ADR: Minimal provider-neutral experiment system: document frontier

Source and coverage

• Source: research/framework-engineering/experiments/architecture.md
• Canonical ID: FE-ADR-EXPERIMENT-SYSTEM-001
• Lifecycle status: accepted-for-pilot
• Artifact type: research-document
• Version: 1.0
• Confidence: not declared
• Inclusion rule: eligible lifecycle status: accepted-for-pilot
• Analysis unit: EXPERIMENT-SYSTEM

Phase 1 — knowledge extraction

• Primary objective: Provide a provider-neutral, immutable-run experiment registry and migration path for current and historical FE experiments.
• Primary claims: The narrow experiment infrastructure is executable and preserves provider/run lineage. Verification of package structure is separate from empirical acceptance.
• Supporting evidence: The source's declared references and the co-analyzed records listed in the analysis-unit frontier.
• Methodology: Architecture definition, migration manifests, immutable-run tooling, fixtures, and automated verification.
• Accepted/current findings: The architecture is accepted-for-pilot and the implementation report is complete in its declared engineering scope.
• Rejected or unsupported findings: Successful experiment-package verification establishes scientific validity. Provider labels alone establish independent replication.
• Assumptions: Write-once run directories and provider metadata are sufficient for the current pilot boundary. Migration manifests faithfully represent historical outputs.
• Limitations: Cross-provider independence semantics, adversarial integrity, and high-consequence promotion remain unvalidated.
• Known uncertainties: Failure behavior under corrupted runs; Migration compatibility; Independence semantics; Operational maintenance cost.
• Declared headings inspected: ADR: Minimal provider-neutral experiment system; Decision; Safety and lineage; Alternatives rejected for this pilot; Consequences.

Phase 2 — unresolved unknowns

• Failure behavior under corrupted runs
• Migration compatibility
• Independence semantics
• Operational maintenance cost

Phase 3 — adversarial challenge

The experiment system has stronger immutability controls than the ROS–FE profile execution validator that indexes some of its research outputs.

Evidence that would materially change this assessment includes a completed independent replication, a preregistered negative control, a matched causal comparison, a governance disposition, or a demonstrated failure of the measurement and integrity instruments relevant to this source.

Top five traceable opportunities

The opportunity records contain source sections, challenged assumptions, supporting evidence, dependencies, methodology, outputs, and success criteria. Repeated opportunities are semantic links to one repository-wide record rather than duplicated records.