MNEME literature evidence maps — method v1

1. Download a report's evidence JSON, engine.mjs, and reproduce.mjs into one folder.
2. With Node.js 20 or newer, run: node reproduce.mjs YOUR-REPORT.json
3. The script checks SHA-256(JSON.stringify(payload without reportHash)) and recomputes counts.
4. Source request URLs, UTC retrieval times, exact query, metadata records, exclusions,
   engine version, coverage limit, and retrieval count are included in each report.
5. The archive index records the engine file's SHA-256 at publication. Compare with
   your downloaded engine. Earlier engine versions may differ in collection limits.

Retrieval: Europe PMC REST search, resultType=core, source MED, first publication date
2020-01-01 through the stated cutoff, newest-first sorting; pagination in batches of 100.
Deduplication: DOI (case-insensitive), otherwise source + ID.
Exclusions: publication-type strings matching retract, withdraw, preprint, published erratum.
Analysis: publication-year counts; metadata publication-type counts (overlapping labels);
title-only, case-insensitive keyword matches (overlapping categories). Exact regex rules
are exported as RULES in engine.mjs. An absence of title words is not absence of evidence.
Abstract text is not analyzed or redistributed. Abstract availability is recorded only.
No manual eligibility screening, risk-of-bias assessment, full-text extraction, meta-analysis,
independent biological replication, or causal inference is performed by this engine.
Some records can concern animals, reviews, unrelated senses of reserve or resilience,
or populations that do not match the proposed next study. Review the source before use.

SHA-256 establishes file consistency, not who created a file or when. This is not an
externally anchored preregistration. The service index can change, so repeating a live
query later may produce a different corpus. The saved normalized snapshot preserves
the actual input to the published counts. A failed API call must not become a report.

Published reports are fixed snapshots. A visitor-triggered scan stays in that browser
session unless downloaded. Shared server-side updates use a separate persistent service.

Shared server-side runs are now stored independently in R2. GET /api/research/status lists their immutable evidence snapshots and change logs. The browser scan remains session-only. The editorial brief has a separate date and is not automatically rewritten.