shelfmark-research-archive

Research a Shelfmark Archive

Build an evidence trail before writing conclusions, and distinguish metadata evidence from content evidence.

Research Workflow

1. Call corpus_stats() and record freshness, corpus coverage, available facets, and whether an email corpus exists.
2. Break the question into a few concrete entities, identifiers, date ranges, and likely folders. Search documents with filename-style or title-style terms.
3. Use browse_folder() to understand a promising project's surrounding files instead of treating isolated search hits as the whole archive.
4. Refine search_docs() with known facets and dates. Call get_file() on leading results to capture exact path, author, authored date, rights, confidentiality, titles, duplicate copies, and residency.
5. When an email corpus is available and correspondence is relevant, call search_emails() without bodies first. Narrow by sender domain or year, then use include_body=true only for the small set needed as evidence.
6. If document-body evidence is required, ask to open selected paths with a separately authorized local file-reading tool. Shelfmark document search and get_file() expose metadata, not document bodies.
7. Produce the requested output with a source list that preserves exact paths or email headers, dates, and governance labels.

Evidence Rules

• Treat document metadata, filenames, and slide titles as discovery evidence, not proof of claims contained in a file.
• Treat returned email excerpts as partial content and mark truncation when Shelfmark does.
• Label conclusions as confirmed, inferred, or unresolved. Tie confirmed claims to content actually read.
• Use shareable_only=true for document searches when the brief or source list may leave the machine. Do not include UNKNOWN, confidential, or restricted material in an external-facing output.
• Mention stale-index warnings and result truncation because both limit completeness.

Output Shape

Keep the result compact:

1. Answer or executive summary.
2. Timeline or grouped findings when useful.
3. Sources with exact provenance and governance status.
4. Gaps, conflicts, and searches that produced no evidence.