How Research Streams Become Research Memory: Evidence Layers, Lineage, and Bounded Autoresearch in the NEP-MIN Mining Economics Observatory

This paper presents a theory and architecture of research observatories as systems for preserving research memory.

Using NEP-MIN, the New Economics Papers report on Mining, as a domain case, the paper argues that most research systems are designed to preserve documents but not the state of learning that accumulates around those documents. Weekly literature streams can identify new papers, but they do not automatically preserve judgments, classifications, review decisions, uncertainty states, lineage relationships, evidence status, or synthesis readiness.

The paper develops a framework in which issue packets, paper records, file records, extraction records, evidence claims, cross-issue indexes, synthesis memos, repair registers, autoresearch runs, registry entries, and provenance records are treated as governed research objects. These objects enable the observatory to preserve not only source materials but also the evolving research state associated with those materials.

The article introduces an evidence-layer model, lineage-repair framework, bounded-autoresearch architecture, and research-memory perspective intended to support longitudinal evidence accumulation while preserving review, provenance, and uncertainty.

This version (v2) preserves the intellectual contribution of the original release while substantially rewriting the manuscript around a clearer human-centered argument. The architecture remains unchanged, but the narrative has been reorganized around the problem of research-memory decay and the conditions required for durable, inspectable, review-gated research memory.

The package includes the manuscript, figure assets, transcript-bearing provenance materials, metadata records, changelog files, manifests, and supporting documentation.