NucLexNuclear Medicine Lexicon and Theranostics Ontology
Stage W0: private living prototype12 of 30 manuscripts drafted, 0 reviewed; 50 lexicon entries in draftWhat each later stage would need to show

Tools and roadmap

Human review of machine proposals

A language model can propose a term, a definition, or a relationship. Only an accountable person can accept it.

MonographBriefNX-M19Proposed length 2,500 to 5,000 words

Writing brief. This monograph has been assigned but the manuscript is not yet written or approved. What follows is the commissioning brief: its question, outline, evidence instructions, and review requirements.

Central question

Who decides whether a proposal is accepted, and what record makes that decision defensible?

Reader promise

The reader will understand the difference between a candidate and an accepted record, will see every artifact that must exist between the two (sources, model and prompt log, reviewer packet, decision, rationale), and will be able to judge any knowledge project by whether it preserves human approval authority or quietly delegates it. The monograph describes the NucLex workflow as a design, drawing on the discussion record's Version 1A agentic layer, and does not claim that the workflow has been run.

Required development

  1. Describe candidate generation: literature surveillance, term extraction, definition suggestion, relation suggestion, and mapping suggestion as distinct tasks with distinct failure modes.
  2. Describe source checks: verifying that a cited source exists, says what the candidate claims, and is of the right kind for the claim (Evidence).
  3. Describe the model and prompt record: model name and version, prompt version, retrieval set, date, and output hash, so that a candidate can be reproduced or at least attributed (Provenance).
  4. Describe decision, rationale, and revision: who may accept, reject, or defer; what rationale is required; how a later correction supersedes an earlier decision without erasing it.
  5. Preserve human approval authority. Explain why automation does not move the locus of accountability, and what sovereignty and legitimacy mean for a terminology others depend on.

Evidence and review

Evidence to obtain: the published editorial and review policies of at least two established terminology or ontology projects (a major clinical terminology's request and review process; a pathway or gene function knowledge base's curation model), with issuing body and version; peer-reviewed studies of automated term or relation extraction reporting precision and recall against a curated gold standard; and published guidance on documenting model and prompt provenance for scientific outputs. The discussion record (sections 10, 32 and 39) is the source for the NucLex design and is cited as a project document, not as external evidence.

Review concern: the manuscript describes an aspiration as if it were practice. Every workflow step must be marked as designed, piloted, or operating; in W0 none is operating.

Human review, Evidence, Provenance, Uncertainty, Identifier, Relationship, Provenance and evidence, and Interactional ontology.

Review requirements

Domain review by an editor or terminologist with governed terminology release experience. Source check of every claim about an external project's review policy against that project's own documents. Editorial check that no NucLex workflow step is described as implemented.

Review gate for this monograph

An evidence packet remains a proposal until reviewed.

The assignment exists. The manuscript has not been written or approved.