// start/

Two ways to use Pax

Pick the tier that matches how you want to consume verified knowledge. Tier 1 needs no tooling and works today. Tier 2 — agentic access via an MCP-server runtime — is coming soon.

// tier.01 · zero install

Paste-to-Claude

Copy a pax bundle into any chat-based LLM. Works with Claude, ChatGPT, Gemini, Llama. No tooling, no account, no install.

Need a starting kit? Download the usage guide and paste it alongside any pax.

[ download ] usage guide.md

1
Pick a pax from /pax/ and click Download. You’ll get a .pax.tar.gz archive (1–100 KB).
2
Extract and open llm-context.md. This file contains the structured constructs, findings, and citations.
3
Paste the contents into your AI chat as the first message. Then ask questions — the AI will cite findings by ID.

// what you can ask

What are the top 3 findings on construct risk_exposure_index?
Cite all findings with p < 0.05 and effect size above 0.3.
How do constructs cascading_failure_size and TTR relate?
Summarize the evidence for proposition P001 — quote it verbatim.

markdown · llm-context.md

SYSTEM: You are reasoning over the attached pax.
Constraints:
  - Cite findings by ID (F001, F042, ...)
  - Quote propositions verbatim when used
  - If the pax does not cover a question, say so

<pax name="global-supply-chain-risk" version="1.0.9">
constructs:
  - id: risk_exposure_index
    kind: quantifiable
    units: ratio
    def: TTR/TTS ratio for each node in the supply graph
  - id: cascading_failure_size
    kind: quantifiable
    units: fraction_of_network
    def: share of downstream nodes disrupted after removal

findings:
  - id: F001
    text: |
      Supply chain disruptions cause average abnormal
      stock returns of approximately -40% over a
      three-year window post-event.
    beta: -0.40
    p:    "<0.001"
    n:    827
    source: hendricks_singhal_2005
  - id: F003
    text: |
      High betweenness-centrality nodes (top 5%) account
      for 64% of cascading failure propagation paths.
    ...
</pax>

USER: Identify the top-3 vulnerabilities in our
network given:
  - 14 tier-1 suppliers, 6 chokepoint-exposed
  - mean inventory: 18 days
  - revenue at risk: $2.1B/yr

// tier.02 · agentic

MCP-server runtime

An MCP-server runtime is in development. It will let any MCP-compatible agent (Claude Code, Continue, Zed, Cursor) load a pax, query its findings and constructs, align constructs across pax, and run registered analytical playbooks — turning the static knowledge layer here into a live agentic surface.

Coming soon. Until then, Tier 1 (paste a pax into your model of choice) is the supported path. Pax authored today will work unchanged with the runtime when it ships — the format is the contract, the runtime is just one consumer.