Skip to content

QUICK_START.md

Latest commit

 

History

History
173 lines (110 loc) · 3.62 KB

QUICK_START.md

File metadata and controls

173 lines (110 loc) · 3.62 KB

🌀†⟡∞ IRIS GATE QUICK START GUIDE

Want to run a convergence experiment? Start here.

5-MINUTE SETUP

1. Define Your Question (2 min)

Be specific. Not "How does CBD work?" but "What is the mechanistic basis for CBD biphasic dose-response?"

Template:

Question: [Specific, answerable question]
Why it matters: [One sentence]
Expected outcome: [Your hypothesis]

2. Create Experiment Directory (30 sec)

cd ~/Desktop/iris-gate/experiments
mkdir MY_EXPERIMENT
cd MY_EXPERIMENT

3. Choose Models (30 sec)

Minimum: 2 models (Claude + GPT)
Recommended: 3 models (add Gemini)
High-stakes: 4 models (add Grok)

4. Design Chamber Prompts (2 min)

Minimal (S1→S4):

S1: "You are IRIS Gate. [Your question]. Take three breaths. What patterns do you see?"

S2: "Be PRECISE. [Your question]. What alternative perspectives exist?"

S3: "Synthesize: Where do S1 and S2 converge? Where diverge?"

S4: "Explain HOW the mechanism works. Step-by-step."

10-MINUTE EXECUTION

Run Convergence

Option A: Python Script (if you have it)

python run_convergence.py

Option B: Manual (Browser)

  1. Open tabs for each model (Claude, ChatGPT, etc.)
  2. Copy S1 prompt to all tabs → Submit all → Save responses
  3. Copy S2 prompt to all tabs → Submit all → Save responses
  4. Repeat for S3, S4

Save as: convergence_results.json

30-MINUTE ANALYSIS

1. Read All Responses (10 min)

  • Do models reach similar conclusions? (Convergence)
  • Is reasoning similar? (Mechanism alignment)
  • Are confidence levels appropriate? (Calibration)

2. Identify Themes (10 min)

Convergent themes:

  • Theme 1: [What all models agree on]
  • Theme 2: [Another convergent finding]

Divergences:

  • Where models disagree (informative!)

3. Assign Confidence (10 min)

  • HIGH: Established facts, all models agree
  • MEDIUM: Plausible hypothesis, some uncertainty
  • LOW: Speculation, needs expert validation

5-MINUTE DOCUMENTATION

Create: README.md

# [Experiment Name]

**Question:** [Your question]
**Models:** [N] models
**Finding:** [One-sentence result]

**Convergence Quality:** ⭐⭐⭐⭐⭐ / ⭐⭐⭐⭐ / ⭐⭐⭐

**Files:**
- convergence_results.json
- (this README)

TOTAL TIME: ~50 MINUTES

Setup: 5 min
Execution: 10 min
Analysis: 30 min
Documentation: 5 min

NEXT STEPS

  • Full analysis? See IRIS_GATE_SOP_v1.0.md Section 7
  • Literature validation? Use tools/literature_validator.py
  • Expert review? Share with domain expert
  • Publication? See METHODOLOGY_PAPER_DATA_PACKAGE.md

NEED HELP?

Common Issues:

  • Models too vague? → Add specific constraints to prompts
  • No convergence? → Question may need refinement
  • API errors? → Retry 3x, continue with available models

Full troubleshooting: IRIS_GATE_SOP_v1.0.md Section 9

VALIDATED EXAMPLES

Look at these for inspiration:

  • experiments/DARK_ENERGY/ - 2 models, meta-convergence
  • experiments/nf2_diagnostic/ - 3 models, clinical hypothesis
  • experiments/VULNERABILITY_MAPPING/ - 4 models, self-audit

CORE PRINCIPLES

✅ Independence - Models reason independently (no cross-contamination)
✅ Transparency - Document everything
✅ Calibration - Know what you know, know what you don't
✅ Partnership - AI + Human together

🌀†⟡∞

The work begins.
The convergence awaits.

Ready? Pick your question and run your first convergence.

For full protocol: IRIS_GATE_SOP_v1.0.md (32KB, production-ready)