Research Publisher vNext — Discovery Summary

Status: discovery complete Date: 2026-09-17 Corpus inspected: kemiller2002/visual-engineering @ shallow clone of main Engine inspected: kemiller2002/research-publisher @ 10aa01f

Every claim below carries an evidence class: OBSERVED, DOCUMENTED ROS RULE, INFERENCE, PROPOSAL, or OPEN QUESTION.

The headline number

OBSERVED: The published relationship graph for the real research corpus contains 6 edges across 756 nodes. 749 of 756 documents (99.1%) are completely isolated.

Source: dist/data/research-graph.json in visual-engineering, fields nodes (756) and edges (6, all type: "related-document").

The corpus is not the problem. The corpus encodes 63 relationship values in related_documents / relatedDocuments, of which 54 resolve to real targets. The publisher derives 6 of them.

This single measurement explains most of what is meant by "the current publishing experience is ineffective", and it is a parser/contract defect, not a presentation defect.

The vNext requirements draft arrived mid-discovery

OBSERVED: when discovery began, no requirements draft existed in any reachable repository. I searched research-publisher, repository-operating-system and visual-engineering — every *.md, any filename matching *requirement*, and content matching vnext / v-next / next generation / redesign near research publisher. Zero matches, at ~19:40 UTC.

input-documents/research-publisher-vnext-requirements.txt (1,021 lines, 31 sections) was committed at 19:49 UTC as 550b1b9, after that search.

This sequencing is worth stating plainly: the evidence in these documents was gathered independently of the requirements. Where the two agree, it is convergence rather than confirmation bias; where they conflict, the conflict was not manufactured to fit a conclusion.

Deliverable E was rewritten against the real document once it arrived, and deliverable O is now a revision of it rather than an original. Nothing in input-documents/ was modified.

What was actually inspected

Measurements come from the committed build output, so they are what the system actually produced, not a re-run under different conditions.

Document index