Problem

Researchers reviewing downloaded biology literature need to trace summaries back to exact files and distinguish contradictory findings from studies conducted under different conditions.

Proposed workflow

Evidence Ledger is an initial UGREEN AI NAS Skill for adult biology researchers. Authorize up to ten local PDFs and three research claims; inventory the sources; retrieve contextual evidence; review source-linked tables and coverage warnings; then optionally approve an exact report and local destination for a verified save.

The workflow keeps file hashes, evidence locators, organisms, measurements, units, and time windows together. It separates retrieved statements, missing context, incomparable studies, and unreadable sources. Partial retrieval must not be treated as proof that evidence is absent.

Technical approach

The Markdown Skill uses the official Uliya Skill Kit 0.7.3 and catalog-supported orchestration through file-agent, document-agent, and knowledge-agent. Planned NAS capabilities cover authorized file discovery and hashing, document metadata and focused evidence retrieval, and approval-controlled text saving. Model data handling depends on the deployment configuration.

Development status

The initial Skill, evidence rules, and validation plan have been authored and passed the kit's local static checker. Installation, expert execution, extraction quality, and verified saving still require validation on the assigned NAS. No NAS runtime or user-test results are claimed.

Validation plan

Synthetic PDFs will test comparable opposing statements with retained locators; different study conditions and duplicate-file aliases; and unreadable or denied sources, empty scope, refused save approval, and destination collisions. Evaluation will measure traceability, coverage accuracy, and prevention of unapproved writes on the assigned NAS.

Limits

Scans, complex tables, missing context, extraction quality, and configured models can limit assessment. File hashes identify bytes rather than authenticity. This is literature-review assistance, not clinical advice, scientific validation, or a comprehensive systematic review.