Evidence and Traceability

This section is the atlas's audit surface. Use it to distinguish a directly captured fact from an interpretation, reproduce the safe parts of the analysis, or move from a prose statement to its supporting artifact and explanatory contract.

Follow a claim

flowchart LR
    accTitle: Evidence and Traceability - Follow a claim
    accDescr: Diagram showing follow a claim in the Evidence and Traceability section.
    Q["Reader question"] --> D["Atlas explanation"]
    D --> C["Version-bound claim ID"]
    C --> E["Static or dynamic evidence"]
    C --> R["Reconstructed contract"]
    E --> L["Known limit or falsification path"]
    R --> L

Need	Start here	Continue to
Understand how observations were collected	Evidence method and reproduction	Scope and publication boundaries
Browse all established claims	Claim catalog	evidence/claims.ndjson
Find the code model behind a claim	Evidence-to-code index	reconstructed/
Check what remains unknown	Version scope and open gaps	Runtime observations
Review the publication boundary	Legal and ethics	Security research and disclosure

Evidence vocabulary

Observed values were captured from the version-matched artifact, signed metadata, CLI output, or a controlled dynamic boundary.

Derived interpretations connect multiple observations without claiming to recover Anthropic's original module boundaries.

Hypothesis models are explicitly open and include a path for confirmation or rejection.