Reproducibility

Artifacts are part of the system.

A cognitive runtime is only useful if its internal state can be preserved. ADAM experiments produce traces, summaries, audits, manifests, and versioned artifacts so later readers can inspect what actually happened.

Runs

Raw trajectories

Runtime records preserve turn-level state, activations, goals, episodes, and ecological diagnostics.

Audits

Human-readable review

Audit files summarize what passed, what failed, and which claims remain out of scope.

Lineage

Versioned evolution

ADAM 4.0, 5.0, and 6.0 preserve different layers of the architecture rather than overwriting one another.

Artifact Policy

The raw record stays separate from the story built on top of it.

Public releases will separate manuscript claims, human-readable summaries, raw runtime artifacts, and code state. This is especially important for concept ecology, where negative results and sparse trajectories can be as informative as successful cascades.

Raw Trace archives remain preserved

Raw outputs allow later verification, alternative analyses, and challenge checks.

Derived Summaries are treated as derived artifacts

Processed JSON, tables, and figures should be tied back to the records that generated them.

Audit Review notes remain attached to versions

Each major line records the reason for changes, known limitations, and next validation steps.

Release Public packages appear after claim lock

Artifact links will be added when each manuscript or public overview is ready for external reuse.