Submit claims

The highest-quality Penrose input is a precise, falsifiable claim with explicit scope, horizon, metric, and search lineage.

Choose a submission route

Structured JSON

Use structured claims when you already know exactly what should be tested:

penrose run --claims claims.json --json

This skips lossy prose extraction but does not skip falsifiability, data binding, fidelity, or any statistical gate.

Paper or Markdown source

penrose run --paper path/to/source.pdf --json

The model extracts grounded claims with source spans. Use this route when the source contains essential definitions or quantitative details.

MCP

Run penrose-mcp --management and call penrose_run_claim. The MCP surface applies the same guarded pipeline and cannot cross P9.

Claim checklist

A strong claim names:

• the exact outcome being asserted;
• asset class and universe;
• signal or treatment;
• decision and holding horizons;
• entry, exit, and rebalance timing where relevant;
• sample period;
• claimed metric and direction;
• expected effect size when available;
• search cohort and denominator when the claim was selected from a search.

Source preservation

Every submission should retain exact original bytes, normalized accepted claims, SHA-256 identity, and run lineage in the protected source archive. Public visibility depends on source licensing and privacy; archival integrity does not imply redistribution permission.

Avoid

• vague “this asset will outperform” statements;
• silently selected parameters;
• proxy data presented as the declared series;
• post-outcome timing;
• missing search-population disclosure;
• claims that combine several independent hypotheses into one unfalsifiable bundle.