.

Process

1. Determine what to prototype: Based on the user's argument and existing research claims. If no argument given, look at the research and suggest the most promising option to test.

2. Build it: Create a working prototype in prototypes/<name>/. This should be:

   • Minimal — just enough to test the hypothesis
   • Runnable — include a README or run script
   • Measurable — produce output that can be evaluated

3. Also generate a demo artifact: Create prototypes/<name>/demo.html — a self-contained HTML page that shows what the prototype does, with screenshots, code snippets, or interactive elements. Non-technical stakeholders should be able to understand the prototype from this page alone.

Claim updates

Every prototype finding becomes a claim with evidence tier tested:

{
  "id": "p001",
  "type": "factual",
  "topic": "<what was tested>",
  "content": "<what we found — be specific with numbers>",
  "source": {
    "origin": "prototype",
    "artifact": "prototypes/<name>/",
    "connector": null
  },
  "evidence": "tested",
  "status": "active",
  "phase_added": "prototype",
  "timestamp": "<ISO timestamp>",
  "conflicts_with": [],
  "resolved_by": null,
  "tags": []
}

Critical: Check if any existing research claims (evidence: "web") are contradicted by prototype results. If so:

• Set conflicts_with on both claims
• The compiler will auto-resolve in favor of tested over web

Update meta.phase to "prototype" in claims.json if this is the first prototype.

Run the compiler

grainulator compile --summary

Report evidence upgrades (research claims superseded by prototype findings).

Version control

Preserve the user’s edits. Commit only when the task explicitly calls for it; never publish automatically.