Evidence Toolkit · Development

Assemble review context without collapsing system responsibilities.

The toolkit is a local-first, sequence-free contract for linking design, selection, and validation-evidence references. Its v0.1 implementation validates synthetic packages; live adapters are a later milestone.

A contract between owners, not one shared truth store.

Each system remains authoritative only for its own record. The toolkit verifies that identifiers and artifact digests agree; it does not copy authority from one system into another.

FactorForge

Owns: Computational design evidence and automated review decisions

Contributes: A versioned, sequence-free Design Evidence record

AgentOS

Owns: Workflow state, selection rationale, and human-checkpoint references

Contributes: A Selection Record; human approval is never inferred

ValidationHub

Owns: Experimental-evidence identity and review state

Contributes: An Evidence Reference Index, not embedded permissioned data

Evidence Toolkit

Owns: Package structure, checksum verification, cross-links, and rendering

Contributes: A deterministic review package and explicit validation result

Included in contract v0.1

  • Versioned JSON contracts for three source-system records
  • SHA-256 integrity and cross-system identity checks
  • Rejection of raw-sequence fields and sequence-like values
  • A standalone HTML review view using synthetic fixtures

Intentionally outside v0.1

  • No raw DNA or protein sequences, wet-lab files, or partner records
  • No scientific sufficiency judgment or human approval generation
  • No regulatory submission assembly or regulatory-acceptance claim
  • No claim of GLP, GMP, or 21 CFR Part 11 compliance

Current connection status

FactorForge's report contract is the first real source format. AgentOS and ValidationHub are represented by explicit sequence-free contracts and synthetic fixtures. Direct API or MCP adapters and controlled artifact storage have not been implemented in this milestone.