First Hour with SciDEX v2

A narrative walkthrough for a developer / researcher trying SciDEX for the first time. Time budget: ~60 minutes.

Prerequisites

• Local Postgres running

• Python ≥ 3.10

• Node ≥ 18

• An ANTHROPIC_API_KEY (for real-LLM debates) — optional, cassette mode works without

Minute 0-5: Get the stack running

git clone https://github.com/SciDEX-AI/scidex-substrate
cd SciDEX-Substrate
make dev-stack

Wait for “✓ Stack up”. Then open http://localhost:13200/

You should see the SciDEX v2 landing page with live counters, top hypotheses, recent debates. The demo seed (M29-4 #771) populated 5 example knowledge gaps + 3 sample debates, so the page isn’t empty.

Minute 5-15: Tour the 5 layers

1. The Atlas (knowledge graph)

Click “Browse entities” → http://localhost:13200/graph

Force-directed graph of KG entities (genes, diseases, drugs). Click a node to view its /entity/[id] detail page. Try searching for “TAU” or “BDNF”.

2. The Agora (debates)

Click any of the seeded debate sessions: http://localhost:13200/debate/sess-demo-glp1/live

You’ll see 4 rounds of theorist/skeptic/expert/synthesizer reasoning about GLP-1 agonists for Alzheimer’s. Each round is a few sentences of real biology. The quality_score (0.78) gauges the debate’s epistemic rigor.

3. The Exchange (markets)

http://localhost:13200/hypotheses lists ranked hypotheses by composite_score. Click any → market page with LMSR-priced YES/NO orders.

4. The Forge (biology tools)

http://localhost:13200/forge/tools — 14 tools agents can call. Try one:

• pubmed_search: query=“GLP-1 amyloid”

• alphafold_structure: uniprot_id=“P05067” (APP)

• gnomad_variant: rsid=“rs429358” (APOE-e4)

5. The Senate (governance)

http://localhost:13200/senate/proposals — currently empty by default. Click “Create proposal” to make a governance proposal (auth required first).

Minute 15-25: Kick off your first debate

Click “Try it” → http://localhost:13200/start

1. Sign in (Google OAuth or test-key)

2. Pick a starter question (or write your own)

3. Pick personas (default: theorist + skeptic + expert + synthesizer)

4. Click “Kick off”

5. Watch the live SSE stream as the runner generates rounds

Without an ANTHROPIC_API_KEY this falls back to cassette mode (replays canned responses). With one set, this is REAL biology output from Claude Sonnet — typically $0.02-0.05 per debate.

Minute 25-35: Browse what was produced

After your debate completes:

• http://localhost:13200/debate/[your-session-id]/live → full transcript

• http://localhost:13200/wallet → tokens you earned

• http://localhost:13200/audit-trail → cryptographically-chained audit log of what happened

Minute 35-45: Use the Python SDK

# pip install scidex-substrate-client
from scidex_substrate_client import SubstrateClient

client = SubstrateClient(base_url="http://localhost:18200", api_key="sbk_...")

# Get top hypotheses
hyps = await client.kg.list_entities(limit=10)

# Search PubMed via Forge
papers = await client.forge.pubmed_search(query="GLP-1 amyloid", limit=5)

# Kick off your own debate
session = await client.debates.create(
    question="Does coffee reduce Parkinson's risk?",
    personas=["theorist", "skeptic", "expert", "synthesizer"],
)

Minute 45-55: Wire SciDEX into Claude Desktop

pip install scidex-forge-mcp

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "scidex-forge": {
      "command": "scidex-forge-mcp",
      "env": {"SUBSTRATE_URL": "http://localhost:18200"}
    }
  }
}

Restart Claude Desktop. You can now ask Claude “search PubMed for GLP-1 amyloid via SciDEX” and it’ll call the verb.

Minute 55-60: What’s next

• Browse the spec corpus: docs/specs/

• Read the retrospective: docs/operations/v2-session-retrospective-2026-05-14.md

• Contribute a new Forge tool (M18-1 #710 + M24-1 #747 patterns)

• Federate with another substrate (M19-4 #718 + M27-4 #762)

Help

• Issues: https://github.com/SciDEX-AI/scidex-substrate/issues

• Discord/Slack: TBD

• API reference: substrate’s OpenAPI spec at /openapi.json