Evidence and records

Keep sources, methods, implementations, plans, evidence, assurance, and claims independently reviewable.

PLLM research uses immutable, linked records rather than one undifferentiated success status. A paper can be acquired without being reproduced. A component can be implemented without benchmark parity, protected execution, or assurance. A measurement can be valid without supporting a broad publication claim.

Use this section to inspect evidence rules and comparison metrics. Source status lives in the paper catalog and research backlog. Every public claim should resolve to the exact source, implementation, plan, workload, environment, result, and limitation that support it.